This talk will explain how high quality image segmentation is critical in biomedical image interpretation for accurate diagnosis and/or assessment of a disease. The FNLCR IVG aims to integrate deep learning into image analysis workflows to produce quantitative, accurate, high throughput, and reproducible results to streamline image interpretation. We trained CNNs for mice tumor segmentation on MRI images for radiomics studies on patient derived xenograft (PDX) models. We trained CNNs and developed software infrastructures for feature quantification of whole slide histology images applied to collagen network analysis and stroma segmentation. Key features include the ability to annotate whole slides, incorporate multiplexed features, and providing an interactive interface for "human-in-the-loop" review and feedback.
We will describe the CoMet application for largescale epistatic Genome-Wide Association Studies (eGWAS) and pleiotropy studies. High performance is attained by transforming the underlying vector comparison methods into generalized distributed dense linear algebra operations. The 2-way and 3-way Proportional Similarity metric and the Custom Correlation Coefficient are implemented using adapted xGEMM kernels optimized for GPU architectures, achieving instruction rates similar to the unmodified kernels. By aggressive overlapping of communications, transfers and computations, and accessing the tensor cores on the Volta GPU, the full computation achieves up to 95 TF per GPU (76% of tensor cores theoretical peak 125 TF) on Summit. 234 x 10^15 element comparisons and 1.88 ExaOps have been reached on 4000 nodes of Summit; full system Summit projected values are 270 x 10^15 comparisons and over 2 ExaOps. Current performance is over 10,000X beyond comparable state of the art. CoMet is currently being used in projects ranging from bioenergy to clinical genomics, including for the genetics of chronic pain and opioid addiction.
The goal of this session is to describe the motivations behind building various tools for Artificial Intelligence Research in an academic Medical Center emphasizing the challenges and practical solutions for successful implementation. The discussion will cover key aspects of the software development cycle in the Healthcare setting including access to data, platforms and infrastructure, collaborative algorithm development, algorithm implementation and integration into clinical workflows.
In this session, we'll discuss the application of OpenSeq2Seq, an Nvidia Research project directed at speech and text processing, to telephone use cases in Healthcare. We will begin by describing the OpenSeq2Seq project and its goals. We will then cover the speech to text use cases in healthcare, related to member interactions with customer service representatives. Finally, we'll discuss our application of OpenSeq2Seq to the datasets, including normalization and modification to the OpenSeq2Seq code (e.g. in order to enable transfer learning) as well as our results
Numerous Fortune 500 customers experience latency and performance issues in their machine learning and data pipelines. Big data platforms and solutions tried to address these challenges with massive infrastructure scale out. But the cost to scale relative to the volume and velocity of current need is prohibitively expensive. NVIDIA is addressing these challenges with RAPIDS, an end-to-end GPU-accelerated data science software stack for enterprises to explore and integrate AI into their core data driven decision-making processes. Learn how to get started with GPU-accelerated data science and quickly identify opportunities to accelerate machine learning workflows in your organization.
Learn how RAPIDS and the open source ecosystem are advancing data science. In this session, we will explore RAPIDS, the NEW open source data science platform from NVIDIA. Deep dive into the RAPIDS platform and learn how to get started leveraging the open-source libraries for easier development and enhanced performance data science on GPUs. See the latest engineering work, including benchmarks and demos. Finally, see how customers are benefiting from early primitives and outperforming CPU equivalents.
The rise of GPU-accelerated data science and AI has come about through a combination of open source innovation and better tooling to support reproducible workflows. However, as the diverse array of deep learning libraries continue to mature, attention is moving to other parts of the AI pipeline, including simulation, ETL, and deployment. In this talk, I'll review open source projects that address these other areas, such as Numba, for implementing custom simulations and data transformations on the GPU, and PyGDF, for GPU accelerated dataframes. I'll discuss how the Anaconda Distribution and its conda packaging system helps data scientists create reproducible environments and deploy models. Finally, I'll talk about how Anaconda Enterprise allows data science teams to collaborate efficiently on GPU-accelerated projects with each other, and supports AI workflows from data exploration all the way to deployment.
