SEARCH SESSIONS

Search All
 
Refine Results:
 
Year(s)

SOCIAL MEDIA

EMAIL SUBSCRIPTION

 
 

GTC On-Demand

AI Application Deployment and Inference
Presentation
Media
Abstract:
This customer panel brings together A.I. implementers who have deployed deep learning at scale using NVIDIA DGX Systems. We'll focus on specific technical challenges we faced, solution design considerations, and best practices learned from i ...Read More
Abstract:

This customer panel brings together A.I. implementers who have deployed deep learning at scale using NVIDIA DGX Systems. We'll focus on specific technical challenges we faced, solution design considerations, and best practices learned from implementing our respective solutions. Attendees will gain insights such as: 1) how to set up your deep learning project for success by matching the right hardware and software platform options to your use case and operational needs; 2) how to design your architecture to overcome unnecessary bottlenecks that inhibit scalable training performance; and 3) how to build an end-to-end deep learning workflow that enables productive experimentation, training at scale, and model refinement.

  Back
 
Topics:
AI Application Deployment and Inference, AI and DL Business Track (high level), Data Center and Cloud Infrastructure, AI for Business, HPC and Supercomputing
Type:
Panel
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8194
Streaming:
Download:
Share:
 
Abstract:
Learn how you can utilize TensorRT and NVIDIA Docker to quickly configure and deploy a GPU-accelerated inference server and start gaining insights from your trained deep neural network (DNN) models. TensorRT is a high-performance tool for low-latency ...Read More
Abstract:
Learn how you can utilize TensorRT and NVIDIA Docker to quickly configure and deploy a GPU-accelerated inference server and start gaining insights from your trained deep neural network (DNN) models. TensorRT is a high-performance tool for low-latency, high-throughput DNN inference. The latest release of TensorRT introduces a novel, framework-agnostic network definition format called universal framework format, which allows TensorRT to support and optimize DNN models trained in multiple deep learning frameworks. We'll leverage the TensorRT Python API to create a lightweight Python Flask application capable of serving multiple DNN models trained using TensorFlow, PyTorch, and Caffe, and also discuss how to containerize this inference service using NVIDIA Docker for ease of deployment at scale. This session will consist of a lecture, live demos, and detailed instructions.  Back
 
Topics:
AI Application Deployment and Inference, Tools and Libraries, Data Center and Cloud Infrastructure
Type:
Tutorial
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8495
Streaming:
Download:
Share:
 
Abstract:
The average human brain has about 100 billion nerve cells. We therefore investigate the question whether there are algorithms for artificial neural networks that are linear in the number of neurons, while the number of connections incident to a neuro ...Read More
Abstract:
The average human brain has about 100 billion nerve cells. We therefore investigate the question whether there are algorithms for artificial neural networks that are linear in the number of neurons, while the number of connections incident to a neuron is bounded by a constant. We offer two approaches to answer this question: First, we derive an algorithm that quantizes a trained artificial neural network such that the resulting complexity is linear. Second, we demonstrate that training networks, whose connections are determined by uniform sampling can achieve a similar precision as compared to using fully connected layers. Due to sparsity upfront, these networks can be trained much faster. Both approaches are made plausible by relating artificial neural units to Monte Carlo integration. We'll demonstrate the results for classic test datasets.  Back
 
Topics:
AI Application Deployment and Inference, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8780
Streaming:
Download:
Share:
 
Abstract:
Inspur has been deploying AI solutions with our customers, such as Microsoft, Alibaba, Baidu, BMW, for many years. We will share AI use cases on how we deploy AI at scale and take a close look at the technologies that enable AI deployments.
 
Topics:
AI Application Deployment and Inference, AI and DL Research, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8996
Streaming:
Share:
 
Abstract:
Now that Deep learning has moved out of the lab and into production, how do you provide training environments to all your internal customers working across business units with different requirements and avoid provisioning separate clusters? IBM has a ...Read More
Abstract:
Now that Deep learning has moved out of the lab and into production, how do you provide training environments to all your internal customers working across business units with different requirements and avoid provisioning separate clusters? IBM has applied decades of HPC experience to build a production ready learning stack, including servers accelerated with NVIDIA GPUs, workload and resource management software, ready to use open source frameworks and it's all covered by IBM support. The solution provides a secure multi-tenant environment so multiple data scientists can share a common set of resources, eliminating silos, while running multiple instances of the same or different applications. The deep learning effort is enhanced with end-to-end pipeline support from data ingestion and preparation, through model training and tuning, to inference. In this session, we will explore what an enterprise deep learning environment looks like and provide insights into the unique IBM value for accelerating the use of deep learning across a wide variety of industries.  Back
 
Topics:
AI Application Deployment and Inference
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81049
Streaming:
Download:
Share:
 
Abstract:
Learn how VisionLabs GPU-powered solutions contribute to creating a safer, smarter Megacity a metropolitan area with a total population in excess of ten million people. We'll do a deep dive into three implemented and ongoing huge scale smart-city ...Read More
Abstract:
Learn how VisionLabs GPU-powered solutions contribute to creating a safer, smarter Megacity a metropolitan area with a total population in excess of ten million people. We'll do a deep dive into three implemented and ongoing huge scale smart-city projects, understand challenges, technical specifics and how GPU computing impacts each of these cases: Face authentication-based immobilizer and driver monitoring systems for municipal service vehicles powered by the NVIDIA Jetson TX2 embedded platform; Megacity scale vehicle traffic analysis and anomalies detection powered by NVIDIA Tesla P40 with over 80 million daily recognition requests; National scale face identification platform for financial services with over 110 million faces in its database. The foundation of all these projects is VisionLabs LUNA a cross-platform object recognition software based on proprietary deep neural networks (DNN) inference framework. To build cost-effective solutions, VisionLabs use know-hows in DNN quantization and acceleration. In terms of accuracy, VisionLabs is recognized as a top three best in the world by National Institute of Standards and Technology's face recognition vendor test, and LFW by University of Massachusetts challenges.  Back
 
Topics:
AI Application Deployment and Inference, NVIDIA Inception Program, Intelligent Video Analytics and Smart Cities, Deep Learning and AI Frameworks, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8584
Streaming:
Download:
Share:
 
Abstract:
We'll delve into the nuts and bolts of how Valve has utilized deep learning to combat cheating in "Counter-Strike: Global Offensive." We'll cover total system details, from the high-level server architecture to the low-level features fed ...Read More
Abstract:
We'll delve into the nuts and bolts of how Valve has utilized deep learning to combat cheating in "Counter-Strike: Global Offensive." We'll cover total system details, from the high-level server architecture to the low-level features fed into the AI. Deep learning has proven to be very effective at identifying cheating behavior without any client-side instrumentation, making it robust against malicious attack by cheaters and cheat vendors. By retraining regularly, the network continues to evolve, picking up new cheating behaviors within hours of their appearance. As a result of this approach, certain types of cheats have been reduced by a factor of 100.  Back
 
Topics:
AI Application Deployment and Inference, AI for Gaming
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8732
Streaming:
Share:
 
Abstract:
Autoregressive wavenets have demonstrated extremely high quality real-time speech synthesis results.  However, the compute requirements and tight latency bounds have made them impractical for deployment on traditional CPU-only systems.  In ...Read More
Abstract:
Autoregressive wavenets have demonstrated extremely high quality real-time speech synthesis results.  However, the compute requirements and tight latency bounds have made them impractical for deployment on traditional CPU-only systems.  In this talk we demonstrate that Volta GPUs provide excellent real-time inference performance on these networks, making practical deployments possible.  We discuss several alternative implementation techniques and demonstrate their achieved performance on a V100 GPU.  Back
 
Topics:
AI Application Deployment and Inference, Speech and Language Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8968
Streaming:
Share:
 
Abstract:
This talk will provide an overview of what is happening in the world of artificial intelligence as it relates to networking, IT infrastructure, and IoT technologies. We will broadly cover AI topics ranging from machine learning and deep learning to s ...Read More
Abstract:
This talk will provide an overview of what is happening in the world of artificial intelligence as it relates to networking, IT infrastructure, and IoT technologies. We will broadly cover AI topics ranging from machine learning and deep learning to symbolic AI. Applied AI as well as general AI and their hybrids are all critical in solving many of today's complex long tail problems in real-time. Just as the capabilities, business opportunities, and positive benefits of AI are growing at a seemingly exponential rate so are the security vulnerabilities, failure modes, and potential adverse business impacts. We will discuss new hybrid neural symbolic approaches that promise to address these issues while simultaneously opening the door to powerful systems that dynamically learn and reason at multiple levels of abstraction, from raw data to high-level symbolic reasoning. We will cover use cases and solutions ranging from smart city, transportation, manufacturing, to security and networking.  Back
 
Topics:
AI Application Deployment and Inference, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8971
Streaming:
Download:
Share:
 
Abstract:
Attendees will learn and understand why AI techniques are so powerful, why developing and deploying optimal AI solutions is complex, why using AI techniques effectively is still difficult--and what Dell Technologies is doing to remove these difficult ...Read More
Abstract:
Attendees will learn and understand why AI techniques are so powerful, why developing and deploying optimal AI solutions is complex, why using AI techniques effectively is still difficult--and what Dell Technologies is doing to remove these difficulties and bring easier, effective AI to everyone. Dell Technologies includes seven companies with a comprehensive portfolio of technology products, services and solutions for global industry, government, and education markets, and aims to be the leader in designing and delivering the best AI solutions for every customer, of every type and scale. From Dell Precision workstations for developers and Gateways for edge sensors, to Dell EMC GPU-optimized PowerEdge Servers and Ready Solutions for Deep Learning and hybrid cloud offerings, Dell is leveraging its leadership in technology and in enterprise relationships to design a world-class portfolio of AI solutions for diverse customer workloads, requirements and objectives. This presentation will cover AI and deep learning in an enterprise context, including customer challenges and needs, and then discuss Dell AI solutions and strategy to empower people to use AI rapidly and effectively.  Back
 
Topics:
AI Application Deployment and Inference
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81046
Streaming:
Share:
 
Abstract:
We'll present a suite of artificial intelligence applications and computation geared towards increasing our understanding of the universe. The intensive collaboration between astrophysics and computer science has long started since Jim Gray and Alex ...Read More
Abstract:
We'll present a suite of artificial intelligence applications and computation geared towards increasing our understanding of the universe. The intensive collaboration between astrophysics and computer science has long started since Jim Gray and Alex Szalay. Nowadays, astrophysics continues to offer rich datasets, which are ideal for exploration with the latest in AI and computer science in general. We'll present successful projects in our space.ml initiative that try to answer a range of fascinating astrophysics questions. We'll show how we can use generative adversarial networks to go slightly beyond the Nyquist resolution limit in images, and to study the host galaxies of powerful quasars. We demonstrate how we can use transfer learning to identify rare galaxy mergers, and how to use variational autoencoders to forward model the processes in cosmology and galaxy evolution. We'll illustrate how we can use GPUs for compressive sensing to better analyze data from radio arrays, and to model the evolution of black holes over the age of the universe. Attendees will not only get our current answers to these questions but also get a taste of how AI is reshaping science today.  Back
 
Topics:
AI Application Deployment and Inference, Astronomy and Astrophysics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8667
Streaming:
Download:
Share:
 
Abstract:
TensorFlow is an open source software library for numerical computation using data flow graphs. NVIDIA TensorRT is an inference optimizer and runtime for runtime deployment. TensorRT provides optimizations for deep neural networks and uses reduced pr ...Read More
Abstract:
TensorFlow is an open source software library for numerical computation using data flow graphs. NVIDIA TensorRT is an inference optimizer and runtime for runtime deployment. TensorRT provides optimizations for deep neural networks and uses reduced precision to increase throughput, reduce latency, while maintaining accuracy. Today we announced tighter integration in TensorFlow for TensorRT through with new TensorFlow APIs, sub-graph optimizations and INT8 calibration to automatically leverage Tensor Cores on Volta GPUs. TensorRT delivers 2.5x faster inference throughput compared to inference without TensorRT. In this session, NVIDIA developers will use an example based workflow to show how to use this new capability.  Back
 
Topics:
AI Application Deployment and Inference, Deep Learning and AI Frameworks
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81009
Streaming:
Download:
Share:
 
Abstract:
Swiss Federal Railways (SBB) operate a 'diagnosis' train fitted with multiple high-resolution cameras that obtain images of tracks - all while traveling at a speed of 75 mph. Current data processing software conducted in real time on the train prod ...Read More
Abstract:
Swiss Federal Railways (SBB) operate a 'diagnosis' train fitted with multiple high-resolution cameras that obtain images of tracks - all while traveling at a speed of 75 mph. Current data processing software conducted in real time on the train produces  a too high rate of false positives/negatives to the extent that railway experts still need to go on the track to physically inspect anomalies. This is not only very dangerous, but sometimes even impossible and in addition it requires a lot of human labor. We describe how deep learning technologies have been developed to massively improve the automatic detection and classification of railway faults. This is not just a nice-to-have, but rather a must-have in order to ensure the safety of future rail transportation.  Back
 
Topics:
AI Application Deployment and Inference, Industrial Inspection
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8944
Streaming:
Download:
Share:
 
Abstract:
There is no shortage of hype around AI, but realizing value through machine and deep learning comes with its challenges. IBM PowerAI removes the inhibitors across each stage of a workflow, allowing enterprises to rapidly realize business value with A ...Read More
Abstract:
There is no shortage of hype around AI, but realizing value through machine and deep learning comes with its challenges. IBM PowerAI removes the inhibitors across each stage of a workflow, allowing enterprises to rapidly realize business value with AI.  Back
 
Topics:
AI Application Deployment and Inference
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81048
Streaming:
Share:
 
Abstract:
NVIDIA's video SDK is a set of APIs for hardware-accelerated video encoding and decoding using NVIDIA GPUs. We'll provide an overview of the APIs, with particular emphasis on the latest features, such as FFmpeg support of NVIDIA-accelerated transco ...Read More
Abstract:
NVIDIA's video SDK is a set of APIs for hardware-accelerated video encoding and decoding using NVIDIA GPUs. We'll provide an overview of the APIs, with particular emphasis on the latest features, such as FFmpeg support of NVIDIA-accelerated transcoding, quality and performance enhancements. We'll discuss some strategies on efficient usage of GPU video hardware acceleration for use cases such as video inferencing, transcoding, and media archiving.  Back
 
Topics:
AI Application Deployment and Inference, Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8601
Streaming:
Download:
Share:
 
Abstract:
We'll take a deep dive into honey bee hive health monitoring with NVIDIA's TX2, TensorRT (a high-performance deep learning inference optimizer), Kineticas insight engine running on DGX-1/DGXStaion, and Microsoft Cognitive Toolkit to rapidly o ...Read More
Abstract:
We'll take a deep dive into honey bee hive health monitoring with NVIDIA's TX2, TensorRT (a high-performance deep learning inference optimizer), Kineticas insight engine running on DGX-1/DGXStaion, and Microsoft Cognitive Toolkit to rapidly optimize, validate, and deploy trained neural networks for inference. In recent years, the media has reported that bees seem to be dying at an unprecedented rate. We'll explore how new accelerated analytics technologies and their corresponding compute platforms can deliver game-changing possibilities for innovation as we follow a honey bee farm scientist in California, who agreed to field test this real-time monitoring solution with her beehives.  See first-hand how adaptable and accessible these complex, cutting-edge technologies have become and how we can use intelligent monitoring technologies to help rescue the honey bee in the real-world environmental analytics opportunity.  Back
 
Topics:
AI Application Deployment and Inference, Deep Learning and AI Frameworks
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8508
Streaming:
Share:
 
Abstract:
Fujikura is pushing forward of implementation of the smart factory with AI and IoT for improving the productivity and production quality. In this presentation, we will present visual inspection system incorporating deep learning in the production pro ...Read More
Abstract:
Fujikura is pushing forward of implementation of the smart factory with AI and IoT for improving the productivity and production quality. In this presentation, we will present visual inspection system incorporating deep learning in the production process of semiconductor lasers. Not only OK/NG classification, but also multiple NG mode classification was performed. The inspection accuracy of 95 % that is equivalent to skilled workers' accuracy was achieved by optimizing the data set and the hyper parameters of a CNN model. The activation map was used for reliability and validity assurance. We will present the difficulty in our practical application in manufacturing industry, such as the small number of some category and small defect/chip size ratio, and also introduce our countermeasures.  Back
 
Topics:
AI Application Deployment and Inference, Industrial Inspection
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8911
Streaming:
Download:
Share:
 
Abstract:
NASA's heliophysics division operates a fleet of spacecraft, the so-called Heliophysics System Observatory, to monitor the Sun's activity and how its changes drive space weather in interplanetary space and in the near-Earth environment. We'll pres ...Read More
Abstract:
NASA's heliophysics division operates a fleet of spacecraft, the so-called Heliophysics System Observatory, to monitor the Sun's activity and how its changes drive space weather in interplanetary space and in the near-Earth environment. We'll present case studies of how a number of challenging problems encountered in heliophysics can be tackled using deep learning: spectropolarimetric inversions for measuring the magnetic field on the solar surface, and mega-Kelvin thermometry of the Sun's corona by using a deep neural network to solve a compressed sensing problem. These low-cost solutions make possible new concepts for deep space missions for space weather monitoring. Some of the work in this presentation was made possible by NASA's Frontier Development Lab, a public-private partnership between the agency and industry partners (including the SETI Institute, NVIDIA, IBM, Intel, kx & Lockheed Martin), whose mission is to use artificial intelligence to tackle problems related to planetary defense and heliophysics.  Back
 
Topics:
AI Application Deployment and Inference, Accelerated Analytics, Astronomy and Astrophysics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8222
Streaming:
Download:
Share:
 
Abstract:
A wide area and city surveillance system solution for running real-time video analytics on thousands of 1080p video streams will be presented. System hardware is an embedded computer cluster based on NVIDIA TX1/TX2 and NXP iMX6 modules. A custom ...Read More
Abstract:

A wide area and city surveillance system solution for running real-time video analytics on thousands of 1080p video streams will be presented. System hardware is an embedded computer cluster based on NVIDIA TX1/TX2 and NXP iMX6 modules. A custom designed system software manages job distribution, resulting in collection and system wide diagnostics including instantaneous voltage, power and temperature readings. System is fully integrated with a custom designed video management software, IP cameras and network video recorders. Instead of drawing algorithm results on the processed video frames, re-encoding and streaming back to the operator computer for display, only the obtained metadata is sent to the operator computer. Video management software streams video sources independently, and synchronizes decoded video frames with the corresponding metadata locally, before presenting the processed frames to the operator.

  Back
 
Topics:
AI Application Deployment and Inference, Intelligent Video Analytics and Smart Cities
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8409
Streaming:
Download:
Share:
 
Abstract:
Deep learning and reinforcement learning are widely used in ads products of JD.com, e.g. ranking model in recommender systems, bidding model in ad exchange business and automatic ads review systems. These technologies have brought great benefits to J ...Read More
Abstract:
Deep learning and reinforcement learning are widely used in ads products of JD.com, e.g. ranking model in recommender systems, bidding model in ad exchange business and automatic ads review systems. These technologies have brought great benefits to JD.com and all of them are built on Nvidia GPUs.  Back
 
Topics:
AI Application Deployment and Inference, Consumer Engagement and Personalization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81016
Streaming:
Download:
Share:
 
Abstract:
Personalized learning has been a promising but often elusive ideal sought after in education. We'll demonstrate the progress made with two concrete examples of personalized learning supports implemented at scale in a massive open online course (MOOC ...Read More
Abstract:
Personalized learning has been a promising but often elusive ideal sought after in education. We'll demonstrate the progress made with two concrete examples of personalized learning supports implemented at scale in a massive open online course (MOOC) and on the UC Berkeley campus in a collaboration with the Office of the Registrar. Both approaches employ long short-term memory to leverage a collaborative signal out of millions of historic learner actions. In the case of the MOOC, the next page a learner is expected to spend considerable time on is predicted and offered as a real-time suggestion. At the university, we consider sequences of millions of historic enrollments over the past eight years. These sequences of course identifiers, when modeled with representation learning approaches most commonly applied to natural language, reveal a tremendous degree of semantic relational information about the courses which can be visualized, reasoned about, and surfaced to students. Our course information platform uses this automatically inferred semantic information to help students navigate the university's offerings and provides personalized course suggestions based on topic preference.  Back
 
Topics:
AI Application Deployment and Inference, Consumer Engagement and Personalization, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8597
Streaming:
Share:
 
Abstract:
Businesses of all sizes are increasingly recognizing the potential value of AI, but few are sure how to prepare for the transformational change it is sure to bring to their organizations. Danny Lange rolled out company-wide AI platforms at Uber ...Read More
Abstract:

Businesses of all sizes are increasingly recognizing the potential value of AI, but few are sure how to prepare for the transformational change it is sure to bring to their organizations. Danny Lange rolled out company-wide AI platforms at Uber and Amazon; now, through Unity Technologies, he's making AI available to the rest of us. He'll also share his thoughts for the most exciting advances that AI will bring over the next year. His insights will help you understand the true potential of AI, regardless of your role or industry.