BlazingDB, the distributed SQL engine on GPUs, will show how we contribute to the Apache GPU Data Frame (GDF) project, and begun to leverage inside BlazingDB. Through the integration of the GDF we have been able to dramatically accelerate our data engine, getting over 10x performance improvements. More importantly, we have built a robust framework to help users bring data from their data lake into GPU accelerated workloads without having to ETL on CPU memory, or separate CPU clusters. Keep everything in the GPU, BlazingDB handles the SQL ETL, and then pyGDF and DaskGDF can take these results to continue machine learning workloads. With the GDF customer workloads can keep the data in the GPU, reduce network and PCIE I/O, dramatically improve ETL heavy GPU workloads, and enable data scientists to run end-to-end data pipelines from the comfort of one GPU server/cluster.
The US opioid crisis is at an all time high. See how analysts and data scientists can use OmniSci (formerly MapD), powered by GPUs, to interactively explore millions of prescription records to find insights to help combat this crisis. Learn how millisecond response time enables rapid and iterative geospatial exploration of this data, to help drive faster action and informed decision-making. The government produces the most socially-impactful and politically-powerful data in the world. New technologies have overcome traditional challenges of making this data both available and consumable, to help improve citizen services. For government agencies leveraging GPUs, there are new opportunities to analyze and visually interact with massive datasets without frustrating lag times.
This talk will present the results of running the following Graph500 and DARPA Graph Challenge benchmarks and highlight the improvements over other platforms: BFS Graph500
• Single Source Shortest Paths Graph500
• PageRank Pipeline Graph Challenge
• Triangle Counting Graph Challenge
• K-Truss Graph Challenge The tremendous performance advantages of the DGX-2 platform for deep-learning has recently gained a lot of publicity. However, that is not the only analytic environment that can take advantage of the DGX-2 architecture. Having sixteen fully connected 32GB Volta GPUs presents an intriguing platform for Graph Analytics. The 512GB of combined GPU memory and full NVLink connection between the GPUs offers a number of advantages over a distributed MPI-based approach.
For all its promise in extracting value from data, machine learning today is still a blend of art and science, and data scientists are in high demand. However, most skilled practitioners still spend a lot of their time on parameter tuning and optimization, and not on creative problem-solving. The next evolution in machine intelligence for government applications will be automating the creation of models or creating AI that designs AI systems. This will allow for subject matter experts and data scientists to stage problems in new ways, and for AI to provide the best model. In this presentation, we will discuss how SparkCognition has developed a new automated model building (AMB) capability designed specifically for time-series and complex problems. We will cover how the high levels of processing power now available in the private and public sectors are key to this new technology, and how SparkCognition AMB solutions maximize the value of organizational data by replicating the mind of a data scientist in automating components of data cleaning, performing latent relationship extraction, and determining the optimal model to solve a given problem. We will discuss how certain steps can be automated within your organization, and the best ways to experiment with the automated creation of deep neural networks to solve simple and complex problems in both the supervised and unsupervised domains. Finally, we will also discuss how SparkCognition has used this technology to quickly build innovative models in real use cases and applications.
Factories contain complex interactions and change at speeds that present challenges for real-time data decision making. To remain competitive, companies must seek deeper understanding of production processes.
The Accelerated Production Insights solution combines Bayesian analytics with GPU acceleration for discovering optimal policies using big data. Unlike other techniques, this approach gives companies insight that moves past dashboard metrics and into actionable solutions.
Attendees will gain knowledge in leveraging explainable machine learning techniques to improve factory performance.
NASA's Heliophysics Division operates a fleet of spacecraft to monitor the Sun's activity and how its changes drive space weather. We show how science and mission capabilities can be enhanced by deep learning:
(a) mega-Kelvin thermometry of the Sun's corona by using a deep neural network (DNN) to solve a compressed sensing problem, and
(b) revival of a spectrograph by using convolutional neural networks (CNNs) to measure the Sun's extreme UV spectral irradiance. This work was done at NASA's Frontier Development Lab, a public-private partnership between NASA and industry partners (including the SETI Institute, NVIDIA, IBM, Lockheed Martin, Google, Intel & kx).
The presentation will compare and contrast the performance of several generations of Nvidia GPU for embedded processing across a wide range of applications. This analysis specifically focuses on the suitability for scientific computing, data analytics, and Deep Learning in power and space constrained environments. Presentation contrasts the performance gains by each successive generation of Jetson card, including the just release Jetson Xavier, and identifies practical limits to the processing capability. The added processing power of the GPU, data bandwidth, and raw throughput results for these advanced embedded systems will be explored.
Deep learning continues to show benefit in significant aspects of sensor systems including computer vision, speech recognition, and cybersecurity. In parallel, radio frequency (RF) systems have become increasingly complex and the number of connected devices will significantly increase as IoT and 5G systems become prevalent. Deep learning within RF systems is a new field of research that shows promise for dealing with a congested spectrum, brining reliability enhancements, and simplifying the ability to build effective signal processing systems. The utilization of deep learning algorithms within RF technology has shown superior results and the ability to classify signals well below the noise floor when compared to traditional signal processing methods. Working with strategic partners, we have designed a software configurable wide-band RF transceiver system capable of performing real-time signal processing and deep learning with an NVIDIA Jetson TX2. We discuss RF specific system performance, collection of RF training data, and the software used by the community to create custom applications. Additionally, we will present data demonstrating applications in the field of deep learning enabled RF systems.
Artificial intelligence is impacting almost every part of the industrial and agricultural supply chain. From robots that quickly adapt to build new products, to automated vehicles that address last-mile challenges for product delivery, to UAVs that can automatically detect failing infrastructure, the world is transitioning from processes that are largely manual to ones that are largely automated. We'll discuss how AI and deep learning are enabling these advances. We'll also analyze a sampling of early successes across different applications. And finally we'll describe some of the remaining challenges to wide-scale deployment, and the work NVIDIA is doing to address those challenges via its Isaac initiative.
Developers are training deep learning algorithms to understand an autonomous vehicle's surrounding environment and follow the rules of the road, but these algorithms still require testing and validation before the system is ready to drive on its own. Many companies have begun testing vehicles on public roads, gathering driving data and exposing the technology to real world experiences. However, this type of validation on its own can be cumbersome - the Rand Corporation estimates it would take hundreds of millions to hundreds of billions of miles driven to prove an autonomous vehicle can safely drive on its own, which translates to nearly a century of driving. Now with the power of simulation, the deep learning algorithms that act as the brain of the self-driving car can drive millions of miles in the fraction of the time it would take to drive that distance in the real world. Photorealistic simulation can create any range of driving scenarios and test them again and again to verify that the car's AI brain can safely navigate them. This session will demonstrate how virtual reality simulation is being utilized to test and validate self-driving hardware and software systems, accelerating the path to safe deployment of autonomous vehicles.
This presentation will provide an overview of Blue River Technology's use of GPUs in developing their See and Spray technology for Precision Agriculture. We will motivate the use of Deep Learning in detection and classification of crops and weeds in production environments, and highlight the ways in which NVIDIA GPUs have provided the tools and platform for training powerful models. NVIDIA GPUs have also helped us perform real-time inference on working machines in the field. This talk will show how these systems perform and provide videos of the machines in operation.
In this talk we will cover the use of synthetic data in training deep neural networks for computer vision tasks. We will explain why this is a critical research area with application ranging from robotics to autonomous vehicles, and we will discuss some important techniques for generating synthetic data using domain randomization and 3D graphics engines.
Rules-based approaches to cyber security detection do not scale and are burdened by a reliance on human engineering. In this session, we explore machine learning approaches to cyber security threats, specifically those related to failed login attempts (often a left-of-compromise indicator of an attack) and credential misuse (abnormal behavior). Rather than apply rules, we use the data processing and analytic capabilities of the GPU Open Analytics Initiative (GOAI) to accelerate model training, inference, and other steps necessary to provide actionable alerts to an analyst in near real-time.