  Back
 
Topics:
AI Application Deployment and Inference, Advanced AI Learning Techniques (incl. GANs and NTMs), AI and DL Business Track (high level), AI for Business
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8729
Streaming:
Share:
 
Abstract:
Come learn how you can optimize the deployment of your trained neural networks using the GPU-accelerated inferencing library called TensorRT. TensorRT is a high-performance tool for low-latency, high-throughput deep neural network (DNN) inference tha ...Read More
Abstract:
Come learn how you can optimize the deployment of your trained neural networks using the GPU-accelerated inferencing library called TensorRT. TensorRT is a high-performance tool for low-latency, high-throughput deep neural network (DNN) inference that runs on NVIDIA GPUs. The latest release of TensorRT introduces a novel, framework-agnostic network definition format called universal framework format, allowing TensorRT to support and optimize DNN models trained in multiple deep learning frameworks like Caffe and TensorFlow. It also provides the capability to run inference at reduced precision, giving developers the ability to take advantage of new GPU hardware features like the Volta Tensor Core architecture. This session will be a combination of lecture and live demos.  Back
 
Topics:
AI Application Deployment and Inference, Tools and Libraries, Performance Optimization, Data Center and Cloud Infrastructure
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8496
Streaming:
Share:
 
Abstract:
Artificial intelligence helps you hire faster and smarter. It also helps you determine your career path, learning, and development. Wondering how? AI platforms have a brain that reads, understands, and analyzes just as human beings do. They can read ...Read More
Abstract:
Artificial intelligence helps you hire faster and smarter. It also helps you determine your career path, learning, and development. Wondering how? AI platforms have a brain that reads, understands, and analyzes just as human beings do. They can read thousands and millions of resumes, JDs, career progressions, and learning content in a matter of seconds. This equips them with intelligence creating a neural network of skills, demographics, industries, occupations, and courses/certifications. This acts as the central intelligence powering search and match algorithms to find accurate matches to job demands in a few seconds. The NLP layer helps understand intent, for example, it differentiates between 'Worked with a PM' and 'Worked as a PM' to determine that the former could work collaboratively and the latter could drive projects. AI platforms mimic a recruiter or hiring manager's brain to find that right match. What takes HR 20-30 days is done in a few seconds by an AI platform. It helps HR leaders in workforce planning by forecasting what skills and domains to invest, maintain, or upgrade in their organizations, which could be a game changer especially for people-centric organizations.  Back
 
Topics:
AI Application Deployment and Inference, Accelerated Analytics, AI and DL Research, AI and DL Business Track (high level)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8303
Streaming:
Share:
 
Abstract:
Using the latest advancements from TensorFlow including the Accelerated Linear Algebra (XLA) Framework, JIT/AOT Compiler, and Graph Transform Tool, we'll demonstrate how to optimize, profile, and deploy TensorFlow models in GPU-based production envi ...Read More
Abstract:
Using the latest advancements from TensorFlow including the Accelerated Linear Algebra (XLA) Framework, JIT/AOT Compiler, and Graph Transform Tool, we'll demonstrate how to optimize, profile, and deploy TensorFlow models in GPU-based production environments. We'll cover many demos based on open source tools. You can completely reproduce all demos through Docker on your own GPU cluster. See http://pipeline.ai for links to the GitHub Repo.  Back
 
Topics:
AI Application Deployment and Inference, NVIDIA Inception Program, Deep Learning and AI Frameworks
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8621
Streaming:
Share:
 
Abstract:
Craig Morioka, UCLA Adjunct Associate Professor of Radiological Sciences, and Dima Lituiev, Postdoctoral Scholar at the University of California San Francisco, Institute for Computational Health Sciences, will discuss how they empower their fellow fa ...Read More
Abstract:
Craig Morioka, UCLA Adjunct Associate Professor of Radiological Sciences, and Dima Lituiev, Postdoctoral Scholar at the University of California San Francisco, Institute for Computational Health Sciences, will discuss how they empower their fellow faculty, staff, and students with the latest techniques in training and deploying deep neural networks through NVIDIAs Deep Learning Institute (DLI) University Ambassador Program - a new AI and Deep Learning education enablement program for universities. This will include a dive into the benefits of an online learning platform, which uses GPUs in the cloud, by stepping through the DLIs online Image Segmentation and Radiomics labs. The Image Segmentation lab leverages an example from medical image analysis where it is often important to separate pixels corresponding to different types of tissue or cells for the purposes of diagnostics and treatment planning. Dima uses image segmentation in his research to facilitate diagnostics of kidney rejection by analyzing histological slides from patients with kidney transplants. We will explore how the Tensorflow code is structured and how the Tensorboard tool can be used to visualize structure and training dynamics of segmentation models. The focus of the Radiomics lab is detection of the 1p19q co-deletion biomarker using deep learning - specifically convolutional neural networks using the Keras and TensorFlow computing frameworks. Attendees will also learn how they can apply to become a DLI University Ambassador and bring the latest in Deep Learning and AI education to their academic communities.    Back
 
Topics:
AI Application Deployment and Inference, Deep Learning and AI Frameworks, AI and DL Business Track (high level)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8823
Streaming:
Share:
 
Abstract:
Learn how to combine computer vision techniques and deep learning to improve the sensitivity of a real-time, GPU-powered safety system. In petawatt laser systems, firing at 10 Hz, suddenly appearing scatterers can damage components. Spreading of dama ...Read More
Abstract:
Learn how to combine computer vision techniques and deep learning to improve the sensitivity of a real-time, GPU-powered safety system. In petawatt laser systems, firing at 10 Hz, suddenly appearing scatterers can damage components. Spreading of damage can be avoided by suspending operation immediately on occurrence of such an event. We'll present our approach for the automatic detection of critical failure states from intensity profiles of the laser beam. By incorporating quick feature detection and learned heuristics for feature classification, both real-time constraints and limited available training data are accommodated. Localization of triggering feature is crucial for when the problem is located in non-sensitive sections and will not be removed from the beam in production.  Back
 
Topics:
AI Application Deployment and Inference, Advanced AI Learning Techniques (incl. GANs and NTMs), Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8330
Streaming:
Share:
 
Abstract:
We'll present an in-car ADAS technology to detect drowsy driving. This technique can be used to alert and awaken the driver, or take corrective actions if required. We employ a CNN-based approach for this technique, which is trained on a mix of synt ...Read More
Abstract:
We'll present an in-car ADAS technology to detect drowsy driving. This technique can be used to alert and awaken the driver, or take corrective actions if required. We employ a CNN-based approach for this technique, which is trained on a mix of synthetic and real images. We'll cover the details of the detection system pipeline and the synthetic dataset generation. We'll also show a demonstration of the detection system in action.  Back
 
Topics:
AI Application Deployment and Inference, Autonomous Vehicles
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8399
Streaming:
Share:
 
Abstract:
What is Deep Learning? In what fields is it useful? How does it relate to artificial intelligence? We'll discuss  deep learning and why this powerful new technology is getting so much attention, learn how deep neural networks are traine ...Read More
Abstract:

What is Deep Learning? In what fields is it useful? How does it relate to artificial intelligence? We'll discuss  deep learning and why this powerful new technology is getting so much attention, learn how deep neural networks are trained to perform tasks with super-human accuracy, and the challenges organizations face in adopting this new approach. We'll also cover some of the best practices, software, hardware, and training resources that many organizations are using to overcome these challenges and deliver breakthrough results.

  Back
 
Topics:
AI Application Deployment and Inference, Deep Learning and AI Frameworks, Deep Learning and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8669
Streaming:
Share:
 
Abstract:
Learn how to use GPUs to accelerate gradient boosting on decision trees. We'll discuss CUDA implementation of CatBoost an open-source library that successfully handles categorical features and shows better quality compared to other open-source gra ...Read More
Abstract:
Learn how to use GPUs to accelerate gradient boosting on decision trees. We'll discuss CUDA implementation of CatBoost an open-source library that successfully handles categorical features and shows better quality compared to other open-source gradient boosted decision trees libraries. We'll provide a brief overview of problems which could be solved with CatBoost. Then, we'll discuss challenges and key optimizations in the most significant computation blocks. We'll describe how one can efficiently build histograms in shared memory to construct decision trees and how to avoid atomic operation during this step. We'll provide benchmarks that shows that our GPU implementation is five to 40 times faster compared to Intel server CPUs. We'll also provide performance comparison against GPU implementations of gradient boosting in other open-source libraries.  Back
 
Topics:
AI Application Deployment and Inference, Tools and Libraries, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8393
Streaming:
Share:
 
Abstract:
We'll present results on speeding up Bayesian inference in NVIDIA DGX-1 server for medical diagnostics. Bayesian inference is an AI technique to reason under uncertainty that is computationally and data intensive. We'll discuss the implications for ...Read More
Abstract:
We'll present results on speeding up Bayesian inference in NVIDIA DGX-1 server for medical diagnostics. Bayesian inference is an AI technique to reason under uncertainty that is computationally and data intensive. We'll discuss the implications for both inference and training of Bayesian networks.  Back
 
Topics:
AI Application Deployment and Inference, Accelerated Analytics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8488
Streaming:
Share:
 
Abstract:
SOFWERX developed a vision-based classifier using commodity hardware and machine learning libraries to satisfy an urgent high-level requirement. To track the usage of tank ammunition, the team had to address challenges involving unavailable training ...Read More
Abstract:
SOFWERX developed a vision-based classifier using commodity hardware and machine learning libraries to satisfy an urgent high-level requirement. To track the usage of tank ammunition, the team had to address challenges involving unavailable training data, varying spatial orientations, and limited power consumption. To resolve these challenges, SOFWERX generated an augmented dataset using synthetic models, implemented spatial transformers, and experimented with different hardware/software optimizations.  Back
 
Topics:
AI Application Deployment and Inference, Performance Optimization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8193
Streaming:
Download:
Share:
 
Abstract:
Artificial intelligence has made great strides in many technology sectors, however, it has yet to impact the design and applications of radio frequency (RF) and wireless systems. This is primarily due to the industry''s preference towards field progr ...Read More
Abstract:
Artificial intelligence has made great strides in many technology sectors, however, it has yet to impact the design and applications of radio frequency (RF) and wireless systems. This is primarily due to the industry''s preference towards field programmable gate array (FPGA) systems. Conversely, the deep learning revolution has been fueled by GPUs and the ease with which they may be programmed for highly parallel computations. The next generation RF and wireless technology will require wide-band systems capable of real-time machine learning with GPUs. Working with strategic partners, we''ve designed a software configurable wide-band RF transceiver system capable of performing real-time signal processing and machine learning with a Jetson TX2. We discuss system performance, collection of RF training data, and the software used by the community to create custom applications. Additionally, we''ll present data demonstrating applications in the field of RF machine learning and deep learning.  Back
 
Topics:
AI Application Deployment and Inference, NVIDIA Inception Program, Cyber Security, IoT, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8375
Streaming:
Share:
 
Abstract:
Learn how a research paper from Adobe Research Labs makes it into a real customer product like Photoshop. We attempted to solve a number of challenging issues about applying the technology to real-world use cases, including large model size, heavy me ...Read More
Abstract:
Learn how a research paper from Adobe Research Labs makes it into a real customer product like Photoshop. We attempted to solve a number of challenging issues about applying the technology to real-world use cases, including large model size, heavy memory consumption, and slow runtime performance.  Back
 
Topics:
AI Application Deployment and Inference
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8550
Streaming:
Download:
Share:
 
Abstract:
OpenNMT is an open source neural machine translation and neural machine sequencing model. Using Volta Tensor Cores and TensorRT, we''re able to improve performance by 100 times over CPU implementation. We''ll discuss OpenNMT and how we implement it v ...Read More
Abstract:
OpenNMT is an open source neural machine translation and neural machine sequencing model. Using Volta Tensor Cores and TensorRT, we''re able to improve performance by 100 times over CPU implementation. We''ll discuss OpenNMT and how we implement it via TensorRT. We''ll show how by using our plugin interface and new TensorRT features, we''re able to implement this network at high performance.  Back
 
Topics:
AI Application Deployment and Inference, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8822
Streaming:
Download:
Share:
 
Abstract:
Organizations everywhere want to AI-infuse every aspect of their business, but need a platform that delivers the scale and flexibility to fit both IT operational constraints, as well as workload performance demanded by data scientists. Attend this se ...Read More
Abstract:
Organizations everywhere want to AI-infuse every aspect of their business, but need a platform that delivers the scale and flexibility to fit both IT operational constraints, as well as workload performance demanded by data scientists. Attend this session to get see the latest advancements in scaling in GPU servers and deep learning software, and hear how the latest solutions from NVIDIA solve your biggest AI platform challenges  Back
 
Topics:
AI Application Deployment and Inference, Data Center and Cloud Infrastructure, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8196
Streaming:
Download:
Share:
 
Abstract:
Deep learning systems are usually developed by data scientists, who are good at mathematics and computer science. But to deploy and operationalize these models for broader use, you need the devops mindset and tools. We''ll show you how to connect the ...Read More
Abstract:
Deep learning systems are usually developed by data scientists, who are good at mathematics and computer science. But to deploy and operationalize these models for broader use, you need the devops mindset and tools. We''ll show you how to connect the workflow between the data scientists and devops. We''ll also explore basic continuous integration and delivery concepts and how they can be applied to deep learning models. Using a number of AWS services, we''ll showcase how you can take the output of a deep learning model and deploy it to perform predictions in real time with low latency and high availability. In particular, we''ll showcase the ease of deploying DL to predict functions using Apache MXNet (a deep learning library), Amazon ECS, Amazon S3, and Amazon ECR, Amazon developer tools, and AWS CloudFormation.  Back
 
Topics:
AI Application Deployment and Inference
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8173
Streaming:
Download:
Share:
 
Abstract:
In order to fulfill customer''s requirement, companies have to guarantee the quality of delivered products, which can often be achieved only by manually inspection of the finished product. Since human-based defect inspection and classification are ti ...Read More
Abstract:
In order to fulfill customer''s requirement, companies have to guarantee the quality of delivered products, which can often be achieved only by manually inspection of the finished product. Since human-based defect inspection and classification are time-consuming and the results vary by individuals, automatic defect detection and classification has the potential to reduce the cost of quality assurance significantly. In this talk, we will demonstrate how to utilize deep learning algorithms, i.e., Fully Convolutional Neural Network to build a general defect inspection and classification model. We will also share experiences on how to effectively collect labelling data, deal with imbalance data, and also how to optimize the model in terms of latency and throughput with TensorRT before deploy the model to the production line.  Back
 
Topics:
AI Application Deployment and Inference, Industrial Inspection, IoT, Robotics & Drones, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8682
Streaming:
Download:
Share:
 
Abstract:
Learn how synthetic data can be used to develop traditional and Convolutional Neural Network (CNN) image segmentation models when labelled training data is limited. We will describe hard drive media defect patterns and how they relate to problems i ...Read More
Abstract:
Learn how synthetic data can be used to develop traditional and Convolutional Neural Network (CNN) image segmentation models when labelled training data is limited. We will describe hard drive media defect patterns and how they relate to problems in the manufacturing line, show why CNN models were chosen for some defect patterns, and how the CNN models were trained using both synthetic and real data. Different architectures using CNNs were explored and the resulting benefits and drawbacks are presented.  Back
 
Topics:
AI Application Deployment and Inference, Industrial Inspection, IoT, Robotics & Drones, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8415
Streaming:
Download:
Share:
 
Abstract:
Machine learning has revolutionized many important fields, ranging from computer vision and natural language processing to healthcare and robotics. In this session, we will discuss how developers can embrace machine learning methods for graphics and ...Read More
Abstract:
Machine learning has revolutionized many important fields, ranging from computer vision and natural language processing to healthcare and robotics. In this session, we will discuss how developers can embrace machine learning methods for graphics and gaming. We''ll cover both gaming use cases and general applications of machine learning as well as how to best leverage recent GPU hardware for machine learning workloads.  Back
 
Topics:
AI Application Deployment and Inference, Graphics and AI, AI for Gaming, Rendering and Ray Tracing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8957
Streaming:
Download:
Share:
 
Abstract:
We''ll discuss anomaly detection on vehicle CAN BUS. We developed a novel solution for neural networks to detect anomalies in CAN data. Due to the inherent characteristics of controller area (CAN) networks, such as lack of authentication and followin ...Read More
Abstract:
We''ll discuss anomaly detection on vehicle CAN BUS. We developed a novel solution for neural networks to detect anomalies in CAN data. Due to the inherent characteristics of controller area (CAN) networks, such as lack of authentication and following a broadcast routing scheme, devices connected to a CAN network are exposed to a broad range of cyberattacks. Our work aims to alleviate this problem by providing an anomaly detection mechanism, that is, identifying deviations from normal network traffic, to enhance the security of CAN networks. This invention is leveraged as one of the intrusion detection methods in a broader NVIDIA embedded software security system deployed in automotive platforms. In this specific application, the embedded system is a car computer -- an embedded system deployed in modern vehicles. Typical examples: infotainment systems, ADAS units, dashboards, head units. The vulnerable endpoints are all the peripherals connected to the computer. Typical examples: sensors, cameras, media devices, local and wide area communication interfaces and devices (for example, WiFi, BT, cellular), specific car network interfaces and devices.  Back
 
Topics:
AI Application Deployment and Inference, Deep Learning and AI Frameworks, Cyber Security, Autonomous Vehicles
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8347
Streaming:
Download:
Share:
 
Abstract:
Learn how to achieve 100% R/W cache hit rate for most intermediate tensors in CNN and over 80% typical DRAM traffic saving, with general applicability to a limited cache size and large tensors. The high-throughput NVIDIA Tensor Core and DLA demand hi ...Read More
Abstract:
Learn how to achieve 100% R/W cache hit rate for most intermediate tensors in CNN and over 80% typical DRAM traffic saving, with general applicability to a limited cache size and large tensors. The high-throughput NVIDIA Tensor Core and DLA demand high memory traffic. Chaining of consecutive layers in CNN can save DRAM traffic by reusing intermediate tensors in cache. This strategy is effective only with small tensors and a large cache. In this work, we slice tensors into small tiles (with halo) and chain these tiles so the requirement for perfect caching can always be fulfilled. Our implementation of this approach is proven to be very effective in saving DRAM traffic. This work allows us to solve the memory bandwidth issue of CNN with a relatively small but high-bandwidth cache.  Back
 
Topics:
AI Application Deployment and Inference, Performance Optimization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8299
Streaming:
Share:
 
Abstract:
We''ll introduce how Bing built a scalable, responsive, and economical object detection API based on NVIDIA GPUs and Azure cloud platforms. Object detection is an important image understanding technique as the entry point or dispatcher to guide users ...Read More
Abstract:
We''ll introduce how Bing built a scalable, responsive, and economical object detection API based on NVIDIA GPUs and Azure cloud platforms. Object detection is an important image understanding technique as the entry point or dispatcher to guide users to more specific scenarios. However, it is very challenging to provide object detection services on web-scale images because it is intrinsically a compute-intensive task and thus resource demanding. We''ll also introduce how to use NVIDIA''s CUDA profiling toolchain and cuDNN to make the system even more cost-effective. The system currently supports billion-level traffic, covering Bing''s entire index.  Back
 
Topics:
AI Application Deployment and Inference, Performance Optimization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8620
Streaming:
Download:
Share:
 
Abstract:
Last year, we began to see promising results of applying Deep Learning in an unexpected space: hardware QA. Fast forward +365, and the efforts have been to expand on what we''ve learned, push the technology broader and into other areas that will ulti ...Read More
Abstract:
Last year, we began to see promising results of applying Deep Learning in an unexpected space: hardware QA. Fast forward +365, and the efforts have been to expand on what we''ve learned, push the technology broader and into other areas that will ultimately aid in our greatest challenge: testing at scale. In this session we will highlight a new piece of the problem we are tackling: VR. We will introduce methodologies for not only addressing the unique problems that VR testing presents, but will also showcase some of the other test process areas where we are applying other Deep Learning models to gain efficiency in our overall production pipeline. From using DL on our bug mining to create a quicker path from tester to developer and back, to analysis on end user issues as a method for task automation, explore how we are enabling speed, accuracy and efficiency.  Back
 
Topics:
AI Application Deployment and Inference, Virtual Reality and Augmented Reality, Tools and Libraries, Graphics and AI, AI for Gaming
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8262
Streaming:
Share:
 
Abstract:
We''ll detail the journey of building Seeing AI, an app from Microsoft AI & Research that narrates the world around you. Designed for the blind and low-vision community, this research project harnesses the power of AI to describe people, text, an ...Read More
Abstract:
We''ll detail the journey of building Seeing AI, an app from Microsoft AI & Research that narrates the world around you. Designed for the blind and low-vision community, this research project harnesses the power of AI to describe people, text, and objects. Seeing AI leverages object classification, detection, image captioning, and more, with several running on the device in real time at more than 15 frames per second. We''ll go over the learnings, challenges, hits, and misses we encountered while developing the application.  Back
 
Topics:
AI Application Deployment and Inference, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8598
Streaming:
Share:
 
Abstract:
We''ll introduce deep learning infrastructure for building and maintaining autonomous vehicles, including techniques for managing the lifecycle of deep learning models, from definition, training and deployment to reloading and life-long ...Read More
Abstract:

We''ll introduce deep learning infrastructure for building and maintaining autonomous vehicles, including techniques for managing the lifecycle of deep learning models, from definition, training and deployment to reloading and life-long learning. DNN autocurates and pre-labels data in the loop. Given data, it finds the best run-time optimized deep learning models. Training scales with data size beyond multi-nodes. With these methodologies, one takes only data from the application and feeds DL predictors to it. This infrastructure is divided into multiple tiers and is modular, with each of the modules containerized to lower infrastructures like GPU-based cloud infrastructure.

  Back
 
Topics:
AI Application Deployment and Inference, Data Center and Cloud Infrastructure, Autonomous Vehicles, Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8531
Streaming:
Download:
Share:
 
Abstract:
Deploying machine learning-based predictive models to the oil field is quite challenging. They are remote, hazardous, and have spotty connectivity to the cloud. The world of operationalizing a model is very different than the perfect lab environment ...Read More
Abstract:
Deploying machine learning-based predictive models to the oil field is quite challenging. They are remote, hazardous, and have spotty connectivity to the cloud. The world of operationalizing a model is very different than the perfect lab environment where the models are born. We'll detail the requirements of our oil and gas customers and how we were able to meet those requirements such that we could deploy a new generation of analytics with a complete software engineering discipline and mentality around it by taking advantage of the Microsoft IoT Edge platform. This is currently a pilot project under way and, due to the engineering principals in place, we are able to complete a loop from the field to the lab and back again.  Back
 
Topics:
AI Application Deployment and Inference, IoT, Robotics & Drones, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8714
Streaming:
Download:
Share:
 
Abstract:
We describes concept of Digital Twin with respect to the Railway Network. Railroad customers across the world manage thousands of miles of Track infrastructure that consists of the Rails, Ballast, Ties, Bridges, Tunnels, Wayside equipment, etc. This ...Read More
Abstract:
We describes concept of Digital Twin with respect to the Railway Network. Railroad customers across the world manage thousands of miles of Track infrastructure that consists of the Rails, Ballast, Ties, Bridges, Tunnels, Wayside equipment, etc. This talk demonstrates a new approach to Track infrastructure monitoring that GE is piloting for customers using the concept of Digital Twin for network. Using an offline GPU infrastructure Deep Learning models are created and trained on large volumes of video data to learn the state of healthy Track and predict anomalies. During the talk, real customer use-case videos will be shown that demonstrate Analytics on videos from Locomotive-mounted cameras with Deep Learning models to calculate health index and display on a map for driving Maintenance decisions.  Back
 
Topics:
AI Application Deployment and Inference, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8614
Streaming:
Download:
Share:
 
Abstract:
How do meteorologists predict weather or weather events such as hurricanes, typhoons, and heavy rain? Predicting weather events were done based on supercomputer (HPC) simulations using numerical models such as WRF, UM, and MPAS. But recently, many de ...Read More
Abstract:
How do meteorologists predict weather or weather events such as hurricanes, typhoons, and heavy rain? Predicting weather events were done based on supercomputer (HPC) simulations using numerical models such as WRF, UM, and MPAS. But recently, many deep learning-based researches have been showing various kinds of outstanding results. We'll introduce several case studies related to meteorological researches. We'll also describe how the meteorological tasks are different from general deep learning tasks, their detailed approaches, and their input data such as weather radar images and satellite images. We'll also cover typhoon detection and tracking, rainfall amount prediction, forecasting future cloud figure, and more.  Back
 
Topics:
AI Application Deployment and Inference, Climate, Weather, Ocean Modeling, Computer Vision, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8816
Streaming:
Download:
Share:
 
Speakers:
Abstract:
We'll share information and lessons learned from developing a scalable visual search engine to handle a massive volatile inventory like eBay. We'll describe how eBay data is challenging for visual search, how to leverage a single deep neural networ ...Read More
Abstract:
We'll share information and lessons learned from developing a scalable visual search engine to handle a massive volatile inventory like eBay. We'll describe how eBay data is challenging for visual search, how to leverage a single deep neural network to perform multiple tasks efficiently, how to deploy our solution in a distributed cloud infrastructure, and which optimizations we have made for a trade-off between relevance and latency. We'll give examples and insights to benefit computer vision practitioners in the industry who intend to build up visual search engines from scratch.  Back
 
Topics:
AI Application Deployment and Inference, Data Center and Cloud Infrastructure, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8766
Streaming:
Share:
 
Abstract:
In this talk we will cover the essential building blocks of the AI platform Nvidia engineers are using to build a world-class automotive perception stack. Through a computer vision application example, we will see how to improve a baseline model to p ...Read More
Abstract:
In this talk we will cover the essential building blocks of the AI platform Nvidia engineers are using to build a world-class automotive perception stack. Through a computer vision application example, we will see how to improve a baseline model to produce better, faster predictions. The talk will focus on: - hyper-parameter optimization, - model complexity reduction (pruning), - target platform optimizations (TensorRT integration), - automation of complex workflows  Back
 
Topics:
AI Application Deployment and Inference, Deep Learning and AI Frameworks
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8633
Streaming:
Download:
Share:
 
Abstract:
We'll talk about how to use Singularity to containerize deep learning applications. We'll provide compelling reasons to choose Singularity over Docker. We'll cover deep learning frameworks, including TensorFlow, NV-Caffe, MXNet, and others. We'll ...Read More
Abstract:
We'll talk about how to use Singularity to containerize deep learning applications. We'll provide compelling reasons to choose Singularity over Docker. We'll cover deep learning frameworks, including TensorFlow, NV-Caffe, MXNet, and others. We'll present the current challenges and workarounds when using Singularity in a HPC cluster. We'll compare the performance of Singularity to bare-metal systems.  Back
 
Topics:
AI Application Deployment and Inference, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8368
Streaming:
Download:
Share:
 
Abstract:
We'll introduce ANI-AL molecular potentials, which are deep learning based potential energy functions for the fast and accurate prediction of quantum mechanical energies and forces of molecular systems. Thanks to GPU acceleration of training and inf ...Read More
Abstract:
We'll introduce ANI-AL molecular potentials, which are deep learning based potential energy functions for the fast and accurate prediction of quantum mechanical energies and forces of molecular systems. Thanks to GPU acceleration of training and inference, we successfully implement an automated sampling method that borrows techniques from active learning to automatically drive the systematic improvement of ANI-AL potentials. We'll also present results from applications of the ANI-AL potential in various problems relating to computational chemistry, such as molecular structure optimization, reaction path prediction, vibrational frequency calculation, and molecular dynamics simulations.  Back
 
Topics:
AI Application Deployment and Inference, Computational Biology and Chemistry
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8827
Streaming:
Share:
 
Abstract:
Come join us and learn how to build a data-centric GPU cluster for artificial intelligence. Mellanox is a leader in high-performance, scalable, low-latency network interconnects for both InfiniBand and Ethernet. We'll present the state of the art te ...Read More
Abstract:
Come join us and learn how to build a data-centric GPU cluster for artificial intelligence. Mellanox is a leader in high-performance, scalable, low-latency network interconnects for both InfiniBand and Ethernet. We'll present the state of the art techniques for distributed machine learning, and discuss what special requirements they impose on the system, followed by an overview of interconnect technologies used to scale and accelerate distributed machine learning including RDMA, NVIDIA's GPUDirect technology, and a special focus on the in-network computing SHARP technology used to accelerate large scale deployments in artificial intelligence and high performance computing.  Back
 
Topics:
AI Application Deployment and Inference, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8635
Streaming:
Share:
 
Abstract:
Voice commands, and advancements in automatic speech recognition algorithms, that help us interact conversationally with devices, appliances and services, are growing within our everyday environment. We will share some highlights and results from wor ...Read More
Abstract:
Voice commands, and advancements in automatic speech recognition algorithms, that help us interact conversationally with devices, appliances and services, are growing within our everyday environment. We will share some highlights and results from work scheduling optimizations in the Kaldi framework. The first part of the talk will describe results focused primarily on optimizing the DNN components of speech pipeline. We will then show results from a GPU optimized fast lattice decode algorithm to achieve high end to end throughput across the whole ASR pipeline from the acoustic model to the language model.  Back
 
Topics:
AI Application Deployment and Inference, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81034
Streaming:
Download:
Share:
 
Abstract:
We'll discuss a detailed scale-up method for accelerating deep learning-based object detection inference engine with INT8 by using NVIDIA's TensorRT. Previously, converting convolutional neural networks (CNNs) from 32-bit floating-point arithmetic ...Read More
Abstract:
We'll discuss a detailed scale-up method for accelerating deep learning-based object detection inference engine with INT8 by using NVIDIA's TensorRT. Previously, converting convolutional neural networks (CNNs) from 32-bit floating-point arithmetic (FP32) to 8-bit integer (INT8) for classification tasks has been researched. However, there is no solid work for accelerating CNN-based object detection tasks. We'll explain how to accelerate YOLO-v2, the state-of-the-art CNN-based object detector with TensorRT using INT8. We improved YOLO-v2 network for better acceleration and more accurate for surveillance and named our network SIDNet. We verified SIDNet on several benchmark object detection and intrusion detection datasets and confirmed that SIDNet with INT8 has only 1% accuracy drop compared with FP32 mode and is 5x faster than the original YOLO-v2 on NVIDIA Tesla P40.  Back
 
Topics:
AI Application Deployment and Inference, Telecommunications, Deep Learning and AI Frameworks, Computer Vision, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8296
Streaming:
Download:
Share:
 
Abstract:
We'll provide insights into how customer support built on the foundation of AI can help streamline customer support for large enterprises, especially manufacturers. With AI technologies like image recognition and natural language processing maturing ...Read More
Abstract:
We'll provide insights into how customer support built on the foundation of AI can help streamline customer support for large enterprises, especially manufacturers. With AI technologies like image recognition and natural language processing maturing, enterprises should strongly consider building an AI-based support platform, especially those with an omni-channel strategy. Delivering an amazing and differentiated user experience will lead to higher net promoter and customer satisfaction scores. By employing AI-based technologies, enterprises can reduce their contacts, consequently reducing their cost and contact. It will also help them sell more replacement parts online.  Back
 
Topics:
AI Application Deployment and Inference, NVIDIA Inception Program, Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8517
Streaming:
Download:
Share:
 
Abstract:
One of the tough aspect of Deep Neural Network resides in its behavior validation. Although actual driving should be achieved with physical cars to train the neural network, there is today no tool to appropriately prepare data acquisition campaign or ...Read More
Abstract:
One of the tough aspect of Deep Neural Network resides in its behavior validation. Although actual driving should be achieved with physical cars to train the neural network, there is today no tool to appropriately prepare data acquisition campaign or go through stress validation before further on-road testing and industrial deployment. This talk will show how hardware and software in the loop on 3DEXPERIENCE CATIA, can now be extended to AI in the loop, with the ability to activate the full system engineering simulation with the actual neural network meant to run in the autonomous vehicle, accurately reproducing the neural network inference and checking overall vehicle behavior in various conditions. Every stage from full 3D synthetic data ingest and real-time software simulation, through actual hardware in the loop validation both use cases leveraging TensorRT GPU inference can now consistently be proofed for appropriate in-depth understanding of the network reactions before it drives on the road. A POC showing TensorRT and DNN behavior validation will be presented in details, opening new opportunities to validate GPU inference but also compare actual performance impact versus CPU  Back
 
Topics:
AI Application Deployment and Inference, Product & Building Design
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8748
Streaming:
Download:
Share:
 
Abstract:
We'll show how GE combines extensive domain knowledge with modern deep learning techniques to build intelligent pipeline inspection systems. GE builds a variety of industrial inspection equipment from ultrasonic pipeline inspection gauges to large-s ...Read More
Abstract:
We'll show how GE combines extensive domain knowledge with modern deep learning techniques to build intelligent pipeline inspection systems. GE builds a variety of industrial inspection equipment from ultrasonic pipeline inspection gauges to large-scale CT scanners. As historical producers of hardware, GE is now leading the transformation of the industrial space by building intelligent ecosystems around industrial equipment and processes. Challenges in this space include the esoteric domain-specific nature of the data, as well as the risk averse nature of the industry. However, by leveraging deep learning on large amounts of inspection data, we have been able to build a production system that enhances the reliability and consistency of the inspection process.  Back
 
Topics:
AI Application Deployment and Inference, Industrial Inspection
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8657
Streaming:
Share:
 
Abstract:
We'll present a novel GPU implementation for batched GBM inferencing. We'll also present detailed performance comparison of our implementation against the state-of-the-art libraries such as XGBoost and Treelite. We'll then compare inference perfor ...Read More
Abstract:
We'll present a novel GPU implementation for batched GBM inferencing. We'll also present detailed performance comparison of our implementation against the state-of-the-art libraries such as XGBoost and Treelite. We'll then compare inference performance on various real-world datasets.  Back
 
Topics:
AI Application Deployment and Inference, Accelerated Analytics, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8873
Streaming:
Download:
Share:
 
Abstract:
We'll discuss why AI and machine learning are a natural fit for serverless computing and a general architecture for scalable and serverless machine learning in production. We'll discuss issues encountered during implementing our own on-demand scali ...Read More
Abstract:
We'll discuss why AI and machine learning are a natural fit for serverless computing and a general architecture for scalable and serverless machine learning in production. We'll discuss issues encountered during implementing our own on-demand scaling over GPU clusters, show how these apply to more general solutions, and present one possible vision for the future of cloud-based machine learning.  Back
 
Topics:
AI Application Deployment and Inference, NVIDIA Inception Program, Accelerated Analytics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8900
Streaming:
Download:
Share:
AI and DL Business Track (high level)
Presentation
Media
Abstract:
Innovation can take many forms, and led by varying stakeholders across an organization. One successful model is utilizing AI for Social Good to drive a proof-of-concept that will advance a critical strategic goal. The Data Science Bowl (DSB) is ...Read More
Abstract:

Innovation can take many forms, and led by varying stakeholders across an organization. One successful model is utilizing AI for Social Good to drive a proof-of-concept that will advance a critical strategic goal. The Data Science Bowl (DSB) is an ideal example, launched by Booz Allen Hamilton in 2014, it galvanizes thousands of data scientists to participate in competitions that will have have far reaching impact across key industries such as healthcare. This session will explore the DSB model, as well as look at other ways organizations are utilizing AI for Social Good to create business and industry transformation.

  Back
 
Topics:
AI and DL Business Track (high level), AI for Business
Type:
Panel
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8953
Streaming:
Download:
Share:
 
Abstract:
From healthcare to financial services to retail, businesses are seeing unprecedented levels of efficiencies and productivity, which will only continue to rise and transform how companies operate. This session will look at how Accenture as an ent ...Read More
Abstract:

From healthcare to financial services to retail, businesses are seeing unprecedented levels of efficiencies and productivity, which will only continue to rise and transform how companies operate. This session will look at how Accenture as an enterprise is optimizing itself in the age of AI, as well as how it guides its customers to success. A look at best practices, insights, and measurement to help the audience inform their AI roadmap and journey.

  Back
 
Topics:
AI and DL Business Track (high level), AI for Business
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8984
Streaming:
Download:
Share:
 
Abstract:
Advancements in deep learning are enabling enterprise companies to make meaningful impacts to bottom-line profits. Enterprises capture thousands of hours of customer phone call recordings per day. This voice data is extremely valuable because it cont ...Read More
Abstract:
Advancements in deep learning are enabling enterprise companies to make meaningful impacts to bottom-line profits. Enterprises capture thousands of hours of customer phone call recordings per day. This voice data is extremely valuable because it contains insights that the business can use to improve customer experience and operations. We'll follow Deepgram CEO Dr. Scott Stephenson's path from working in a particle physics lab two miles underground to founding a deep learning company for voice understanding. We'll describe applications of cutting-edge AI techniques to make enterprise voice datasets mineable for valuable business insights. Companies today use these insights to drive the bottom line.  Back
 
Topics:
AI and DL Business Track (high level), Telecommunications, Speech and Language Processing, NVIDIA Inception Program
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8274
Streaming:
Download:
Share:
 
Abstract:
We have all heard about Facial Expression and Recognition Systems (FERS) and emotion capture but curiosity looms large. Is it training sets born of Generative Adversarial Networks (GANs) along with GPU architectures that will catapult this technolog ...Read More
Abstract:
We have all heard about Facial Expression and Recognition Systems (FERS) and emotion capture but curiosity looms large. Is it training sets born of Generative Adversarial Networks (GANs) along with GPU architectures that will catapult this technology forward? To be sure, but, something much deeper - a revolution within Computer Science programs in the schools - will accelerate its arrival in consumer platforms. It's called Social Signal Processing and women technologists have a competitive advantage in inventing and enhancing the deep learning algorithms that will fuel it. Come and listen to an industry veteran with 28 years in Artificial Intelligence, including her driving Watson into consumer platforms and a graduate of Stanford University, bolstered by her solid research in Symbolic Systems, discuss their patent-pending technology in the exciting area of Social Signal Processing and FERS. They are both frequent speakers on the ethics of AI usage and will offer their thoughts about how this new class of technology offers a new deal for women to shape the future of AI.  Back
 
Topics:
AI and DL Business Track (high level), AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8939
Streaming:
Share:
 
Abstract:
An organization''s data science needs change dramatically as they move through stages of data science maturity--their ability to consume, adopt, and deploy advanced analytics solutions. Understanding the maturity stage of your organization will help ...Read More
Abstract:
An organization''s data science needs change dramatically as they move through stages of data science maturity--their ability to consume, adopt, and deploy advanced analytics solutions. Understanding the maturity stage of your organization will help you choose projects that can bring value, grow your ability to derive greater value in the future, and help you make good decisions when growing your data science team. A data scientist might be a journeyman model builder, or a data scientist consultant, or a software engineer, or a developer of new deep learning algorithms. The data scientist that would be successful in a mature organization may well fail in an organization new to data science. Hiring and growing data scientists based on skill sets in line with your data science maturity stage and maximizes your probability of success. We''ll discuss a framework to determine your level of data science readiness, explore a tool to assess the skill sets of data scientists, and find which skills can maximize your organization''s probability of success at each stage.  Back
 
Topics:
AI and DL Business Track (high level)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8954
Streaming:
Download:
Share:
 
Abstract:
We are still in the early stages of AI, and its impact on industries is already significant - from healthcare to financial services to retail. Businesses are seeing unprecedented levels of efficiencies and productivity, which will only continue to ri ...Read More
Abstract:
We are still in the early stages of AI, and its impact on industries is already significant - from healthcare to financial services to retail. Businesses are seeing unprecedented levels of efficiencies and productivity, which will only continue to rise and transform how companies operate. This session will explore the progress of AI adoption over the last year, the industries that are leaping ahead, new AI innovations that will serve cross-industry concerns, and what businesses should expect in terms of adoption maturity in 2018.  Back
 
Topics:
AI and DL Business Track (high level)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8952
Streaming:
Download:
Share:
 
Abstract:
Has your team developed an AI proof-of-concept with promising metrics? Next step is to broaden the scope to impact larger areas of the enterprise. With its unique challenges and complexities, scaling POCs across multiple business units is a significa ...Read More
Abstract:
Has your team developed an AI proof-of-concept with promising metrics? Next step is to broaden the scope to impact larger areas of the enterprise. With its unique challenges and complexities, scaling POCs across multiple business units is a significant part of any company''s AI roadmap. This session will look at best practices, insights and success, rooted in Element AI''s experience with enterprise customers.  Back
 
Topics:
AI and DL Business Track (high level), NVIDIA Inception Program
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8989
Streaming:
Download:
Share:
 
Abstract:
For enterprises daunted by the prospect of AI and investing in a new technology platform, the reality is that AI can leverage already-in-place big data and cloud strategies. This session will explore AI and deep learning use cases that are desig ...Read More
Abstract:

For enterprises daunted by the prospect of AI and investing in a new technology platform, the reality is that AI can leverage already-in-place big data and cloud strategies. This session will explore AI and deep learning use cases that are designed for ROI, and look at how success is being measured and optimized.