Investigation teams have a love/hate relationship with event logs. The ever-increasing volumes and richness of data opens many possibilities, but also makes day-to-day operations a slog. GPU acceleration is changing basic assumptions around what is possible. From incident response and threat hunting to anti-money-laundering and anti-fraud, Graphistry has been working with F500 and federal teams on more scalable approaches to human-in-the-loop analytics. In particular, we have been bringing end-to-end GPU acceleration to visual graph analytics for visually connecting virtually any log data. Using examples from malware outbreaks to human trafficking, we'll demonstrate what can now be achieved, and dig into the supporting technologies like hypergraphs, Apache Arrow, GoAi, and visual playbooks.
Pacific Northwest National Laboratory (PNNL) is a federally funded research and development center that is part of the Department of Energy National Lab complex. As part of our internally funded research portfolio, PNNL is developing a range of cyber analytics and defense tools using GPU-enabled deep learning models. In this talk, I will cover PNNL's overall GPU-enabled deep learning and investment strategy as well as highlight two specific examples of cyber analytics and defense tools developed at the lab.
Learn how to effectively schedule and manage your system workload using Slurm; the free, open source and highly scalable cluster management and job scheduling system for Linux clusters. Slurm is in use today on roughly half of the largest systems in the world servicing a broad spectrum of applications.
Slurm developers have been working closely with NVIDIA to provide capabilities specifically focused on the needs of GPU management. This includes a multitude of new options to specify GPU requirements for a job in various ways (GPU count per job, node, socket and/or task), additional resource requirements for allocated GPUs (CPUs and/or memory per GPU), how spawned tasks should be bound to allocated GPUs, and control over GPU frequency and voltage.
An introduction to Slurm's design and capabilities will be presented with a focus on managing workloads for GPUs.
Growth in the geospatial sector has led to an explosive amount of data being collected to characterize our changing planet. The National Geospatial Intelligence Agency's (NGA) Artificial Intelligence, Automation and Augmentation (AAA) strategy is focused on helping our nation and its allies harness these new resources to more quickly respond to time sensitive missions like humanitarian assistance and disaster response. Learn how Radiant Solutions is combining machine learning on NVIDIA GPUs with crowdsourcing capabilities to help operationalize AAA. Both speakers share success stories on how AAA is currently being applied to augment a variety of NGA missions and is helping close the gap between data collection and decisions.
What is Deep Learning? In what fields is it useful, and how does it relate to artificial intelligence? During this session, we'll get an understanding of 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.
In this session, you will learn why NAS and legacy file systems cannot sufficiently support deep learning workloadsÃ¢â¬âthey leave GPUs idling and waiting for data. We will explain why implementing an NVMe-native next-generation file system, like WekaIO MatrixÃ¢âÂ¢, into the AI architecture delivers the low latency, high throughput capabilities necessary to keep a system like the DGX2 saturated with data. We will profile two autonomous vehicles use cases that found Matrix to be more performant and scalable than NAS or an AFA. We will also present real life TensorFlow benchmarks comparing WekaIO performance to a local SSD file system showing that Matrix is the only coherent shared storage that is even faster than the current caching solutions. It also allows customers to linearly scale performance by adding more GPU servers.
This technical session will explore the objectives for building the DGX-2, along with the inspired, innovative technology and architecture used to eliminate traditional bottlenecks and to enable multi-GPU training at unprecedented scale. This talk led by the DGX product team will present on the following topics:
-the innovation and architecture found in NVSwitch, which enables the AI network fabric for the DGX-2 platform
-the design challenges and hardware architecture employed to enable 16 V100's to operate as one
-feature by feature walkthrough highlighting the most important innovations that accelerate deep learning workflow and training performance
-use cases that were previously unaddressable on a GPU platform, now solved with DGX-2
In this talk, we'll explore two case studies in medical imaging, where TensorFlow was used to develop tools to assist doctors detect disease. First, I'll cover recent work in Pathology, and explain the workflow to detect cancerous cells in gigapixel images. Next, we'll explore foundational work in detecting Diabetic Retinopathy. I'll close with recent work in developing an augmented reality microscope, detecting skin cancer using a mobile device, and educational resources for you to learn more.