  Back
 
Topics:
AI and DL Business Track (high level), AI for Business
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8983
Streaming:
Share:
 
Abstract:
Get the latest information on how the proliferation of mobile, cloud, and IoT devices has brought us into a new era: The Extreme Data Economy. There''s a greater variety of data than ever before, and exponentially more of it, streaming in real time. ...Read More
Abstract:
Get the latest information on how the proliferation of mobile, cloud, and IoT devices has brought us into a new era: The Extreme Data Economy. There''s a greater variety of data than ever before, and exponentially more of it, streaming in real time. Across industries, companies are turning data into an asset, above and beyond any product or service they offer. But unprecedented agility is required to keep business in motion and succeed in this post-big data era. To enable this level of agility, companies are turning to instant insight engines that are powered by thousands of advanced GPU cores, bringing unparalleled speed, streaming data analysis, visual foresight, and machine learning to break through the old bottlenecks. Learn about new data-powered use cases you''ll need to address, as well as advances in computing technology, particularly accelerated parallel computing, that will translate data into instant insight to power business in motion.  Back
 
Topics:
AI and DL Business Track (high level), NVIDIA Inception Program
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8997
Streaming:
Download:
Share:
 
Abstract:
In this session, you will learn how Google Cloud helps enterprises make the most out of data, and deliver customer value. We will provide an in-depth overview of the Cloud AI and Data Analytics offering that helps enterprises manage their ML lifecycl ...Read More
Abstract:
In this session, you will learn how Google Cloud helps enterprises make the most out of data, and deliver customer value. We will provide an in-depth overview of the Cloud AI and Data Analytics offering that helps enterprises manage their ML lifecycle, from data ingestion to insights and prediction. We will also demonstrate some breakthrough solutions, like AutoML, that are making ML accessible to everyone.  Back
 
Topics:
AI and DL Business Track (high level)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8976
Streaming:
Share:
 
Abstract:
We'll examine business value drivers for artificial intelligence and machine learning in retail and consumer goods industries. Traditionally, traction in AI and ML has been in deep research, scientific, and technical communities. Retailers and consu ...Read More
Abstract:
We'll examine business value drivers for artificial intelligence and machine learning in retail and consumer goods industries. Traditionally, traction in AI and ML has been in deep research, scientific, and technical communities. Retailers and consumer products companies are finding great success applying AI and ML technology to distinct use cases and business challenges. Join us to hear project descriptions and customer examples where AI and ML can impact the business by increasing revenue, protecting margin, and improving consumer satisfaction.  Back
 
Topics:
AI and DL Business Track (high level), Virtual Reality and Augmented Reality, Consumer Engagement and Personalization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8131
Streaming:
Download:
Share:
 
Abstract:
We'll review three practical use cases of applying AI and deep learning in the marketing and retail industries. For each use case, we'll cover business situations, discuss potential approaches, and describe final solutions from both the AI and infr ...Read More
Abstract:
We'll review three practical use cases of applying AI and deep learning in the marketing and retail industries. For each use case, we'll cover business situations, discuss potential approaches, and describe final solutions from both the AI and infrastructural points of view. Attendees will learn about applications of AI and deep learning in marketing and advertising; AI readiness criteria; selecting the right AI and deep learning methods, infrastructure, and GPUs for specific use cases; and avoiding potential risks.  Back
 
Topics:
AI and DL Business Track (high level), Predictive Analytics for Retail, Consumer Engagement and Personalization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8265
Streaming:
Download:
Share:
 
Abstract:
Learn how recent advances in Earth observation are opening up a new exciting area for exploration of satellite image data with deep learning. Focusing on real-world scenarios, we will teach you how to analyze this exciting remote sensing data source ...Read More
Abstract:
Learn how recent advances in Earth observation are opening up a new exciting area for exploration of satellite image data with deep learning. Focusing on real-world scenarios, we will teach you how to analyze this exciting remote sensing data source with deep neural networks. An automated satellite image understanding is of high interest for various research fields and industry sectors such as the insurance, agriculture or investing industry. You will learn how to apply deep neural networks in natural disaster situations and for the classification of land-use, land-cover and building types.  Back
 
Topics:
AI and DL Business Track (high level), GIS, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81028
Download:
Share:
AI and DL Research
Presentation
Media
Abstract:
We will cover the techniques for training DNNs with Tensor Cores described in "S8923 - Training Neural Networks with Mixed Precision: Theory and Practice". These methods were introduced for AI processing with the Volta GPU architecture. T ...Read More
Abstract:
We will cover the techniques for training DNNs with Tensor Cores described in "S8923 - Training Neural Networks with Mixed Precision: Theory and Practice". These methods were introduced for AI processing with the Volta GPU architecture. Tensor Cores provide up to 120 TFlops throughput, mixing operations on IEEE half- and single-precision floats. Techniques used will include loss-scaling, master weights copy, and choosing the proper precision for a given operation. For each of TensorFlow and PyTorch we will describe a fp32 network definition and then demonstrate the same network using mixed precision techniques.  Back
 
Topics:
AI and DL Research, Algorithms and Numerical Techniques
Type:
Tutorial
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81012
Streaming:
Download:
Share:
 
Abstract:
This tutorial will cover the issues encountered when deploying NVIDIA DGX-1/DGXStation into secure environment. For security reasons, some installations require that systems be isolated from the internet or outside networks. Since most DGX-1 softwar ...Read More
Abstract:
This tutorial will cover the issues encountered when deploying NVIDIA DGX-1/DGXStation into secure environment. For security reasons, some installations require that systems be isolated from the internet or outside networks. Since most DGX-1 software updates are accomplished through an over-the-network process with NVIDIA servers, this session will walk the participants through how updates can be made by maintaining an intermediary server. This session will be a combination of lecture, live demos and along with detailed instructions.  Back
 
Topics:
AI and DL Research, Data Center and Cloud Infrastructure
Type:
Tutorial
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8568
Streaming:
Share:
 
Abstract:
In this session, participants will get a taste of state-of-the-art techniques for scaling Deep Learning on GPU clusters. We present SuperML, a general and efficient communication layer for machine learning, which can scale neural network training to ...Read More
Abstract:
In this session, participants will get a taste of state-of-the-art techniques for scaling Deep Learning on GPU clusters. We present SuperML, a general and efficient communication layer for machine learning, which can scale neural network training to hundreds of GPU nodes. SuperML builds on three main ideas: decentralization, which allows algorithms to converge without a centralized coordinator (parameter server) or all-to-all communication, communication quantization, which significantly speeds up point-to-point messaging, and structured sparsity, by which SuperML induces model updates which only have a limited number of non-zero entries. From the technical perspective, SuperML provides a new implementation of the classic MPI standard, re-designed and re-implemented to provide efficient support for quantization and sparsity. We illustrate the performance characteristics of SuperML on CSCS Piz Daint, Europe's most powerful supercomputer, and on Amazon EC2, improving upon other highly optimized implementations such as CrayMPI and NVIDIA NCCL.  Back
 
Topics:
AI and DL Research, Accelerated Analytics, HPC and Supercomputing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8668
Streaming:
Download:
Share:
 
Abstract:
The field of wireless engineering is on the cusp of a revolution, driven by deep learning, that will define the next paradigm in wireless system design. While wireless communications technology has advanced considerably since its invention in the 189 ...Read More
Abstract:
The field of wireless engineering is on the cusp of a revolution, driven by deep learning, that will define the next paradigm in wireless system design. While wireless communications technology has advanced considerably since its invention in the 1890s, the fundamental design methodology has remained unchanged throughout its history - expert engineers hand-designing radio systems for specific applications. Deep learning enables a new, radically different approach, where systems are learned from wireless channel data. As the world becomes more connected and the Internet of Things becomes a reality, it is difficult to overstate the enormity of the impact to both commercial and military systems. This talk will provide a high-level overview of deep learning applied to wireless communications, discuss the current state of the technology and research, and present a vision for the future of wireless engineering.  Back
 
Topics:
AI and DL Research, Telecommunications
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8791
Streaming:
Share:
 
Abstract:
We'll introduce the basic concept of domain adaptation and how to use adversarial training to achieve unsupervised domain adaptation. We'll then describe how the technique is used in two tasks: improving semantic segmentation across cities, and tr ...Read More
Abstract:
We'll introduce the basic concept of domain adaptation and how to use adversarial training to achieve unsupervised domain adaptation. We'll then describe how the technique is used in two tasks: improving semantic segmentation across cities, and transferring language style for image captioning. In particular, we combine domain adaptation with policy gradient-based reinforcement learning approach to transfer language style. The details and results of both tasks are published in ICCV 2017.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8200
Streaming:
Share:
 
Abstract:
We'll discuss applications of deep learning to radio frequency (RF) data including specific signal and digital modulation scheme classification, identification of nefarious activities, and a general overview of the unique challenges and solutions fo ...Read More
Abstract:
We'll discuss applications of deep learning to radio frequency (RF) data including specific signal and digital modulation scheme classification, identification of nefarious activities, and a general overview of the unique challenges and solutions for AI in this domain. With the ubiquity of RF communication signals in our lives, deep learning can be leveraged to ensure accurate signal transmission and safer communities.  Back
 
Topics:
AI and DL Research, Computational Physics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8826
Streaming:
Download:
Share:
 
Abstract:
We'll introduce attendees to a new deep learning approach to object-localization. Instead of bounding boxes, our network estimates the center pixel locations for a variable number of targets in a scene while simultaneously extracting a characteristi ...Read More
Abstract:
We'll introduce attendees to a new deep learning approach to object-localization. Instead of bounding boxes, our network estimates the center pixel locations for a variable number of targets in a scene while simultaneously extracting a characteristic feature-set. We'll outline the overall approach and describe the underlying network architecture and training. We'll also present the results of our network as applied to the cars overhead with context dataset and discuss the current and future possibilities of this approach.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8191
Streaming:
Share:
 
Abstract:
In this technical deep dive, get an in-depth look at the deep learning containers on NVIDIA GPU Cloud (NGC) and learn how they can simplify your AI projects. NVIDIA pre-integrates and optimizes the top deep learning frameworks such as TensorFlow, PyT ...Read More
Abstract:
In this technical deep dive, get an in-depth look at the deep learning containers on NVIDIA GPU Cloud (NGC) and learn how they can simplify your AI projects. NVIDIA pre-integrates and optimizes the top deep learning frameworks such as TensorFlow, PyTorch, and MXNet, and makes them available on NVIDIA GPU Cloud, removing time consuming do-it-yourself software integration. We'll look at the NVIDIA framework optimizations, such as reducing GPU memory overhead, improving multi-GPU scaling, and reducing latency. And we'll talk about the integration of runtimes and drivers in the containers to ensure the correct versions of software are working together for peak performance. You'll leave with an understanding of what make an NVIDIA GPU-optimized deep learning container tick.  Back
 
Topics:
AI and DL Research, Deep Learning and AI Frameworks, Data Center and Cloud Infrastructure
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8497
Streaming:
Download:
Share:
 
Abstract:
Matchbox is an open source PyTorch-based tool that lets users implement their deep learning models as imperative code that applies to individual data samples, then efficiently train and validate them on batched data using GPUs. By automatically keepi ...Read More
Abstract:
Matchbox is an open source PyTorch-based tool that lets users implement their deep learning models as imperative code that applies to individual data samples, then efficiently train and validate them on batched data using GPUs. By automatically keeping track of batch-level masking and padding and rewriting data-dependent control flow, Matchbox simplifies model code, eliminates a class of implementation bugs, and allows programmers to work directly at a more natural level of abstraction.  Back
 
Topics:
AI and DL Research, Deep Learning and AI Frameworks
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8977
Streaming:
Download:
Share:
 
Abstract:
We'll introduce new concepts and algorithms that apply deep learning to radio frequency (RF) data to advance the state of the art in signal processing and digital communications. With the ubiquity of wireless devices, the crowded RF spectrum ...Read More
Abstract:

We'll introduce new concepts and algorithms that apply deep learning to radio frequency (RF) data to advance the state of the art in signal processing and digital communications. With the ubiquity of wireless devices, the crowded RF spectrum poses challenges for cognitive radio and spectral monitoring applications. Furthermore, the RF modality presents unique processing challenges due to the complex-valued data representation, large data rates, and unique temporal structure. We'll present innovative deep learning architectures to address these challenges, which are informed by the latest academic research and our extensive experience building RF processing solutions. We'll also outline various strategies for pre-processing RF data to create feature-rich representations that can significantly improve performance of deep learning approaches in this domain. We'll discuss various use-cases for RF processing engines powered by deep learning that have direct applications to telecommunications, spectral monitoring, and the Internet of Things.

  Back
 
Topics:
AI and DL Research, Telecommunications, Federal
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8267
Streaming:
Download:
Share:
 
Abstract:
This presentation shows in-depth comparisons of several neural network models for 3D object classification. Object classification from 2D image is studied thoroughly and widely adopted during last few years by following the advances of deep neural ne ...Read More
Abstract:
This presentation shows in-depth comparisons of several neural network models for 3D object classification. Object classification from 2D image is studied thoroughly and widely adopted during last few years by following the advances of deep neural networks. From then, 3D object classification methods are actively studied, and yet not completely mature. Point cloud is most basic format of 3D objects. In this work, we present many neural network models that can be learned from 3D point cloud. It includes directly learning from 3D point cloud, projected 2D pixels, and voxelated volumes. This work uses Princeton ModelNet datasets and ShapeNetCore.v2 dataset, and then provides the comparisons of those neural network models.  Back
 
Topics:
AI and DL Research, Graphics and AI, Rendering and Ray Tracing, Real-Time Graphics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8453
Streaming:
Download:
Share:
 
Abstract:
We'll introduce GPU-accelerated unsupervised reinforcement and auxiliary learning (UNREAL) algorithm. Recent state-of-the-art deep reinforcement learning algorithms, such as A3C and UNREAL, are designed to train on a single device with only CPUs. Us ...Read More
Abstract:
We'll introduce GPU-accelerated unsupervised reinforcement and auxiliary learning (UNREAL) algorithm. Recent state-of-the-art deep reinforcement learning algorithms, such as A3C and UNREAL, are designed to train on a single device with only CPUs. Using GPU acceleration for these algorithms results in low GPU utilization, which means the full performance of the GPU is not reached. Motivated by the architecture changes made by the GA3C algorithm, which gave A3C better GPU acceleration, together with the high learning efficiency of the UNREAL algorithm, we extend GA3C with the auxiliary tasks from UNREAL to create GUNREAL. We show that our GUNREAL system finished training faster than UNREAL and reached higher scores than GA3C.  Back
 
Topics:
AI and DL Research, Performance Optimization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8219
Streaming:
Download:
Share:
 
Abstract:
To acquire rich repertoires of skills, robots must be able to learn from their own autonomously collected data. We'll describe a video-prediction model that predicts what a robot will see next, and show how this model can be used to solve complex ma ...Read More
Abstract:
To acquire rich repertoires of skills, robots must be able to learn from their own autonomously collected data. We'll describe a video-prediction model that predicts what a robot will see next, and show how this model can be used to solve complex manipulations tasks in real-world settings. Our model was trained on 44,000 video sequences, where the manipulator autonomously pushes various objects. Using the model, the robot is capable of moving objects that were not seen during training to desired locations, handling multiple objects and pushing objects around obstructions. Unlike other methods in robotic learning, video-prediction does not require any human labels. Our experiments show that the method achieves a significant advance in the range and complexity of skills that can be performed entirely with self-supervised robotic learning. This session is for attendees that possess a basic understanding of convolutional and recurrent neural networks.  Back
 
Topics:
AI and DL Research, IoT, Robotics & Drones, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8629
Streaming:
Download:
Share:
 
Abstract:
We'll discuss how we could use deep generative modeling in two application domains; in speech synthesis, and in sensor data modeling. We'll give an overview of what generative modeling is and how it could be used for practical AI tasks through the ...Read More
Abstract:
We'll discuss how we could use deep generative modeling in two application domains; in speech synthesis, and in sensor data modeling. We'll give an overview of what generative modeling is and how it could be used for practical AI tasks through these examples. We'll also give a flavor of latent space methods, which we can use to learn more about our data so as to transform them in meaningful ways, with uses in both reconstruction and in generation.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8617
Streaming:
Download:
Share:
 
Abstract:
The current generation AI systems are mostly moving towards dialogue generation and question answering. Human like conversation and dialogue based interaction has been proposed as the interface for tomorrow, which would obliterate key-boards and trac ...Read More
Abstract:
The current generation AI systems are mostly moving towards dialogue generation and question answering. Human like conversation and dialogue based interaction has been proposed as the interface for tomorrow, which would obliterate key-boards and track-pads from computers as we know them. We present two important current developments in these fields. First we talk about a neural dialogue generation system which can be deployed to engage humans in a multi-turn conversation. Next we talk about a segmented question answering module which can find answers from the web. The combination of these two techniques has the potential to unlock numerous new verticals, such as travel, retail etc. We will talk about the technical details as well as the higher level design choices.  Back
 
Topics:
AI and DL Research, Speech and Language Processing, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8151
Streaming:
Share:
 
Abstract:
We'll show how deep reinforcement learning can be greatly sped up by separating perception and action, with a reward function specified in terms of objects and their motions, which are supplied by the perceptual system. In the past five years, reinf ...Read More
Abstract:
We'll show how deep reinforcement learning can be greatly sped up by separating perception and action, with a reward function specified in terms of objects and their motions, which are supplied by the perceptual system. In the past five years, reinforcement learners have become vastly more powerful by incorporating deep learning techniques, playing Atari, Mario, Go, and other games with superhuman skill. However, these learners require vast amounts of training data to become skilled. For example, to master Pong, state-of-the-art reinforcement learners require tens of millions of game frames, equivalent to months of play time at human speed. We show that endowing the learner with a minimal perceptual system, capable of detecting and tracking objects, greatly reduces the number of frames needed for learning. This shifts the learning bottleneck from the amount of training data available to computations easily accelerated with GPUs.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8581
Streaming:
Download:
Share:
 
Abstract:
We'll describe the latest advances in neural machine translation from three different perspectives. We'll start with character-level, multilingual neural machine translation, which aims at harnessing positive language transfer among multiple langua ...Read More
Abstract:
We'll describe the latest advances in neural machine translation from three different perspectives. We'll start with character-level, multilingual neural machine translation, which aims at harnessing positive language transfer among multiple languages to improve the translation quality and the robustness of such a multilingual translation model to intra-sentence code-switching and typos. We'll then discuss the recent research on exploiting data beside oft-used parallel corpora. We'll discuss how another modality, such as vision, can be used to enable zero-resource machine translation, and how purely unsupervised neural machine translation can be done by exploiting the similarity between language distributions of two languages. Finally, we'll discuss a recent trend of retrieval-based approaches to deep learning with a specific example of non-parametric neural machine translation.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8609
Streaming:
Download:
Share:
 
Abstract:
We'll discuss ongoing work at NVIDIA on deep active learning. Attendees can expect to learn what active learning is and some of the challenges of applying it to deep neural network training.
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8692
Streaming:
Share:
 
Abstract:
We'll introduce a GAN-based framework for unsupervised image-to-image translation. It leverages a shared latent space assumption to learn to translate an image in one domain to a corresponding image in another domain without requiring any pair of co ...Read More
Abstract:
We'll introduce a GAN-based framework for unsupervised image-to-image translation. It leverages a shared latent space assumption to learn to translate an image in one domain to a corresponding image in another domain without requiring any pair of corresponding images in the two domains in the training dataset. We'll show examples on translating street scene images, from sunny day to rainy day or from day time to night time. We also show image translation results on dog breed conversions and cat species conversion as well as human face translation based on attributes.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8114
Streaming:
Download:
Share:
 
Abstract:
Reinforcement learning aims to determine a mapping from observations to actions that maximize a reward criterion. The agent starts off exploring the environment for rewards with random search, which is only likely to succeed in all but simplest of se ...Read More
Abstract:
Reinforcement learning aims to determine a mapping from observations to actions that maximize a reward criterion. The agent starts off exploring the environment for rewards with random search, which is only likely to succeed in all but simplest of settings. Furthermore, measuring and designing reward functions for real-world tasks is non-trivial. Inspired by research in developmental psychology, in this talk I will discuss how reinforcement learning agents might use curiosity and knowledge accumulated from experience for efficient exploration. I will present results illustrating an agent learning to play the game of Mario and learning to navigate without rewards, a study quantifying the kinds of prior knowledge used by humans for efficient exploration and some robotic manipulation experiments including the use of an anthropomorphic hand for grasping objects.   Back
 
Topics:
AI and DL Research, IoT, Robotics & Drones, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8217
Streaming:
Share:
 
Abstract:
To convert phonemes of telephone conversations and responses at meetings into texts in real time, pass the text to the computational model created by DGX-1, label with a learning without teacher, and add the clusters, we are developing a system which ...Read More
Abstract:
To convert phonemes of telephone conversations and responses at meetings into texts in real time, pass the text to the computational model created by DGX-1, label with a learning without teacher, and add the clusters, we are developing a system which compares objects and analyzes meaning of conversation and profiles of interlocutors. With this technology, customers can receive appropriate responses at the beginning of a conversation with a help desk, and patients can receive correspondence during a remote diagnosis with a doctor based solely off of their dialogue and examination results. By using TensorFlow as a platform and running the K-Means method, Word2vec, Doc2Vec, etc. in DGX-1 clustered environment on DGX-1, the result of arithmetic processing is found at high speed conversation. Even if the amount of sentences is increased, the learning effect increases linearly, demonstrating that the proportion of validity can be raised without taking grammar of languages ??other than English (e.g. Japanese) into account.  Back
 
Topics:
AI and DL Research, Speech and Language Processing, NVIDIA Inception Program
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8371
Streaming:
Download:
Share:
 
Abstract:
Building intelligent agents that possess the ability to perceive the rich visual environment around us, communicate this understanding in natural language to humans and other agents, and execute actions in a physical environment, has been a long-term ...Read More
Abstract:
Building intelligent agents that possess the ability to perceive the rich visual environment around us, communicate this understanding in natural language to humans and other agents, and execute actions in a physical environment, has been a long-term goal of Artificial Intelligence. In this talk, I will present my recent work on an instantiation of this goal -- Embodied Question Answering (EQA) -- where an agent that is spawned at a random location in an environment (a house or building) is asked a natural language question ("What color is the car?"). The agent perceives its environment through first-person vision and can perform a few 'atomic' actions: move-{forward, backward, right, left}, and turn-{right, left}. The objective of the agent is to explore the environment and gather visual information necessary to answer the question ("orange"). I'll introduce our OpenGL-based environments, a large-scale dataset of expert demonstrations for this task and deep models, trained end-to-end using reinforcement learning, from raw pixels to multi-step navigation control to visual question answering.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8582
Streaming:
Download:
Share:
 
Abstract:
Horovod makes it easy to train a single GPU TensorFlow model on many GPUs; both on a single server and across multiple servers. We'll cover Uber's explorations of distributed deep learning, how to use Horovod, and what kind of performance you ...Read More
Abstract:
Horovod makes it easy to train a single GPU TensorFlow model on many GPUs; both on a single server and across multiple servers. We'll cover Uber's explorations of distributed deep learning, how to use Horovod, and what kind of performance you can get on standard models, such as Inception V3 and ResNet-101. Learn how to speed up training of your TensorFlow model with Horovod.  Back
 
Topics:
AI and DL Research, Deep Learning and AI Frameworks, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8152
Streaming:
Share:
 
Abstract:
We'll present a unique framework for cross-modal image and sentence matching; namely selective multimodal long short-term memory (LSTM) that incorporates a new deep learning module as multimodal context-modulated attention network to selectively att ...Read More
Abstract:
We'll present a unique framework for cross-modal image and sentence matching; namely selective multimodal long short-term memory (LSTM) that incorporates a new deep learning module as multimodal context-modulated attention network to selectively attend to pairwise semantic concepts. In detail, effective image and sentence matching depends on measuring their global visual-semantic similarity. Based on the observation that such a global similarity arises from a complex aggregation of multiple local similarities between pairwise instances of image (objects) and sentence (words), we propose a selective multimodal LSTM network (sm-LSTM) for instance-aware image and sentence matching. The sm-LSTM includes a multimodal context-modulated attention scheme at each timestep that can selectively attend to a pair of instances of image and sentence by predicting pairwise instance-aware saliency maps for image and sentence. For selected pairwise instances, their representations are obtained based on the predicted saliency maps, and then compared to measure their local similarity. By similarly measuring multiple local similarities within a few timesteps, the sm-LSTM sequentially aggregate.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8281
Streaming:
Download:
Share:
 
Abstract:
We are witnessing unprecedented advances in computer vision and AI. What lies next for AI? We believe that the next generation of intelligent systems (say the next generation of Google's Assistant, Facebook's M, Apple's Siri, Amazon's Alexa) will ...Read More
Abstract:
We are witnessing unprecedented advances in computer vision and AI. What lies next for AI? We believe that the next generation of intelligent systems (say the next generation of Google's Assistant, Facebook's M, Apple's Siri, Amazon's Alexa) will need to possess the ability to perceive their environment (through vision, audition, or other sensors), communicate (i.e., hold a natural language dialog with humans and other agents), and act (e.g., aid humans by executing API calls or commands in a virtual or embodied environment), for tasks such as: aiding visually impaired users in understanding their surroundings; interacting with an AI assistant (Human: 'Alexa can you see the baby in the baby monitor?', AI: 'Yes, I can', Human: 'Is he sleeping or playing?'); robotics applications (e.g. search and rescue missions) where the operator may be situationally blind and operating via language. We'll present work from our lab on a range of projects on such visually grounded conversational agents.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8571
Streaming:
Download:
Share:
 
Abstract:
In this session we present a Kubernetes deployment on Amazon AWS GPUs that provide customized computer vision to a large number of users. Reza offers an overview of Matroid's pipeline and demonstrates how to customize computer vision neural network ...Read More
Abstract:
In this session we present a Kubernetes deployment on Amazon AWS GPUs that provide customized computer vision to a large number of users. Reza offers an overview of Matroid's pipeline and demonstrates how to customize computer vision neural network models in the browser, followed by building, training, and visualizing TensorFlow models, which are provided at scale to monitor video streams.  Back
 
Topics:
AI and DL Research, Data Center and Cloud Infrastructure, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8610
Streaming:
Download:
Share:
 
Abstract:
We'll explain the concept and the importance of audio recognition, which aims to understand literally all the information contained in the audio, not limiting its scope to speech recognition. It includes the introduction of various types of non ...Read More
Abstract:
We'll explain the concept and the importance of audio recognition, which aims to understand literally all the information contained in the audio, not limiting its scope to speech recognition. It includes the introduction of various types of non-verbal information contained in the audio such as acoustic scenes/events, speech, and music. This session is helpful to the people who are not familiar with audio processing but are interested in the context-aware system. Also, it might be inspiring for someone who develops AI applications such as AI home assistant, a humanoid robot, and self-driving cars. It also covers the potential use-cases and creative applications, including a video demonstration of the audio context-aware system applied to media-art performance for real-time music generation.  Back
 
Topics:
AI and DL Research, Speech and Language Processing, NVIDIA Inception Program, GIS
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8696
Streaming:
Download:
Share:
 
Abstract:
We'll present how deep reinforcement learning (DRL) and memory extended networks can be used to train agents, which optimize asset allocations or propose trading actions. The memory component is crucial for improved mini-batch parallelization and he ...Read More
Abstract:
We'll present how deep reinforcement learning (DRL) and memory extended networks can be used to train agents, which optimize asset allocations or propose trading actions. The memory component is crucial for improved mini-batch parallelization and helps mitigate catastrophic forgetting. We also address how concepts from risk-sensitive and safe reinforcement learning apply to improve the robustness of the learned policies. The DRL approach has several advantages over the industry standard approach, which is still based on the mean variance portfolio optimization. The most significant benefit is that the information bottleneck between the statistical return model and the portfolio optimizer is removed, and available market data and trade history are used much more efficiently.  Back
 
Topics:
AI and DL Research, Algorithms and Numerical Techniques, Advanced AI Learning Techniques (incl. GANs and NTMs), Finance
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8679
Streaming:
Download:
Share:
 
Abstract:
The paradigm for robot programming is changing with the adoption of the deep learning approach in the field of robotics. Instead of hard coding a complex sequence of actions, tasks are acquired by the robot through an active learning procedure. This ...Read More
Abstract:
The paradigm for robot programming is changing with the adoption of the deep learning approach in the field of robotics. Instead of hard coding a complex sequence of actions, tasks are acquired by the robot through an active learning procedure. This introduces new challenges that have to be solved to achieve effective training. We'll show several issues that can be encountered while learning a close-loop DNN controller aimed at a fundamental task like grasping, and their practical solutions. First, we'll illustrate the advantages of training using a simulator, as well as the effects of choosing different learning algorithms in the reinforcement learning and imitation learning domains. We'll then show how separating the control and vision modules in the DNN can simplify and speed up the learning procedure in the simulator, although the learned controller hardly generalizes to the real world environment. Finally, we'll demonstrate how to use domain transfer to train a DNN controller in a simulator that can be effectively employed to control a robot in the real world.  Back
 
Topics:
AI and DL Research, IoT, Robotics & Drones, Computer Vision, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8132
Streaming:
Download:
Share:
 
Abstract:
We'll cover the four known methods for emotion detection: vision, speech, sentiment analysis, and wearable technology. We'll provide a quick dive through each presented solution, and then introduce a novel approach aimed for the future of autonomou ...Read More
Abstract:
We'll cover the four known methods for emotion detection: vision, speech, sentiment analysis, and wearable technology. We'll provide a quick dive through each presented solution, and then introduce a novel approach aimed for the future of autonomous vehicles.  Back
 
Topics:
AI and DL Research, Consumer Engagement and Personalization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8352
Streaming:
Download:
Share:
 
Abstract:
Join a special presentation from our 2017-2018 Graduate Fellowship recipients to learn "what's next" out of the world of research and academia. Sponsored projects involve a variety of technical challenges, including distributed systems for ...Read More
Abstract:
Join a special presentation from our 2017-2018 Graduate Fellowship recipients to learn "what's next" out of the world of research and academia. Sponsored projects involve a variety of technical challenges, including distributed systems for large-scale deep learning; dynamic data structures for massively parallel machine learning; machine learning techniques for biomedical image analysis; visual dynamics; and compilation frameworks for high-performance graphics systems. We believe that these minds lead the future in our industry and we're proud to support the 2016-2017 NVIDIA Graduate Fellows. We'll also announce the 2017-2018 Graduate Fellows at this session. For more information on the NVIDIA Graduate Fellowship program, visit www.nvidia.com/fellowship.  Back
 
Topics:
AI and DL Research, Virtual Reality and Augmented Reality, Graphics and AI, Computational Biology and Chemistry, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8793
Streaming:
Share:
 
Abstract:
Estimation of 3D motion in a dynamic scene from a pair of images is a core task in many scene understanding problems. In real world applications, a dynamic scene is commonly captured by a moving camera (i.e., panning, tilting or hand-held), increasin ...Read More
Abstract:
Estimation of 3D motion in a dynamic scene from a pair of images is a core task in many scene understanding problems. In real world applications, a dynamic scene is commonly captured by a moving camera (i.e., panning, tilting or hand-held), increasing the task complexity because the scene is observed from different viewpoints. The main challenge is the disambiguation of the camera motion from scene motions, which becomes more difficult as the amount of rigid parts observed decreases. In this talk, We introduce a method to learn a rigidity of a scene from a large collection of dynamic scene data, and directly infer a rigidity mask from two sequential RGB-D images in a supervised manner. With the learned network, we show how we can effectively estimate camera motion and projected scene flow using computed 2D optical flow and the inferred rigidity mask. Through evaluations, we show that our methods can make the scene flow estimation more robust and stable over state-of-the-art methods in challenging dynamic scenes. The expected audiences will include people who are interested in computer vision algorithms, but not limited to any audiences interested in AI and machine learning in general. We'll cover: the motivation behind scene flow estimation, potential applications, how we train two networks for the scene flow estimation, and how we evaluate the algorithm with popular benchmark dataset, SINTEL. We'll also show a new semi-synthetic dataset and its generation method where we mix real video footage with virtually rendered foreground scenes.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8798
Streaming:
Download:
Share:
 
Abstract:
During this presentation we will review a deep neural network architecture and its training approaches used for producing high volume of estimations of travel times on a road graph with historical routes and traffic. This includes initial and continu ...Read More
Abstract:
During this presentation we will review a deep neural network architecture and its training approaches used for producing high volume of estimations of travel times on a road graph with historical routes and traffic. This includes initial and continuous online training, finding various sources to produce training data, challenges of quality control, and, of course, the invaluable role of GPU's for computation during both training and inference.  Back
 
Topics:
AI and DL Research, Product & Building Design, Intelligent Video Analytics and Smart Cities, GIS, Autonomous Vehicles
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8156
Streaming:
Download:
Share:
 
Abstract:
Recurrent neural networks are used in state-of-the-art models in domains such as speech recognition, machine translation, and language modeling. Sparsity is a technique to reduce compute and memory requirements of deep learning models. Sparse RNNs ar ...Read More
Abstract:
Recurrent neural networks are used in state-of-the-art models in domains such as speech recognition, machine translation, and language modeling. Sparsity is a technique to reduce compute and memory requirements of deep learning models. Sparse RNNs are easier to deploy on devices and high-end server processors. Even though sparse operations need less compute and memory relative to their dense counterparts, the speed-up observed by using sparse operations is less than expected on different hardware platforms. To address this issue, we prune blocks of weights in a layer instead of individual weights. Using these techniques, we can create block-sparse RNNs with sparsity ranging from 80% to 90% with a small loss in accuracy. This technique allows us to reduce the model size by 10x. Additionally, we can prune a larger dense network to recover this loss in accuracy while maintaining high block sparsity and reducing the overall parameter count. Our technique works with a variety of block sizes up to 32x32. Block-sparse RNNs eliminate overheads related to data storage and irregular memory accesses while increasing hardware efficiency compared to unstructured sparsity.  Back
 
Topics:
AI and DL Research, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8924
Streaming:
Download:
Share:
 
Abstract:
Constructing an accurate prediction model from limited data is one of the important tasks in machine learning. We'll introduce unsupervised domain adaptation and a learning method using interclass patterns as a method to construct accurate predictio ...Read More
Abstract:
Constructing an accurate prediction model from limited data is one of the important tasks in machine learning. We'll introduce unsupervised domain adaptation and a learning method using interclass patterns as a method to construct accurate prediction models from limited data. Regarding unsupervised domain adaptation, we use three networks asymmetrically. Two networks are used to label unlabeled target patterns, and one network is trained by the pseudo-labeled patterns to obtain target-discriminative representations. About the learning method using interclass patterns, we generate interclass patterns by mixing two patterns belonging to different classes with a random ratio and train the model to output the mixing ratio form the mixed patterns. Although the algorithm is very simple, the proposed method significantly improves classification performance on sound recognition and image recognition. In addition, we'll briefly introduce various topics, including WebDNN, which our team is working on.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8786
Streaming:
Download:
Share:
 
Abstract:
We'll focus on recent developments in deep learning-based generative models for image and video creation. The last two to three years have seen an explosive growth in the development of generative adversarial networks, variational autoencoders, and ...Read More
Abstract:
We'll focus on recent developments in deep learning-based generative models for image and video creation. The last two to three years have seen an explosive growth in the development of generative adversarial networks, variational autoencoders, and related autoregressive methods that have been made it possible to automatically generate images and videos, by harnessing the power of GPUs and deep learning libraries. These methods present interesting possibilities in automatic generation of datasets for training machine learning methods, as well as in real-world applications for image and video processing such as morphing, editing, advertising, design, and art. We'll present the technical details of these methods and recent results in various settings.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs), Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8784
Streaming:
Share:
 
Abstract:
Maps are a key component in image-based camera localization and visual SLAM systems: they are used to establish geometric constraints between images, correct drift in relative pose estimation, and relocalize cameras after lost tracking. The exact def ...Read More
Abstract:
Maps are a key component in image-based camera localization and visual SLAM systems: they are used to establish geometric constraints between images, correct drift in relative pose estimation, and relocalize cameras after lost tracking. The exact definitions of maps, however, are often application-specific and hand-crafted for different scenarios (e.g., 3D landmarks, lines, planes, bags of visual words). We propose to represent maps as a deep neural net called MapNet, which enables learning a data-driven map representation. Unlike prior work on learning maps, MapNet exploits cheap and ubiquitous sensory inputs like visual odometry and GPS in addition to images and fuses them together for camera localization. Geometric constraints expressed by these inputs, which have traditionally been used in bundle adjustment or pose-graph optimization, are formulated as loss terms in MapNet training and also used during inference. In addition to directly improving localization accuracy, this allows us to update the MapNet (i.e., maps) in a self-supervised manner using additional unlabeled video sequences from the scene.  Back
 
Topics:
AI and DL Research, Autonomous Vehicles, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8792
Streaming:
Share:
 
Abstract:
Densely connected neural networks were originally introduced to avoid the problem of layer-wise vanishing gradients when CNNs are stacked in a very deep fashion, specifically for image recognition tasks. Inspired by these works, we've explored the u ...Read More
Abstract:
Densely connected neural networks were originally introduced to avoid the problem of layer-wise vanishing gradients when CNNs are stacked in a very deep fashion, specifically for image recognition tasks. Inspired by these works, we've explored the use of dense networks connections within LSTM models for the task of automatic speech recognition. By introducing additional connections, to connect (almost) every layer to at least one other layer, we mitigate the vanishing gradient effect between LSTM layers and enable error signals to propagated back to the very first layer during training. In this presentation, we'll present the fundamentals of speech recognition and introduce different neural network model structures that have been shown to be effective for this task. We'll then introduce identity, highway, and dense connections and demonstrate how they improve the performance of these models. We'll evaluate the performance of these models across different datasets, and show that with a lattice-based system combination, densely connected LSTMs significantly contributed to reaching the marks of 5.0% and 9.1% in word error rate (WER) for the Switchboard and CallHome testsets.  Back
 
Topics:
AI and DL Research, Speech and Language Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8903
Streaming:
Share:
 
Abstract:
In this talk we will present four applications of deep learning in e-commerce. 1) A deep neural net architecture which has been successfully deployed as a large scale Visual Search and Recommendation system for e-commerce. The deployment has been at ...Read More
Abstract:
In this talk we will present four applications of deep learning in e-commerce. 1) A deep neural net architecture which has been successfully deployed as a large scale Visual Search and Recommendation system for e-commerce. The deployment has been at Flipkart, India's largest e-Commerce vendor, over a catalog of 50M products, supporting 2K queries per second. Our results beat state of the art on the on the Exact Street2Shop dataset. 2) Visual Semantic embedding of e-Commerce products for enhanced searchability and product ranking. 3) Neural Network based click prediction. 4) A novel neural network architecture for demand prediction.  Back
 
Topics:
AI and DL Research, Deep Learning and AI Frameworks, Consumer Engagement and Personalization, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8684
Streaming:
Share:
 
Abstract:
We'll discuss training techniques and deep learning architectures for high-precision landmark localization. In the first part of the session, we'll talk about ReCombinator Networks, which aims at maintaining pixel-level image information ...Read More
Abstract:

We'll discuss training techniques and deep learning architectures for high-precision landmark localization. In the first part of the session, we'll talk about ReCombinator Networks, which aims at maintaining pixel-level image information, for high-accuracy landmark localization. This model combines coarse-to-fine features to first observe global (coarse) image information and then recombines local (fine) information. By using this model, we report SOTA on three facial landmark datasets. This model can be used for other tasks that require pixel-level accuracy (for example, image segmentation, image-to-image translation). In the second part, we'll talk about improving landmark localization in a semi-supervised setting, where less labeled data is provided. Specifically, we consider a scenario where few labeled landmarks are given during training, but lots of weaker labels (for example, face emotions, hand gesture) that are easier to obtain are provided. We'll describe training techniques and model architectures that can leverage weaker labels to improve landmark localization.