Machine learning is rapidly advancing the state-of-the-art in algorithm performance for wireless telecommunications systems. Building on our work presented at GTC Silicon Valley, recasting fundamental wireless signal processing problems as data-centric deep learning problems, we present further evidence that learned signal processing algorithms can empower the next generation of wireless systems with significant reductions in power consumption and improvements in density, throughput, and accuracy when compared to the brittle and manually designed systems of today. This talk will introduce the core enabling technologies and fundamental approaches, share our latest work and results in deep learning-based sensing and learned communications, and discuss applications such as 5G and IoT, commercial cyber-threat sensing, and defense RF sensing to illustrate the wide range of fields these technologies will impact over the next several years.
Learn how Lilt automatically adapts neural machine translation models in real time to match translator word choice and writing style using sparse gradient updates of residual tensors. This presentation will describe recent research results and practical production tips that apply to a wide variety of interactive machine learning systems that offer personalized results.
Artificial intelligence is touching a growing number of jobs and will be an enormous part of the future workforce. How do we make sure today's students are equipped for a future AI-powered economy? What steps can educators and parents take now to get children excited about creating, building and innovating with AI? Learn what two youth-focused organizations, Iridescent and Girls Computing League, are doing to prepare K-12 students, including those from underserved communities, for an AI-powered future.
Many examples of AI are reported daily that enhance traditional products and servicesÃ¢â¬â but the benefits have only just begun to scratch the surface. Entire industries will be transformed and massive benefits will be realized in the next wave of AI deployment. Learn about the next generation of AI and how it will add over 60M jobs and $13 trillion to the global economy if policy makers, businesses and the AI community adopt the right strategies, initiatives and platforms to harness this incredible new technology.
This presentation will walk through the research and development Harris has performed in creating an automated pipeline for synthetic label data generation for training deep learning remote sensing algorithms. The benefit of artificial intelligence, machine learning and specifically deep learning to various disaster response, humanitarian and economic assessment applications has become obvious but has also exposed some challenges and hurdles to adoption. In the realm of remote sensing, the availability of labeled training data has proven to be a costly barrier to entry. Remote sensing physics-based modeling and simulation provide a solution to this challenge by synthesizing radiometrically accurate labeled training data in mass quantities. Pipeline capabilities, performance and use cases will be discussed in this presentation.
The audience will learn examples and common practices for using Kubernetes to leverage NVIDIA GPU computing power when building DL models. We use a converged data platform to serve as data infrastructure, providing distributed file system and key-value storage and streams. Kubernetes is an orchestration layer that manages containers to scale out the training and deployment of DL models using heterogeneous GPU clusters. We also leverage the ability to publish and subscribe to streams on the platform to build next-gen applications with DL models, and monitor the model performance and shift of feature distributions.
We'll do a dive deep into best practices and real world examples of leveraging the power and flexibility of local GPU workstations, such as the DGX Station, to rapidly develop and prototype deep learning applications. This journey will take you from experimenting and iterating fast and often, to obtaining a trained model, to eventually deploying scale-out GPU inference servers in a datacenter. Tools available, that will be explained, are NGC (NVIDIA GPU Cloud), TensorRT, TensorRT Inference Server, and YAIS. NGC is a cloud registry of docker images, TensorRT is an inference optimizing compiler, TensorRT Inference Server is a containerized microservice that maximizes GPU utilization and runs multiple models from different frameworks concurrently on a node, and YAIS is a C++ library for developing compute intensive asynchronous microservices using gRPC.
This session will consist of a lecture, live demos, and detailed instructions about different inference compute options, pre- and post-processing considerations, inference serving options, monitoring considerations, and scalable workload orchestration using Kubernetes.
This talks brings together A.I. implementers who have deployed deep learning at scale using NVIDIA DGX Systems. We'll focus on specific technical challenges they faced, solution design considerations, and best practices learned from implementing deep learning platforms. 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.
Although normal is an ambiguous term, many operations, such operational platforms (e.g. ship), electricity use, computer operations, and network traffic, depend on an understanding of normal conditions. Recognizing abnormalities can raise a concern before it becomes critical and help identify unusual situations worthy of investigation even ones formally unrecognized. Fortunately, data exists across these environments from the many "sensors" that exist. Outlined methods enable machine learning to form a predictive fabric. The fabric is constructed using two deep learning methods; self-organizing maps (SOM) create the envelopes of normal states and recurrent neural networks (RNN) predict sequences. The two combine to form a high performance prediction model to determine normal/abnormal conditions. The methods "evolve" the predictive models to incorporate "normal" changes over time. Additionally, the methods reveal the sensitivity of the various data sources to determine the key contributors. Both methods depend on a powerful architecture based on GPUs.
Over the past three decades, the prevalence of dedicated advanced monitoring systems on complex systems has grown extensively. The data aggregated from these monitoring systems, as well as other data sources such as flight test, supply chain, maintenance, safety, operator information, design specs, and logistics, provide valuable insight into how the assets are being operated over their lifespans and enables a wide range of advanced analytics that supports fleet sustainment. Using Sikorsky helicopter fleets as a case study, this talk will discuss how large data sets collected from fleets of assets are used to enhance safety and enable intelligent decision making about the sustainment of the assets including usage monitoring, optimizing maintenance & supply chain, maximizing availability, focusing troubleshooting, and enabling proactive and timely support. This is all enabled by massively scalable data platforms that facilitate the collection, storage, algorithm development, and retrieval of data collected from these fleets of assets. The architecture of this platform and the software and hardware used will be discussed, including the software framework that was created for scaling these resources out to engineering subject matter experts. Specific deep learning applications will also be presented including how deep learning was used for multivariate time series anomaly detection as well as feature extraction and classification.
Facebook's strength in AI innovation comes from its ability to quickly bring cutting-edge research into large scale production using a multi-faceted toolset. Learn how ONNX and PyTorch 1.0 are helping to accelerate the path from research to production by making AI development more seamless and interoperable. We'll share the latest on PyTorch 1.0 and discuss Facebook's initiatives around ethical and responsible AI development.
In this talk we discuss a new type of variational autoencoder that we call Semantic Distortion Networks (SDNs). These networks incorporate semantic constraints in the form of distortion functions (as in rate-distortion theory) during training that result in learning a latent space data manifold which is structured to emphasize task relevant information and de-emphasize task irrelevant information. We illustrate the impact of different distortion functions on the organization of latent space using image data. We will then discuss the application of Semantic Distortion to a large scale graph knowledge graph embedding problem with applications to natural language processing and the use of high performance GPU acceleration (DGX1s) for learning graph embeddings optimized with respect to multiple semantic objectives.
Deep learning and NVIDIA GPUs are exciting technologies for taking artificial intelligence to the next level. Lockheed Martin Space (LM Space) Advanced Technology Center (ATC) has a long history of developing innovative next generation space solutions. We understand the importance and potential of NVIDIA GPUs and are discovering new benefits in our applications. This presentation will give a brief overview of LM Space ATC's Analytics and Data Exploitation organization and outline how we are incorporating deep learning and NVIDIA GPUs.
As AI science has boomed, visions of its potential have expanded to match. AI's power, combined with its typical opacity, can make it seem like magic. Engineering the science into tangible solutions for government, however, requires practitioners to understand and accommodate domain-specific constraints that often differ materially from those in academic or commercial applications. In this session, we will present a three-part recipe that leaders can use to assess a problem's AI readiness, effectively target investments, and preempt the challenges that many organizations face when implementing AI. Critically, leaders using this methodology can also identify how to attain AI readiness for problems that have not yet reached that stage.
Every disaster event provides new lessons about how best to effectively collaborate and respond to rapidly relieve human suffering and achieve societal resilience. This presentation will focus on the challenges of Humanitarian Assistance and Disaster Response and Lockheed Martin's commitment to the development and implementation of current and future enhanced Artificial Intelligence capabilities to provide real time actionable information to enhance planning, preparedness, and response efforts. Collaboration efforts with Government, Defense, Industry, Academia, and the Private Sector will be highlighted.