  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8406
Streaming:
Download:
Share:
 
Abstract:
Using only randomized simulated images, we'll present a system to infer and simply execute a human-readable robotic program after watching a real-world task demonstration. The system is comprised of a series of deep neural network modules, each lear ...Read More
Abstract:
Using only randomized simulated images, we'll present a system to infer and simply execute a human-readable robotic program after watching a real-world task demonstration. The system is comprised of a series of deep neural network modules, each learned entirely in simulation. During training, images are generated in a gaming engine and made transferable to the real world by domain randomization. After training, the system is straightforwardly deployed on a real robot with no retuning of the neural networks and having never previously seen a real image. We demonstrate the system on a Baxter robot performing block tower construction tasks.  Back
 
Topics:
AI and DL Research, IoT, Robotics & Drones, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8439
Streaming:
Share:
 
Abstract:
Learn how to generate long answers for non-factoid questions in quality assurance community sites by using the encoder-decoder framework. We'll present our novel extension of the encoder-decoder framework, called the ensemble network, that goes beyo ...Read More
Abstract:
Learn how to generate long answers for non-factoid questions in quality assurance community sites by using the encoder-decoder framework. We'll present our novel extension of the encoder-decoder framework, called the ensemble network, that goes beyond a single short sentence. It handles several sentences (i.e. two major sentence types that organize answers for non-factoid questions, conclusion statements, and its supplementary ones) to generate complicated non-factoid answers.  Back
 
Topics:
AI and DL Research, Speech and Language Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8301
Streaming:
Share:
 
Abstract:
Robust object tracking requires knowledge and understanding of the object being tracked: its appearance, motion, and change over time. A tracker must be able to modify its underlying model and adapt to new observations. We present Re3, a real-ti ...Read More
Abstract:

Robust object tracking requires knowledge and understanding of the object being tracked: its appearance, motion, and change over time. A tracker must be able to modify its underlying model and adapt to new observations. We present Re3, a real-time deep object tracker capable of incorporating temporal information into its model. Rather than focusing on a limited set of objects or training a model at test-time to track a specific instance, we pretrain our generic tracker on a large variety of objects and efficiently update on the fly; Re3 simultaneously tracks and updates the appearance model with a single forward pass. This lightweight model is capable of tracking objects at 150 FPS, while attaining competitive results on challenging benchmarks. We also show that our method handles temporary occlusion better than other comparable trackers using experiments that directly measure performance on sequences with occlusion.

  Back
 
Topics:
AI and DL Research, Intelligent Video Analytics and Smart Cities, Autonomous Machines, Computer Vision, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8298
Streaming:
Share:
 
Abstract:
We''ll present a multi-node distributed deep learning framework called ChainerMN. Even though GPUs are continuously gaining more computation throughput, it is still very time-consuming to train state-of-the-art deep neural network models. For better ...Read More
Abstract:
We''ll present a multi-node distributed deep learning framework called ChainerMN. Even though GPUs are continuously gaining more computation throughput, it is still very time-consuming to train state-of-the-art deep neural network models. For better scalability and productivity, it is paramount to accelerate the training process by using multiple GPUs. To enable high-performance and flexible distributed training, ChainerMN was developed and built on top of Chainer. We''ll first introduce the basic approaches to distributed deep learning and then explain the design choice, basic usage, and implementation details of Chainer and ChainerMN. To demonstrate the scalability and efficiency of ChainerMN, we''ll discuss the remarkable results from training ResNet-50 classification model on ImageNet database using 1024 Tesla P100 GPUs and our in-house cluster, MN-1.    Back
 
Topics:
AI and DL Research, NVIDIA Inception Program, Deep Learning and AI Frameworks, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8889
Streaming:
Download:
Share:
 
Abstract:
To effectively leverage the progress in Artificial Intelligence (AI) to make our lives more productive, it is important for humans and AI to work well together in a team. Traditionally, research has focused primarily on making AI more accurate, and ( ...Read More
Abstract:
To effectively leverage the progress in Artificial Intelligence (AI) to make our lives more productive, it is important for humans and AI to work well together in a team. Traditionally, research has focused primarily on making AI more accurate, and (to a lesser extent) on having it better understand human intentions, tendencies, beliefs, and contexts. The latter involves making AI more human-like and having it develop a theory of our minds. In this talk, I will argue that for human-AI teams to be effective, humans must also develop a Theory of AI''s Mind get to know its strengths, weaknesses, beliefs, and quirks. I will present some (very) initial results in the context of visual question answering and visual dialog where the AI agent is trained to answer natural language questions about images.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8560
Streaming:
Download:
Share:
 
Abstract:
We''ll review our study of the use of artificial intelligence to augment various domains of computational science in order to improve time to solution for various HPC problems. We''ll discuss the current state-of-the-art approaches and performance ga ...Read More
Abstract:
We''ll review our study of the use of artificial intelligence to augment various domains of computational science in order to improve time to solution for various HPC problems. We''ll discuss the current state-of-the-art approaches and performance gains where applicable. We''ll also investigate current barriers to adoption and consider possible solutions.  Back
 
Topics:
AI and DL Research, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8242
Streaming:
Download:
Share:
 
Abstract:
We''ll explore how deep learning approaches can be used for perceiving and interpreting the driver''s state and behavior during manual, semi-autonomous, and fully-autonomous driving. We''ll cover how convolutional, recurr ...Read More
Abstract:

We''ll explore how deep learning approaches can be used for perceiving and interpreting the driver''s state and behavior during manual, semi-autonomous, and fully-autonomous driving. We''ll cover how convolutional, recurrent, and generative neural networks can be used for applications of glance classification, face recognition, cognitive load estimation, emotion recognition, drowsiness detection, body pose estimation, natural language processing, and activity recognition in a mixture of audio and video data.

  Back
 
Topics:
AI and DL Research, Autonomous Vehicles, Autonomous Driving
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8626
Streaming:
Share:
 
Abstract:
Learn how to predict a dense depth image from a sparse set of depth measurements and a single RGB image. This approach can be applied to serve as a plug-in module in simultaneous localization and mapping to convert sparse maps to dense maps, and as a ...Read More
Abstract:
Learn how to predict a dense depth image from a sparse set of depth measurements and a single RGB image. This approach can be applied to serve as a plug-in module in simultaneous localization and mapping to convert sparse maps to dense maps, and as a super-resolution of LiDAR depth data. We''ll describe the performance of our prediction method, explain how to train the depth prediction network, and showcase examples of its applications. Codes and video demonstration are also publicly available. This session is for registrants who are already familiar with basic machine learning techniques.  Back
 
Topics:
AI and DL Research, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8216
Streaming:
Share:
 
Abstract:
We''ll present a framework that can learn a compute-intensive deep neural networks (DNNs) task using multiple AI blocks and evolve better confidence by combining estimates. We''ll consider the example of establishing the identity of a user using spee ...Read More
Abstract:
We''ll present a framework that can learn a compute-intensive deep neural networks (DNNs) task using multiple AI blocks and evolve better confidence by combining estimates. We''ll consider the example of establishing the identity of a user using speech and image data. The system consists of two blocks - the AI block and Arbiter block. The AI block uses multiple DNNs (voice-based and image-based DNNs that generate a low confidence estimate initially). These AI blocks assist each other using Arbiter blocks and build confidence, improve accuracy, and learn salient features over time. Arbiter can store recent unacquainted data at run time in noisy and distorted environments and train the AI blocks periodically or on an on-demand basis. This concept could potentially improve the automatic speech recognition capabilities and allow detection of faces even when variable features of faces change with time. The GPU is the ideal choice as the task requires inferencing as well as training on the go.  Back
 
Topics:
AI and DL Research, Intelligent Video Analytics and Smart Cities, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8331
Streaming:
Download:
Share:
 
Abstract:
We have long envisioned that machines one day can perform human-like perception, reasoning, and expression across multiple modalities including vision and language, which will augment and transform the ways humans communicate with each other and with ...Read More
Abstract:
We have long envisioned that machines one day can perform human-like perception, reasoning, and expression across multiple modalities including vision and language, which will augment and transform the ways humans communicate with each other and with the real world. With this vision, we''ll introduce the latest work of developing a deep attention GAN for fine-grained language-to-image synthesis. We''ll discuss the open problems behind the task that we''re thrilled to solve, including image and language understanding, joint reasoning across both modalities, and expressing abstract concepts into full imagination, which are of fundamental importance to reaching general intelligence.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8867
Streaming:
Download:
Share:
 
Abstract:
We find 99.9 percent of the gradient exchange in distributed SGD is redundant, and we propose deep gradient compression (DGC) to greatly reduce the communication bandwidth and improve the scalability of distributed training. To preserve accuracy duri ...Read More
Abstract:
We find 99.9 percent of the gradient exchange in distributed SGD is redundant, and we propose deep gradient compression (DGC) to greatly reduce the communication bandwidth and improve the scalability of distributed training. To preserve accuracy during this compression, DGC employs four methods: momentum correction, local gradient clipping, momentum factor masking, and warm-up training. We have applied DGC to image classification, speech recognition, and language modeling with multiple datasets including Cifar10, ImageNet, Penn Treebank, and Librispeech Corpus. In all these scenarios, DGC achieves a gradient compression ratio from 270x to 600x without losing accuracy, cutting the gradient size of ResNet-50 from 97MB to 0.35MB, and for DeepSpeech from 488MB to 0.74MB. DGC enables large-scale distributed training on inexpensive commodity 1Gbps Ethernet and facilitates distributed training on mobile.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8607
Streaming:
Share:
 
Abstract:
In this talk, we will survey how Deep Learning methods can be applied to personalization and recommendations. We will cover why standard Deep Learning approaches don''t perform better than typical collaborative filtering techniques. Then ...Read More
Abstract:

In this talk, we will survey how Deep Learning methods can be applied to personalization and recommendations. We will cover why standard Deep Learning approaches don''t perform better than typical collaborative filtering techniques. Then we will survey we will go over recently published research at the intersection of Deep Learning and recommender systems, looking at how they integrate new types of data, explore new models, or change the recommendation problem statement. We will also highlight some of the ways that neural networks are used at Netflix and how we can use GPUs to train recommender systems. Finally, we will highlight promising new directions in this space.

  Back
 
Topics:
AI and DL Research, Consumer Engagement and Personalization, Deep Learning and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81011
Streaming:
Download:
Share:
 
Abstract:
We''ll talk about the challenges in a large-scale distributed, GPU-based deep learning, and propose an efficient communication algorithm to achieve state-of-the-art scalability. In detail, we''ll explain various ways to speed up GPU-based deep learni ...Read More
Abstract:
We''ll talk about the challenges in a large-scale distributed, GPU-based deep learning, and propose an efficient communication algorithm to achieve state-of-the-art scalability. In detail, we''ll explain various ways to speed up GPU-based deep learning, and motivate the large-scale deep-learning in the performance context. Then, we will state that efficient communication is a grand challenge in the large-scale deep-learning, especially with upcoming more powerful GPUs such as Volta architecture Tesla V100. We''ll present the technical details on a proposed communication algorithm along with the supporting data collected with more than 100 GPUs.  Back
 
Topics:
AI and DL Research, Deep Learning and AI Frameworks
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8479
Download:
Share:
 
Abstract:
The growth in density of housing in cities like London and New York has resulted in the higher demand for efficient smaller apartments. These designs challenge the use of space and function while trying to ensure the occupants have the perceptio ...Read More
Abstract:

The growth in density of housing in cities like London and New York has resulted in the higher demand for efficient smaller apartments. These designs challenge the use of space and function while trying to ensure the occupants have the perception of a larger space than provided. The process of designing these spaces has always been the responsibility and perception of a handful of designers using 2D and 3D static platforms as part of the overall building design and evaluation, typically constraint by a prescriptive program and functional requirement. A combination of human- and AI-based agents creating and testing these spaces through design and virtual immersive environments (NVIDIA Holodeck) will attempt to ensure the final results are efficient and best fit for human occupancy prior to construction.

  Back
 
Topics:
AI and DL Research, Virtual Reality and Augmented Reality
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8398
Streaming:
Download:
Share:
 
Abstract:
End-to-end learning is a powerful new strategy for training neural networks from perception to control. While such systems have been shown to perform well for reactionary control, the representation learned is not usable for higher level decision mak ...Read More
Abstract:
End-to-end learning is a powerful new strategy for training neural networks from perception to control. While such systems have been shown to perform well for reactionary control, the representation learned is not usable for higher level decision making, such as navigation. We''ll discuss the latest methodologies for training end-to-end systems for parallel autonomy, and demonstrate some of the shortcomings when such decision making capability is needed.  Back
 
Topics:
AI and DL Research, Autonomous Vehicles
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8605
Streaming:
Download:
Share:
 
Abstract:
Driver monitoring systems are used to detect many driver attributes like gaze, head pose, eye openness, and other features pertaining to attention and assistance. We''ll present a synthetic method of generating data for training DNNs, which caters to ...Read More
Abstract:
Driver monitoring systems are used to detect many driver attributes like gaze, head pose, eye openness, and other features pertaining to attention and assistance. We''ll present a synthetic method of generating data for training DNNs, which caters to the above mentioned features of the subject. We use blender for generating synthetic images, powered by NVIDIA GPUs, which can be scaled to match training needs. Synthetic data generatation allows precise control over data points that are difficult to control in a real environment, like pupil dialation. This approach avoids noisy measurements and results in high accuracy without the need for a high-precision 3D sensor.  Back
 
Topics:
AI and DL Research, Autonomous Vehicles, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8324
Streaming:
Download:
Share:
 
Abstract:
Go beyond working with a single sensor and enter the realm of Intelligent Multi-Sensor Analytics (IMSA). We''ll introduce concepts and methods for using deep learning with multi-sensor, or heterogenous, data. There are many resources and ...Read More
Abstract:

Go beyond working with a single sensor and enter the realm of Intelligent Multi-Sensor Analytics (IMSA). We''ll introduce concepts and methods for using deep learning with multi-sensor, or heterogenous, data. There are many resources and examples available for learning how to leverage deep learning with public imagery datasets. However, few resources exist to demonstrate how to combine and use these techniques to process multi-sensor data. As an example, we''ll introduce some basic methods for using deep learning to process radio frequency (RF) signals and make it a part of your intelligent video analytics solutions. We''ll also introduce methods for adapting existing deep learning frameworks for multiple sensor signal types (for example, RF, acoustic, and radar). We''ll share multiple use cases and examples for leveraging IMSA in smart city, telecommunications, and security applications.

  Back
 
Topics:
AI and DL Research, Intelligent Video Analytics and Smart Cities, Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8260
Streaming:
Download:
Share:
 
Abstract:
Many scientific and engineering fields increasingly rely on complex and time consuming computational simulation as part of the modern scientific workflow. In many applications, such as High Energy Particle Physics, Cosmology, Geophysics, and others, ...Read More
Abstract:
Many scientific and engineering fields increasingly rely on complex and time consuming computational simulation as part of the modern scientific workflow. In many applications, such as High Energy Particle Physics, Cosmology, Geophysics, and others, simulations are the computational bottleneck for producing and testing results. We introduce the usage of Generative Adversarial Networks (GAN) as a potential tool for speeding up expensive theoretical models and simulations in scientific and engineering applications, ushering in a new era of deep learning-powered scientific discovery. We will show that using a GAN-based High Energy Physics fast simulator on GPUs can provide speedups of up to 100,000x when compared to traditional simulation software, while retaining high levels of precision. Finally, we will discuss modeling and architectural considerations in this domain with the hope of directly empowering scientists and engineers in other fields to experiment with Generative Adversarial Networks in order to speed up simulation across scientific domains.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs), HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81001
Streaming:
Download:
Share:
 
Abstract:
Deep reinforcement learning (deep RL) has emerged as a promising direction for autonomous acquisition of complex behaviors due to its ability to process complex sensory input and to acquire elaborate behavior skills, using general-purpose neural netw ...Read More
Abstract:
Deep reinforcement learning (deep RL) has emerged as a promising direction for autonomous acquisition of complex behaviors due to its ability to process complex sensory input and to acquire elaborate behavior skills, using general-purpose neural network representations. Since learning expressive function approximators requires large quantities of data, deep RL has been mostly applied to simulated domains, such as video games and simulated robotic locomotion and manipulation tasks, where the data collection can occur faster than real time and be trivially parallelized. We''ll address techniques that have been proposed to enable deep RL for real-world robotics, and discuss how the maximum-entropy principle can be leveraged to reduce the required amount of real-world interaction.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8603
Streaming:
Download:
Share:
 
Speakers:
Abstract:
Designing neural network architectures are critical for deep learning applications, but it is so complex and depends on AI experts. We''ll demonstrate how you can learn how to construct neural networks automatically without the human intervention. Th ...Read More
Abstract:
Designing neural network architectures are critical for deep learning applications, but it is so complex and depends on AI experts. We''ll demonstrate how you can learn how to construct neural networks automatically without the human intervention. There are two fundamental limiters to the performance of auto-generated neural networks: accuracy and efficiency, which is caused by searching overhead. We''ll also explore new techniques to make auto-generated neural network methods accurate and efficient, including: end-to-end technology to construct neural network within reinforcement learning, adaptive random search and bayesian optimization framework for different AI domains, such as computer vision, IoT acoustics, NLP and finance; using historical knowledge bases to reduce the searching overhead; and scheduling the execution of searching tasks over multiple NVIDIA GPUs to speed up the searching process. Also, we''ll give both the theoretical analysis and experiment results, which show significant improvement of accuracy and substantial reduction of searching time.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8234
Streaming:
Download:
Share:
 
Abstract:
We''ll introduce several attempts for modeling the long-term sequence dependence to help improve the action recognition performance. First, we''ll introduce a fused feature of deep and hand-crafted features to prove the complementation between them. ...Read More
Abstract:
We''ll introduce several attempts for modeling the long-term sequence dependence to help improve the action recognition performance. First, we''ll introduce a fused feature of deep and hand-crafted features to prove the complementation between them. We''ll also introduce an attempt of attention model to illustrate the effectiveness of attention mechanism on action recognition. We''ll then introduce shuttleNet, which is a biologically-inspired neural network. Finally, we''ll give some divergent experiments on action recognition to show the potential research direction.  Back
 
Topics:
AI and DL Research, Computer Vision, Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8229
Streaming:
Download:
Share:
 
Abstract:
As the race to full autonomy accelerates, the in-cab transportation experience is also being redefined. Future vehicles will sense the passengers'' identities and activities, as well as their cognitive and emotional states, to adapt and ...Read More
Abstract:

As the race to full autonomy accelerates, the in-cab transportation experience is also being redefined. Future vehicles will sense the passengers'' identities and activities, as well as their cognitive and emotional states, to adapt and optimize their experience. AI capable of interpreting what we call "people analytics" captured through their facial and vocal expressions, and aspects of the context that surrounds them will power these advances. We''ll give an overview of our Emotion AI solution, and describe how we employ techniques like deep learning-based spatio-temporal modeling. By combining these techniques with a large-scale dataset, we can develop AI capable of redefining the in-cab experience.

  Back
 
Topics:
AI and DL Research, NVIDIA Inception Program, Deep Learning and AI Frameworks, Autonomous Vehicles
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8758
Streaming:
Download:
Share:
 
Abstract:
In this talk we will discuss the work Columbia University, in partnership with NYC government, is using deep learning and GPUs to develop smart city traffic management facilitating support for navigation/movement of multitude of vehicles (including a ...Read More
Abstract:
In this talk we will discuss the work Columbia University, in partnership with NYC government, is using deep learning and GPUs to develop smart city traffic management facilitating support for navigation/movement of multitude of vehicles (including autonomous cars) in dense urban environments with many pedestrians. We will describe our work in real-time tracking of cars and pedestrians, prediction of movement based on historical observations of the intersection, backed by ultra-low latency wireless communications and edge computing nodes.  Back
 
Topics:
AI and DL Research, Intelligent Video Analytics and Smart Cities, Autonomous Vehicles
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8201
Streaming:
Share:
 
Abstract:
We''ll discuss recent research in deep reinforcement learning (RL), with a focus on the application of intuitions, from planning to neural network architectures for deep RL. Planning in complex visual environments has thus far been held back by the d ...Read More
Abstract:
We''ll discuss recent research in deep reinforcement learning (RL), with a focus on the application of intuitions, from planning to neural network architectures for deep RL. Planning in complex visual environments has thus far been held back by the difficulty of learning accurate predictive models. To address this, we embedded a model inside a differentiable, dynamically-constructed tree-planning architecture, so that we identify an effective model when used within that planner. We''ll share our work on developing these architectures, as well as our approaches to various technical obstacles associated with the efficient optimization of deep tree-structured models on GPU.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8787
Streaming:
Download:
Share:
 
Abstract:
This session will describe an approach to building personalized recommendations using (very) deep autoencoders. We will explore effects of different activation functions, network depth and novel algorithmic approaches. The model is trained end-to-end ...Read More
Abstract:
This session will describe an approach to building personalized recommendations using (very) deep autoencoders. We will explore effects of different activation functions, network depth and novel algorithmic approaches. The model is trained end-to-end without any layer-wise pre-training and our PyTorch-based code is publicly available.  Back
 
Topics:
AI and DL Research, Consumer Engagement and Personalization
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8212
Streaming:
Download:
Share:
 
Abstract:
Learn how VUE.ai''s model generator uses conditional GANs to produce product-specific images suitable for replacing photographs in catalogs. We''ll present networks that generate images of fashion models wearing specific garments, using an image of t ...Read More
Abstract:
Learn how VUE.ai''s model generator uses conditional GANs to produce product-specific images suitable for replacing photographs in catalogs. We''ll present networks that generate images of fashion models wearing specific garments, using an image of the garment as a conditioning variable. Network architecture variants, training, and manipulation of latent variables to control attributes such as model pose, build, or skin color will be addressed.  Back
 
Topics:
AI and DL Research, NVIDIA Inception Program, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8776
Streaming:
Download:
Share:
 
Abstract:
We provide a unified framework on learning affinity in pure data-driven fashion using a linear propagation structure. This is a GPU and deep learning friendly pairwise learning module that does not require solving linear equation, iterative inference ...Read More
Abstract:
We provide a unified framework on learning affinity in pure data-driven fashion using a linear propagation structure. This is a GPU and deep learning friendly pairwise learning module that does not require solving linear equation, iterative inferences or manually defined kernels. Specifically, we develop a three-way connection for the linear propagation model, which formulates a sparse transformation matrix, where all elements can be the output from a deep CNN, but results in a dense affinity matrix that effectively models any task-specific pairwise similarity matrix. The spatial propagation network can be applied to many affinity-related tasks, such as image matting, segmentation and colorization, to name a few. Essentially, the model can learn semantically aware affinity relations for high-level vision tasks due to the powerful learning capability of the deep CNN. We validate the framework on the task of refinement for image segmentation boundaries. Experiments on face parsing and semantic segmentation tasks show that the spatial propagation network provides a general, effective, and efficient solution for generating high-quality segmentation results.  Back
 
Topics:
AI and DL Research, Computer Vision, Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8312
Streaming:
Download:
Share:
 
Abstract:
Deep learning creates advances following a virtuous recipe: model architecture search, creating large training datasets, and scaling computation. Baidu Research''s Silicon Valley AI Lab develops state-of-the-art conversational user interfaces followi ...Read More
Abstract:
Deep learning creates advances following a virtuous recipe: model architecture search, creating large training datasets, and scaling computation. Baidu Research''s Silicon Valley AI Lab develops state-of-the-art conversational user interfaces following this DL recipe. We research new model architectures and features for speech recognition (Deep Speech 3), speech generation (Deep Voice 3), and natural language processing. To deploy these models in impactful products, we want a deep understanding of how recipe components coordinate to drive accuracy improvements. Through large-scale empirical studies, we find intriguing results about how deep learning is likely to scale: As training set size increases, DL model generalization error and model sizes scale as particular power-law relationships. For a fixed dataset size, as model size grows, training time remains roughly constant -- larger models require fewer steps to converge to the same accuracy. These scaling relationships have significant implications on DL research, practice, and systems. They can assist model debugging, setting accuracy targets, and decisions about dataset growth and future computing system design.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8899
Streaming:
Download:
Share:
 
Abstract:
Training AI agents that can successfully generalize requires large amounts of diverse labeled training data. Collecting and labeling data is a significant cost in the development of AI applications, which, in some cases, may not even be feasib ...Read More
Abstract:
Training AI agents that can successfully generalize requires large amounts of diverse labeled training data. Collecting and labeling data is a significant cost in the development of AI applications, which, in some cases, may not even be feasible. We'll describe computer graphics facial models that we are developing to generate large labeled synthetic facial data for training deep neural networks. Facial analysis is central to many vision applications that involve human-computer interaction, including robotics, autonomous cars, rehabilitation, and extended usability. Generating and animating human faces with high realism is a well-studied problem in computer graphics; however, very few computer vision AI techniques take advantage of rendered facial data to augment or replace manually collected training data. We'll share key insights of how we successfully use synthetic facial data for training facial analysis classifiers. We'll also demonstrate many sub-tasks on which synthetic data helps to significantly improve accuracy and reduces the need for manual data collection.
 
  Back
 
Topics:
AI and DL Research, Intelligent Video Analytics and Smart Cities
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8794
Streaming:
Share:
 
Abstract:
We propose a novel computational strategy based on deep and reinforcement learning techniques for de-novo design of molecules with desired properties. This strategy integrates two deep neural networks generative and predictive to generate novel c ...Read More
Abstract:
We propose a novel computational strategy based on deep and reinforcement learning techniques for de-novo design of molecules with desired properties. This strategy integrates two deep neural networks generative and predictive to generate novel chemical structures with the desired properties. In the first phase of the method, generative and predictive models are separately trained with supervised learning algorithms. In the second phase, both models are jointly trained with reinforcement learning approach to bias newly generated chemical structures towards those with desired physical and biological properties. In this proof-of-concept study, we have employed this strategy to design chemical libraries biased toward compounds with either maximal, minimal, or specific range of physical properties, such as melting point and hydrophobicity, as well as to develop novel putative inhibitors of JAK2. This new approach can find a general use for generating targeted chemical libraries optimized for a single desired property or multiple properties.  Back
 
Topics:
AI and DL Research, Computational Biology and Chemistry
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8254
Streaming:
Download:
Share:
 
Abstract:
We discuss one of the first attempts to teach computers to imagine or generate videos with controlled content using deep learning generative modeling techniques. To this end, we assume visual information in a natural video can be decomposed into two ...Read More
Abstract:
We discuss one of the first attempts to teach computers to imagine or generate videos with controlled content using deep learning generative modeling techniques. To this end, we assume visual information in a natural video can be decomposed into two major components: content and motion. While content encodes the objects present in the video, motion encodes the object dynamics. Based on this prior, we propose the motion and content decomposed generative adversarial network (MoCoGAN) framework for video generation. The proposed framework generates a video clip by sequentially mapping random noise vectors to video frames. We divide a random noise vector into content and motion parts. By controlling these parts we generate both the content of the video and the action that is being performed. We perform quantitative and qualitative analysis on several video datasets, including artificial shape motion, facial expression, and tai-chi videos.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs), Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8477
Streaming:
Share:
 
Abstract:
Spin up a deep learning (DL) proof-of-concept on a budget. We'll walk you through a DL workflow in the cloud leveraging DIGITS, then download a trained model, and run inference on a Jetson TX2. This session considers multiple options such as Nimbix, ...Read More
Abstract:
Spin up a deep learning (DL) proof-of-concept on a budget. We'll walk you through a DL workflow in the cloud leveraging DIGITS, then download a trained model, and run inference on a Jetson TX2. This session considers multiple options such as Nimbix, AMI, and NGC on Tesla P100, Tesla V100, and NVIDIA DGX-1 servers. This tutorial will be a combination of lecture, live demos, and detailed instructions.  Back
 
Topics:
AI and DL Research, Accelerated Analytics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8286
Download:
Share:
 
Abstract:
We'll discuss the development of a novel model for video prediction and analysis -- the parallel multi-dimensional long short-term memory (PMD-LSTM). PMD-LSTM is a general model for learning from higher dimensional data such as images, videos, and b ...Read More
Abstract:
We'll discuss the development of a novel model for video prediction and analysis -- the parallel multi-dimensional long short-term memory (PMD-LSTM). PMD-LSTM is a general model for learning from higher dimensional data such as images, videos, and biomedical scans. It is an extension of the popular LSTM recurrent neural networks to higher dimensional data with a rearrangement of the recurrent connections to dramatically increase parallelism. This gives the network the ability to compactly model the effect of long-range context in each layer, unlike convolutional networks, which need several layers to cover a larger input context. We'll discuss the blind spot problem in recent work on video prediction, and show how PMD-LSTM based models are fully context-aware for each predicted pixel. These models outperform comparatively complex state-of-the-art approaches significantly in a variety of challenging video prediction scenarios such as car driving, human motion, and diverse human actions.  Back
 
Topics:
AI and DL Research, NVIDIA Inception Program, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8713
Streaming:
Download:
Share:
 
Abstract:
We'll present the results of the SpaceNet 2017-2018 Challenge, preview future SpaceNet Challenges, and how developers can generally access open labeled satellite image training data through SpaceNet on AWS. To date, three SpaceNet Challenges ha ...Read More
Abstract:
We'll present the results of the SpaceNet 2017-2018 Challenge, preview future SpaceNet Challenges, and how developers can generally access open labeled satellite image training data through SpaceNet on AWS. To date, three SpaceNet Challenges have been designed to apply computer vision techniques to satellite imagery which examine building footprint extraction, road network extraction, and off-nadir object detection. SpaceNet on AWS is an online repository of openly available satellite imagery, co-registered map data to train algorithms for developers and data scientists to access for research. This first-of-its-kind open innovation project for the geospatial industry launched in August 2016 as a collaboration between AWS, CosmiQ Works, DigitalGlobe, and NVIDIA. The SpaceNet Roads Challenge, launching in November, builds on labeled training datasets consisting of building footprints across Khartoum, Las Vegas, Paris, and Shanghai by providing over 8,000 km of mapped road networks. It uses a novel metric motivated by graph theory concepts that focused competitors on routing rather than just static road pixel identification.  Back
 
Topics:
AI and DL Research, GIS
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8553
Streaming:
Share:
 
Abstract:
We'll cover the theory and practice for training DNNs with Tensor Cores, introduced for AI processing with the Volta GPU architecture. Tensor Cores provide up to 120 TFlops throughput, mixing operations on IEEE half- and single-precision floats. In ...Read More
Abstract:
We'll cover the theory and practice for training DNNs with Tensor Cores, introduced for AI processing with the Volta GPU architecture. Tensor Cores provide up to 120 TFlops throughput, mixing operations on IEEE half- and single-precision floats. In the theory portion of the talk, we'll review the half-precision format, values that arise in DNN computations, and techniques that maximize utilization of fp16 format by these values. Techniques include loss-scaling, master weights, and choosing the proper precision for a given operation. In the practice portion of this talk, we'll survey various models that have been trained in mixed precision, matching the accuracy of fp32 training sessions while using the same hyperparameters. Models include various architectures (feed forward, recurrent, generative) as well as cover diverse tasks (image, speech, and language processing). We'll also provide network design and training guidelines to maximize speed when using Tensor Cores.  Back
 
Topics:
AI and DL Research, Algorithms and Numerical Techniques
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8923
Streaming:
Download:
Share:
 
Abstract:
At NVIDIA, we're busy applying deep learning to diverse problems, and this talk will give an overview of a few of these applications. We'll discuss our resume matching system, which helps match candidates to job openings at NVIDIA, as well as an op ...Read More
Abstract:
At NVIDIA, we're busy applying deep learning to diverse problems, and this talk will give an overview of a few of these applications. We'll discuss our resume matching system, which helps match candidates to job openings at NVIDIA, as well as an open-source sentiment analysis project trained on unsupervised text that is improving our marketing capabilities. We'll discuss a blind image quality metric that we're using to lower the cost of raytracing photorealistic graphics, and a generative model that we've built to create realistic graphics from simplistic sketches.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8672
Streaming:
Share:
 
Abstract:
AI is one of the most rapidly-evolving areas of computer science today and datascientists are constantly pushing the boundaries of the possible -- wanting to explore new data types, new algorithms, and diverse and heterogenous models. In this talk we ...Read More
Abstract:
AI is one of the most rapidly-evolving areas of computer science today and datascientists are constantly pushing the boundaries of the possible -- wanting to explore new data types, new algorithms, and diverse and heterogenous models. In this talk we'll explore two key productivity factors for datascience -- first, speed and the ability to explore many models and sets of data quickly; and second, ability to explore broad types of models, incorporating both machine learning and deep learning. We will talk about results of 40x and 50x productivity through system+software co-design and novel algorithms which leverage Power Systems and GPUs for both deep learning and key areas of classical machine learning. System+software co-design and co-optimization can result in dramatic efficiency improvements, enable creation of large models, exploration of large datasets, and realize productivity gains for datascientists, freeing them up to focus on the fundamental science of deep and machine learning -- gaining accuracy, functionality, and generalizability of their models.  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81025
Streaming:
Share:
 
Abstract:
Recent advances in earth observation are opening up a new exciting area for exploration of satellite image data. We'll teach you how to analyse this new data source with deep neural networks. Focusing on emergency response, you will learn how to app ...Read More
Abstract:
Recent advances in earth observation are opening up a new exciting area for exploration of satellite image data. We'll teach you how to analyse this new data source with deep neural networks. Focusing on emergency response, you will learn how to apply deep neural networks for semantic segmentation on satellite imagery. We will specifically focus on multimodal segmentation and the challenge of overcoming missing modality information during inference time. It is assumed that registrants are already familiar with fundamentals of deep neural networks.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8596
Streaming:
Download:
Share:
 
Abstract:
This session will present a proof of concept where a deep neural network was trained with pairs of Iray ray traced images (one arbitrary ray tracing iteration number and one fully converged image) and theirs structural similarity index (SSIM). Origin ...Read More
Abstract:
This session will present a proof of concept where a deep neural network was trained with pairs of Iray ray traced images (one arbitrary ray tracing iteration number and one fully converged image) and theirs structural similarity index (SSIM). Originally thought as a method for measuring the similarity between two images, SSIM index can also be viewed as a quality measure versus a reference image or, in our case, as a ray tracing rendering progress. The DNN can now from any render iteration of arbitrary scene infer a rendering progress estimator but also provides heat map pictures of the scenes that can be used for adaptive rendering, focusing ray tracing engine power on appropriate zones.  Back
 
Topics:
AI and DL Research, Graphics and AI, Rendering and Ray Tracing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8788
Streaming:
Download:
Share:
 
Abstract:
We'll showcase how you can apply a wealth of unlabeled image data to significantly improve accuracy and speed of single-shot object-detection (SSD) techniques. Our approach, SSD++, advances the state-of-the-art of single shot multibox-based object d ...Read More
Abstract:
We'll showcase how you can apply a wealth of unlabeled image data to significantly improve accuracy and speed of single-shot object-detection (SSD) techniques. Our approach, SSD++, advances the state-of-the-art of single shot multibox-based object detectors (such as SSD, YOLO) by employing a novel combination of convolution-deconvolution networks to learn robust feature maps, thus making use of unlabeled dataset, and the fresh approach to have confluence of convolution and deconvolution features to combine generic as well as semantically rich feature maps. As a result, SSD++ drastically reduces the requirement of labeled datasets, works on low-end GPUs, identifies small as well as large objects with high fidelity, and speeds up inference process by decreasing the requirement of default boxes. SSD++ achieves state-of-the-art results on PASCAL VOC and MS COCO datasets. Through ablation study, we'll explain the effectiveness of different components of our architecture that help us achieve improved accuracy on the above datasets. We'll further show a case study of SSD++ to identify shoppable objects in fashion, home decor, and food industry from images in the wild.  Back
 
Topics:
AI and DL Research, NVIDIA Inception Program, Computer Vision, Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8159
Streaming:
Share:
 
Abstract:
Learn how to apply deep learning technologies for building robust and scalable dialogue systems with deeper understanding of the classic pipelines as well as detailed knowledge on the benchmark of the models of the prior work. We'll start with an ov ...Read More
Abstract:
Learn how to apply deep learning technologies for building robust and scalable dialogue systems with deeper understanding of the classic pipelines as well as detailed knowledge on the benchmark of the models of the prior work. We'll start with an overview of the dialogue research and allow the audience to dive deep into the state-of-the-art work about neural-based language understanding, dialogue management, and language generation towards end-to-end neural dialogue systems.  Back
 
Topics:
AI and DL Research, Speech and Language Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8542
Streaming:
Download:
Share:
 
Abstract:
Deep residual networks (ResNets) made a recent breakthrough in deep learning. The core idea of ResNets is to have shortcut connections between layers that allow the network to be much deeper while still being easy to optimize avoiding vanishing ...Read More
Abstract:

Deep residual networks (ResNets) made a recent breakthrough in deep learning. The core idea of ResNets is to have shortcut connections between layers that allow the network to be much deeper while still being easy to optimize avoiding vanishing gradients. These shortcut connections have interesting properties that make ResNets behave differently from other typical network architectures. In this talk we will use these properties to design a network based on a ResNet but with parameter sharing and adaptive computation time, we call it IamNN. The resulting network is much smaller than the original network and can adapt the computational cost to the complexity of the input image. During this talk we will provide an overview of ways to design compact networks, give an overview of ResNets properties and discuss how they can be used to design compact dense network with only 5M parameters for ImageNet classification.

  Back
 
Topics:
AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8456
Streaming:
Download:
Share:
 
Abstract:
This talk presents a novel framework named multimodal memory model for video captioning, which builds a visual and textual shared memory to model the long-term visual-textual dependency and further guide visual attention on described visual targets t ...Read More
Abstract:
This talk presents a novel framework named multimodal memory model for video captioning, which builds a visual and textual shared memory to model the long-term visual-textual dependency and further guide visual attention on described visual targets to solve visual-textual alignments. Video captioning which automatically translates video clips into natural language sentences is a very important task in computer vision. By virtue of recent deep learning technologies, video captioning has made great progress. However, learning an effective mapping from the visual sequence space to the language space is still a challenging problem due to the long-term multimodal dependency modelling and semantic misalignment. Inspired by the facts that memory modelling poses potential advantages to long-term sequential problems and working memory is the key factor of visual attention, the proposed model attaches an external memory to store and retrieve both visual and textual contents by interacting with video and sentence with multiple read and write operations.  Back
 
Topics:
AI and DL Research, Computer Vision, Video and Image Processing
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8311
Streaming:
Download:
Share:
 
Abstract:
Want to get started using TensorFlow together with GPUs? Then come to this session, where we will cover the TensorFlow APIs you should use to define and train your models, and the best practices for distributing the training workloads to multipl ...Read More
Abstract:

Want to get started using TensorFlow together with GPUs? Then come to this session, where we will cover the TensorFlow APIs you should use to define and train your models, and the best practices for distributing the training workloads to multiple GPUs. We will also look at the underlying reasons why are GPUs are so great to use for Machine Learning workloads?