Attendees will learn from how to address the challenges of building AI systems based on the design principles and technologies that have proven successful in Penguin Computing AI deployments for customers in the Top 500. Lessons learned will focus on end-to-end aspects of designing and deploying large scale GPU clusters including datacenter and environmental challenges, network performance and optimization, data pipeline and storage challenges as well as workload orchestration and optimization. Attendees will hear about real life deployments for private organizations and government labs, including those using OCP technology.
Attend this half-day session on Monday, October 22 from 1:00-5:00pm to hear how AI is transforming operations now and how deep learning will influence the future. Session topics include: "Bootstrap AI Projects within the Government", "Building Blocks of AI, Improving ROI with Computer Vision and Natural Language Processing", "How to effectively recruit and train employees while building your AI capability" and "The Future of AI".
The “Black Box” approach to AI may well be good enough for simple applications, but what happens when life or death decisions are entrusted to neural networks. Without baked-in “explainability”, can we really trust the results the machines are giving us?
If these decisions are for the purposes of a tank identifying a target, then the decisions they support need to be understandable. We will show how the explainability ethos can be built into the design of AI architectures, replacing large black box models with smaller explainable components. Ensembles and hierarchies of these smaller AI components can perform dedicated tasks enabling a decision audit trail to be maintained, allowing decisions to be tracked and errors to be diagnosed. We will discuss the difference between training and inferencing and how this impacts possible deployment from the cloud, to the dedicated server to embedded devices such as smart phones. Low latency requirements can be facilitated by efficient components optimised with TensorRT and deployed on an embedded device, while at the other end of the scale we will demonstrate how we use large-scale batch deconvolution inferencing for explaining decisions made by CNNs using state-of-the-art enterprise GPU cards.
We will discuss hybrid deployment of AI components for real-world use cases from the automotive, insurance and voice industries, and demonstrate how clever architecture and explainable AI components can provide instantaneous feedback to the user, resulting in improved data collection quality for inferencing and training purposes.
In the DARPA RFMLS program, Expedition Technology, Inc. (EXP) is creating new or adapt existing, ML practices, structures, and algorithms specialized for the radiofrequency (RF) domain. Four key capabilities form the foundation for the eventual RFML System as well the overall structure of the RFMLS program: Feature Learning, Attention & Saliency, Autonomous RF Sensor Configuration, and Waveform Synthesis. EXP is applying these RF ML capabilities under RFMLS for RF Forensics (RF Feature Learning and RF Waveform Synthesis) and for RF Spectrum Awareness and Autonomous RF sensor configuration. This presentation will describe the initial progress in this ongoing program, describing the RF training dataset, deep learning architecture, and computationally demanding aspects of RF deep learning.
We will present on how Expedition Technology has developed a set of deep learning algorithms used for change detection, tracking, and track analytics, including anomaly detection and activity recognition from overhead imagery. Our trained 3-frame change detection algorithm utilizes a convolutional network whose outputs serve as seed points for an object tracker that robustly handles non-trivial motions (e.g. move-stop-move), object obscuration, and low SNR scenarios without requiring background subtraction. The tracker utilizes a Siamese network-based deep learning architecture and learns features of the object of interest to strengthen track consistency through obscuration and over long tracks.
The Turing architecture introduces many new, never before seen graphics shading technologies as well as features designed to accelerate the graphics pipeline. In this talk, we will focus on Variable Rate Shading, Multi-View Rendering, Mesh Shading, and Texture-Space Shading. We will cover details of how the features work, provide examples of algorithms that leverage these features, and explain details on how to get started using these features in your graphics applications today.
The OptiX AI denoiser introduced at Siggraph 2017 is now integrated into the most popular ray tracing applications including, Autodesk Arnold, Chaos Vray, Redshift, Isotropix and many more. A new feature of the upcoming OptiX SDK leverages AI to determine the quality of an image automatically and without a reference image. In this talk we will describe the evolution of these features, how they are trained and how they can be used to accelerate ray tracing.
We'll discuss parallel implementations to resampling techniques, commonly used in particle filtering, and their performance on NVIDIA GPUs, including the embedded TX2. A novel parallel approach to implementing systematic and stratified schemes is the highlight, but we'll also feature an optimized version of the Metropolis resampling technique. There are two main challenges that have been addressed: Traditional systematic and stratified techniques are serial by nature, but our approach breaks the algorithm up in a way to allow implementation on a GPU while producing identical results to the serial method. Secondly, while the Metropolis method is well suited for a GPU, its naive implementation does not utilize coalesced accesses to global memory.
The vision of the Exascale Computing Project, initiated in 2016 as a formal U.S. Department of Energy project executing through 2022, is to accelerate innovation with exascale simulation and data science solutions. After a brief overview of this, we will give illustrative examples on how the ECP teams are leveraging, exploiting, and advancing accelerated-node software technologies and applications on hardware such as the powerful GPUs provided by NVIDIA. We will summarize best practices and lessons learned from these accelerated-node experiences along with ECP's plans moving into the exascale era, which is on the now near-term horizon.
These solutions will enhance U.S. economic competitiveness, change our quality of life, and strengthen our national security. ECP's mission is to deliver exascale-ready applications and solutions that address currently intractable problems of strategic importance and national interest; create and deploy an expanded and vertically integrated software stack on DOE HPC exascale and pre-exascale systems, defining the enduring US exascale ecosystem; and leverage U.S. HPC vendor research activities and products into DOE HPC exascale systems. The project is a joint effort of two DOE programs: the Office of Science Advanced Scientific Computing Research Program and the National Nuclear Security Administration Advanced Simulation and Computing Program. ECP's RD&D activities, which encompass the development of applications, software technologies, and hardware technologies and architectures, is carried out by over 100 small teams of scientists and engineers from the DOE national laboratories, universities, and industries.
Most AI researchers and industry pioneers agree that the wide availability and low cost of highly-efficient and powerful GPUs and accelerated computing parallel programming tools (originally developed to benefit HPC applications) catalyzed the modern revolution in AI/Deep Learning. Now, AI methods and tools are starting to be applied to HPC applications to great effect. This talk will describe an emergent workflow that uses traditional HPC numeric simulations to generate the labeled data sets required to train machine learning algorithms, then employs the resulting AI models to predict the computed results, often with dramatic gains in efficiency, performance, and even accuracy. Some compelling success stories will be shared, and the implications of this new HPC + AI workflow on HPC applications and system architecture in a post-Moore's Law world considered.
For the first time in human history we're using Artificial Intelligence technology to automate tasks and decision-making, and in most cases we don't understand how the technology works. This lack of understanding creates distrust and can disenfranchise the users the technology is intended to benefit the most. This is compounded in highly-regulated spaces, such as the U.S. government. In this session, we'll cover the shortcomings of how Machine Learning and AI technologies are being applied in the USG today and how you can establish a trusted environment for successful human and machine collaboration.
This talk explores how DeepStream enables developers to create high-stream density applications with deep learning and accelerated multimedia image processing, building IVA solutions at scale. Leverage a heterogeneous concurrent neural network architecture to bring in different deep learning techniques for more intelligent insights. The framework makes it easy to create flexible and intuitive graph-based applications, resulting in highly optimized pipelines for maximum throughput.
Learn how we overcame the odds of certifying computer vision and AI systems in an industry as risk adverse as the air traffic control sector. We use off-the-shelf cameras deployed in an airport environment to provide an out the window view of the airfield, create an enriched augmented reality view for better situational awareness, contingency and redundancy. In this talk, we take you through the steps from developing an AI using Nvidia frameworks, to deploying a camera system at an airport for air traffic control use as an imaging system as well as a tracking system using AI technology such as artificial neural networks. All the way through user acceptance tests and certification. This talk is intended as a lessons learned for your next project in smart cities or aerospace. The main focus of this talk lays on the tools used to develop AI and the tools used to understand and visualize neural networks.
Cities are always looking for new ways to maintain high standards of living, better connect with citizens and find ways to save moneyÃ¢â¬âall while serving growing populations. As city population densities increase and cities strive to increase walkability and mobility for their citizens, they have a big focus on a holistic approach to traffic safety. As part of their efforts to become smarter, more and more cities are turning to the Internet of Things (IoT) and Machine-to-Machine (M