  Back
 
Topics:
AI and DL Research, Deep Learning and AI, Developer Talk
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8946
Streaming:
Download:
Share:
 
Abstract:
Humans are remarkably proficient at controlling their limbs and tools from a wide range of viewpoints, diverse environments and in the presence of distractors. In robotics, this ability is referred to as visual servoing. Standard visual servoing appr ...Read More
Abstract:
Humans are remarkably proficient at controlling their limbs and tools from a wide range of viewpoints, diverse environments and in the presence of distractors. In robotics, this ability is referred to as visual servoing. Standard visual servoing approaches have limited generalization as they typically rely on manually designed features and calibrated camera. We exhibit generalizable visual servoing in the context of robotic manipulation and navigation tasks learned through visual feedback and by deep reinforcement learning (RL) without needing any calibrated setup. By highly randomizing our simulator, we train policies that generalize to novel environments and also to the challenging real world scenarios. Our domain randomization technique addresses the high sample complexity of deep RL, avoids the dangers of trial-and-error and also provides us with the liberty to learn recurrent vision-based policies for highly diverse tasks where capturing sufficient real robot data is impractical. An example of such scenario is learning view-invariant robotic policies which leads into learning physical embodiment and self-calibration purely through visual feedback.  Back
 
Topics:
AI and DL Research, IoT, Robotics & Drones, Robotics & Autonomous Machines
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8955
Streaming:
Share:
 
Abstract:
We saw the huge success of the deep learning paradigm and the superhuman capability in numerous benchmarks in image, video, audio, or text. However, it poses huge challenges as adopting the methods in industrial applications (mainly due to the lack o ...Read More
Abstract:
We saw the huge success of the deep learning paradigm and the superhuman capability in numerous benchmarks in image, video, audio, or text. However, it poses huge challenges as adopting the methods in industrial applications (mainly due to the lack of quality tracking data) as the neural networks consume enormous parameters and require relatively huge quality training data. We'll aim for investigating the "data augmentation" strategies increasing quality training data for robust inference across different learning problems mainly in image, video, 3D, and IoT data streams. We'll first quantify the importance of training data for deep neural networks then review numerous strategies, such as crawling from the web, utilizing generative models, 3D computer graphics, augmented reality, engagement in social media, gaming, etc. We'll compare the effectiveness among the diverse strategies. As generally taking the data from other domains, we also need to deal with the cross-domain learning problem. We'll provide detailed insights from our recent work published in top conferences (e.g., CVPR, ICCV, AAAI, etc.) and those cases in industrial applications.  Back
 
Topics:
AI and DL Research, Advanced AI Learning Techniques (incl. GANs and NTMs)
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8391
Streaming:
Download:
Share:
 
Abstract:
We'll review recent efforts to compress fully connected layers in machine learning via tensor networks, including the Tensor Train format, the Tensor Contraction Layer, the Tensor Regression Layer, and a Tensor Ring decomposition. These decompositio ...Read More
Abstract:
We'll review recent efforts to compress fully connected layers in machine learning via tensor networks, including the Tensor Train format, the Tensor Contraction Layer, the Tensor Regression Layer, and a Tensor Ring decomposition. These decompositions, in supplementing or replacing fully connected layers, are shown to dramatically reduce the number of parameters required by the network without resorting to sparsity and without loss in error. We've shown 55-80 percent compression of the entire network with less than one percent loss in accuracy. These Tensor layers can be used in end-to-end training, fine-tuning, and transfer-learning by initializing the decomposition with a pre-trained fully connected layer. Furthermore, because the forward and backward passes of the network rely on dense Tensor contractions, we show that these methods retain high computational intensity and can be efficiently evaluated on GPUs.  Back
 
Topics:
AI and DL Research, Algorithms and Numerical Techniques, HPC and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8807
Streaming:
Download:
Share:
 
Abstract:
The Department of Energy (DOE) entered into a partnership with the National Cancer Institute (NCI) of the National Institutes of Health (NIH) to accelerate cancer research. This "Cancer Moonshot" aims to tackle three main objectives: better ...Read More
Abstract:
The Department of Energy (DOE) entered into a partnership with the National Cancer Institute (NCI) of the National Institutes of Health (NIH) to accelerate cancer research. This "Cancer Moonshot" aims to tackle three main objectives: better understand the mechanisms of cancer, use large amounts of diverse medical data for predictive models, and enable precision medicine by providing guidance for treatment to individual patients. Leveraging the compute expertise of DOE in high performance computing (HPC) and new methods for deep learning in artificial intelligence, this HPC+AI approach aims to create a single scalable deep neural network code called CANDLE (CANcer Distributed Learning Environment) that will be used to address all three challenges. This talk aims to give an overview of the project and highlight how GPU accelerated systems in the DOE ecosystem, Summit and Sierra, have contributed to the project.  Back
 
Topics:
AI and DL Research, HPC and AI, Medical Imaging and Radiology
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81033
Streaming:
Share:
 
Abstract:
Voxelized representation of 3D objects is commonly used for training 3D-Convolutional Neural Networks for object detection and classification. However, high-resolution voxelization of CAD models are memory intensive and hence, it is not possible to l ...Read More
Abstract:
Voxelized representation of 3D objects is commonly used for training 3D-Convolutional Neural Networks for object detection and classification. However, high-resolution voxelization of CAD models are memory intensive and hence, it is not possible to load multiple models in the GPU for training. We have developed a GPU-accelerated voxelization technique that generates multi-level voxel grids of 3D objects. Instead of creating a single high-resolution voxel grid for the whole object, this technique generates selective region-based high-resolution voxel grids to represent detailed features in the object. We have also developed a multi-resolution 3D-Convolutional Neural Network that uses this hybrid voxelization for accurate object recognition and classification.  Back
 
Topics:
AI and DL Research, Industrial Inspection, Computer Vision
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8389
Streaming:
Download:
Share:
 
Abstract:
We'll present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks. Conditional GANs have enabled a variety of applications, but the results are often limited ...Read More
Abstract:
We'll present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks. Conditional GANs have enabled a variety of applications, but the results are often limited to low-res and still far from realistic. We'll show that we're capable of generating 2048x1024 visually appealing results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures. Furthermore, we extend our framework to interactive visual manipulation with two additional features. First, we incorporate object instance segmentation information, which enables object manipulations such as removing/adding objects and changing the object category. Second, we propose a method to generate diverse results given the same input, allowing users to edit the object appearance interactively. Human opinion studies demonstrate that our method significantly outperforms existing methods, advancing both the quality and the resolution of deep image synthesis and editing.  Back
 
Topics:
AI and DL Research, Graphics and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8918
Streaming:
Share:
 
Abstract:
In this session, you will learn about the latest IBM PowerAI solution, IBM Cloud GPU offerings and see a price-performance comparison, with supporting data, on the number of CPUs required to optimize GPU performance. We've also aggregated extensive ...Read More
Abstract:
In this session, you will learn about the latest IBM PowerAI solution, IBM Cloud GPU offerings and see a price-performance comparison, with supporting data, on the number of CPUs required to optimize GPU performance. We've also aggregated extensive test data to determine general best practices such as half-precision deep learning advantages on the Tesla V100 and the implications of neural-network model variable distribution and gradient aggregation techniques on your performance results. Join us to see why NVIDIA GPUs on IBM Cloud offer superior results.  Back
 
Topics:
AI and DL Research, Accelerated Analytics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81013
Streaming:
Download:
Share:
AI for Gaming
Presentation
Media
Abstract:
GPU accelerated creative development platforms are no longer just for games, they're revolutionizing areas from film to automotive. See how Unity is being used to enable unheard-of levels of productivity and create even deeper collaboration between ...Read More
Abstract:
GPU accelerated creative development platforms are no longer just for games, they're revolutionizing areas from film to automotive. See how Unity is being used to enable unheard-of levels of productivity and create even deeper collaboration between teams.  Back
 
Topics:
AI for Gaming, Graphics and AI, Real-Time Graphics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S81010
Streaming:
Download:
Share:
 
Abstract:
Advances in deep reinforcement learning have allowed autonomous agents to perform well on Atari games, often outperforming humans, using only raw pixels to make their decisions. However, most of these games take place in 2D environments that are full ...Read More
Abstract:
Advances in deep reinforcement learning have allowed autonomous agents to perform well on Atari games, often outperforming humans, using only raw pixels to make their decisions. However, most of these games take place in 2D environments that are fully observable to the agent. We present the first architecture to tackle 3D environments in first-person shooter games that involve partially observable states. Typically, deep reinforcement learning methods only utilize visual input for training. We present a method to augment these models to exploit game feature information, such as the presence of enemies or items, during the training phase. Our model is trained to simultaneously learn these features along with minimizing a Q-learning objective, which is shown to dramatically improve the training speed and performance of our agent. Our architecture is also modularized to allow different models to be independently trained for different phases of the game. We show that the proposed architecture substantially outperforms built-in AI agents of the game as well as average humans in deathmatch scenarios.  Back
 
Topics:
AI for Gaming, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8467
Streaming:
Download:
Share:
 
Abstract:
Over the last couple of years, neural nets have enabled significant breakthroughs in computer vision, voice generation and recognition, translation, and self-driving cars. Neural nets will also be a powerful enabler for future game development. We'l ...Read More
Abstract:
Over the last couple of years, neural nets have enabled significant breakthroughs in computer vision, voice generation and recognition, translation, and self-driving cars. Neural nets will also be a powerful enabler for future game development. We'll give an overview of the potential of neural nets in game development, as well as provide an in-depth look at how we can use neural nets combined with reinforcement learning for new types of game AI.  We will also show some new exciting results from applying deep reinforcement learning to AAA games.  Back
 
Topics:
AI for Gaming, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8715
Streaming:
Download:
Share:
 
Abstract:
Real-time games have an extremely small budget for computations of each frame. Learn the right way to approach real-time performance with inference workloads, taking advantage of the newest technologies available.
 
Topics:
AI for Gaming
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8742
Streaming:
Share:
 
Abstract:
We''ll examine tools and technologies that NVIDIA''s GameWorks team is building to leverage the power of deep learning for content creation, demonstrating recent research in ways that neural networks can be used to generate realistic looking human an ...Read More
Abstract:
We''ll examine tools and technologies that NVIDIA''s GameWorks team is building to leverage the power of deep learning for content creation, demonstrating recent research in ways that neural networks can be used to generate realistic looking human animation. We''ll talk about how to apply GPUs for high-performance runtime inferencing of these networks for use in games or real-time VFX scenarios.  Back
 
Topics:
AI for Gaming
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8743
Streaming:
Download:
Share:
 
Abstract:
We''ll present a machine learning solution that enables cost-efficient creation of large amounts of high-quality facial animation for digital doubles in games. Remedy Entertainment, NVIDIA, and the University of Southern California recently published ...Read More
Abstract:
We''ll present a machine learning solution that enables cost-efficient creation of large amounts of high-quality facial animation for digital doubles in games. Remedy Entertainment, NVIDIA, and the University of Southern California recently published "Production-Level Facial Performance Capture Using Deep Convolutional Neural Networks" as part of the Symposium on Computer Animation. We''ll cover topics including recording a facial animation dataset for an actor, setting up a deep learning project, preprocessing the data, training a deep convolutional neural network, and evaluating the results. We''ll also present a summary of the findings and discuss potential future work.  Back
 
Topics:
AI for Gaming, Graphics and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8734
Streaming:
Share:
 
Abstract:
We''ll present an overview of the StarCraft II machine learning environment, including some basic API examples using C++ and Python.
 
Topics:
AI for Gaming, Graphics and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8739
Streaming:
Download:
Share:
 
Abstract:
Unity ML-Agents is an open source AI toolkit that enables machine learning developers and researchers to train agents in realistic, complex scenarios with decreased technical barriers. ML-Agents offers a flexible way to use the Unity Editor and Engin ...Read More
Abstract:
Unity ML-Agents is an open source AI toolkit that enables machine learning developers and researchers to train agents in realistic, complex scenarios with decreased technical barriers. ML-Agents offers a flexible way to use the Unity Editor and Engine to develop and test new AI algorithms quickly and efficiently across games, robotics, and beyond. We''ll walk you through new learning methods that are bundled with the latest version of Unity ML-Agents. This includes, (1) Imitation Learning: Train agents to mimic human behavior. (2) Multi-agent Reinforcement Learning: Train multiple agents together to fulfill cooperative, competitive, and general tasks. We''ll showcase these new learning methods in some interesting training scenarios with real game examples.  Back
 
Topics:
AI for Gaming, Graphics and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8740
Streaming:
Download:
Share:
 
Abstract:
We have been developing Multi-Player Online Battle Arena (MOBA) Game AI to push AI research boundaries in pursuit of general AI. One of the key challenges in 5v5 MOBA AI development is how to process massive game replays and feed the data to model tr ...Read More
Abstract:
We have been developing Multi-Player Online Battle Arena (MOBA) Game AI to push AI research boundaries in pursuit of general AI. One of the key challenges in 5v5 MOBA AI development is how to process massive game replays and feed the data to model training in an efficient and reliable manner. To address this, we have built a large-scale game AI platform where millions of CPUs and thousands of GPUs are efficiently scheduled. Powered by our game AI platform and scheduling schemes, our MOBA AI is capable of learning upon billions of high-quality user replay samples per day using both deep learning and self-play.  Back
 
Topics:
AI for Gaming, Data Center and Cloud Infrastructure, AI and DL Research
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8922
Streaming:
Download:
Share:
 
Abstract:
The artistic manpower needed to create a video-game has been increasing exponentially over the years. Thanks to the computational power of NVIDIA GPUs, new AI accelerated workflows are poised to solve this problem, saving artists and studio ...Read More
Abstract:

The artistic manpower needed to create a video-game has been increasing exponentially over the years. Thanks to the computational power of NVIDIA GPUs, new AI accelerated workflows are poised to solve this problem, saving artists and studios time and money, and driving greater creativity. Artomatix is the leading pioneer in this space, its AI-based approach to content creation helps automate many of the mundane, tedious and repetitive tasks artists and designers face every day. This talk introduces the academic theory and history behind Creative AI and then delves into specific use cases and applications such as: Texture Synthesis, Material Enhancement, Hybridization and Style Transfer. Finally, this talk presents the next generation of tools for the creative industries, powered by AI, and gives case studies on how they've been solving some of the game industries largest problems over the past year. Join this session to gain an insight to the future of game creation.

  Back
 
Topics:
AI for Gaming, NVIDIA Inception Program, Graphics and AI
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8735
Streaming:
Share:
AI in Healthcare
Presentation
Media
Abstract:
The increasing availability of large medical imaging data resources with associated clinical data, combined with the advances in the field of machine learning, hold large promises for disease diagnosis, prognosis, therapy planning and therapy mo ...Read More
Abstract:

The increasing availability of large medical imaging data resources with associated clinical data, combined with the advances in the field of machine learning, hold large promises for disease diagnosis, prognosis, therapy planning and therapy monitoring. As a result, the number of researchers and companies active in this field has grown exponentially, resulting in a similar increase in the number of papers and algorithms. A number of issues need to be addressed to increase the clinical impact of the machine learning revolution in radiology. First, it is essential that machine learning algorithms can be seamlessly integrated in the clinical workflow. Second, the algorithm should be sufficiently robust and accurate, especially in view of data heterogeneity in clinical practice. Third, the additional clinical value of the algorithm needs to be evaluated. Fourth, it requires considerable resources to obtain regulatory approval for machine learning based algorithms. In this workshop, the ACR and MICCAI Society will bring together expertise from radiology, medical image computing and machine learning, to start a joint effort to address the issues above.

  Back
 
Topics:
AI in Healthcare, Medical Imaging and Radiology
Type:
Panel
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8897
Streaming:
Download:
Share:
 
Abstract:
Learn how to apply deep learning for detecting and segmenting suspicious breast masses from ultrasound images. Ultrasound images are challenging to work with due to the lack of standardization of image formation. Learn the appropriate data augme ...Read More
Abstract:

Learn how to apply deep learning for detecting and segmenting suspicious breast masses from ultrasound images. Ultrasound images are challenging to work with due to the lack of standardization of image formation. Learn the appropriate data augmentation techniques, which do not violate the physics of ultrasound imaging. Explore the possibilities of using raw ultrasound data to increase performance. Ultrasound images collected from two different commercial machines are used to train an algorithm to segment suspicious breast with a mean dice coefficient of 0.82. The algorithm is shown to perform at par with conventional seeded algorithm. However, a drastic reduction in computation time is observed enabling real-time segmentation and detection of breast masses.

  Back
 
Topics:
AI in Healthcare, Medical Imaging and Radiology
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8525
Streaming:
Download:
Share:
 
Abstract:
It is not always easy to accelerate a complex serial algorithm with CUDA parallelization. A case in point is that of aligning bisulfite-treated DNA (bsDNA) sequences to a reference genome. A simple CUDA adaptation of a CPU-based implementation c ...Read More
Abstract:

It is not always easy to accelerate a complex serial algorithm with CUDA parallelization. A case in point is that of aligning bisulfite-treated DNA (bsDNA) sequences to a reference genome. A simple CUDA adaptation of a CPU-based implementation can improve the speed of this particular kind of sequence alignment, but it's possible to achieve order-of-magnitude improvements in throughput by organizing the implementation so as to ensure that the most compute-intensive parts of the algorithm execute on GPU threads.

  Back
 
Topics:
AI in Healthcare, Bioinformatics & Genomics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8130
Streaming:
Download:
Share:
 
Abstract:
Fast, inexpensive and safe, ultrasound imaging is the modality of choice for the first level of medical diagnostics. The emerging solutions of portable and hand-held 2/3D scanners, advanced imaging algorithms, and deep learning promise further d ...Read More
Abstract:

Fast, inexpensive and safe, ultrasound imaging is the modality of choice for the first level of medical diagnostics. The emerging solutions of portable and hand-held 2/3D scanners, advanced imaging algorithms, and deep learning promise further democratization of this technology. During the session, we will present an overview of ultrasound imaging techniques in medical diagnostics, explore the future of ultrasound imaging enabled by GPU processing, as well as set out the path to the conception of a portable 3D scanner. We will also demonstrate our hardware developments in ultrasound platforms with GPU-based processing. Having started with one large research scanner, we have begun our migration towards more commercially-viable solutions with a small hand-held unit built on the mobile GPU NVidia Tegra X1.

  Back
 
Topics:
AI in Healthcare, Medical Imaging and Radiology
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8421
Streaming:
Download:
Share:
 
Abstract:
We'll disscuss how GPUs are playing a central role in making advances in Ion Torrent's targeted sequencing workflow and talk about the S5 DNA sequencer from Ion Torrent that is enabling democratization of sequencing market and accel ...Read More
Abstract:

We'll disscuss how GPUs are playing a central role in making advances in Ion Torrent's targeted sequencing workflow and talk about the S5 DNA sequencer from Ion Torrent that is enabling democratization of sequencing market and accelerating research in precision medicine at a breathtaking pace with the help of GPUs. We'll highlight our work in liquid biopsy and non-invasive prenatal testing and how the breadth in technology offerings in semiconductor chips gives us the scale of sequencing from small panels to exomes. We'll discuss our analysis pipeline and the latest and greatest in algorithm development and acceleration on GPUs as well as our experiences ranging from Fermi to Pascal GPU architectures. 

  Back
 
Topics:
AI in Healthcare, Bioinformatics & Genomics
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8419
Streaming:
Download:
Share:
 
Abstract:
How can we train medical deep learning models at a petabyte scale and how can these models impact clinical practice? We will discuss possible answers to these questions in the field of Computational Pathology. Pathology is in the midst of a revo ...Read More
Abstract:

How can we train medical deep learning models at a petabyte scale and how can these models impact clinical practice? We will discuss possible answers to these questions in the field of Computational Pathology. Pathology is in the midst of a revolution from a qualitative to a quantitative discipline. This transformation is fundamentally driven by machine learning in general and computer vision and deep learning in particular. With the help of PAIGE.AI we are building a clinical-grade AI at Memorial Sloan Kettering Cancer Center. The models are trained based on petabytes of image and clinical data on top of the largest DGX-1 V100 cluster in pathology. The goal is not only to automated cumbersome and repetitive tasks, but to impact diagnosis and treatment decisions in the clinic. This talk will focus on our recent advances in deep learning for tumor detection and segmentation, on how we train these high capacity models with annotations collected from pathologists, and how the resulting systems are implemented in the clinic.

  Back
 
Topics:
AI in Healthcare, Medical Imaging and Radiology
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8960
Streaming:
Download:
Share:
 
Abstract:
Machine Learning in Precision Medicine: Patient-Specific Treatment Enabled by Quantitative Medical Imaging, Artificial Intelligence, and GPU Efficiency The attendees will learn about the need for and use of machine learning in today's patien ...Read More
Abstract:

Machine Learning in Precision Medicine: Patient-Specific Treatment Enabled by Quantitative Medical Imaging, Artificial Intelligence, and GPU Efficiency The attendees will learn about the need for and use of machine learning in today's patient-centered healthcare. The talk will focus on general approaches requiring machine learning to obtain image-based quantitative features, reach patient diagnoses, predict disease outcomes, and identify proper precision-treatment strategies. While the presented methods are general in nature, examples from cardiovascular disease management will be used to demonstrate the need for and power of machine learning enabled by the performance advantages of GPU computation.

  Back
 
Topics:
AI in Healthcare, Medical Imaging and Radiology
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8892
Streaming:
Download:
Share:
 
Abstract:
AI in medical imaging has the potential to provide radiology with an array of new tools that will significantly improve patient care. To realize this potential, AI algorithm developers must engage with physician experts and navigate domains such ...Read More
Abstract:

AI in medical imaging has the potential to provide radiology with an array of new tools that will significantly improve patient care. To realize this potential, AI algorithm developers must engage with physician experts and navigate domains such as radiology workflow and regulatory compliance. This session will discuss a pathway for clinical implementation, and cover ACR's efforts in areas such as use case development, validation, workflow integration, and monitoring.

  Back
 
Topics:
AI in Healthcare, Medical Imaging and Radiology
Type:
Talk
Event:
GTC Silicon Valley
Year:
2018
Session ID:
S8994
Download:
Share: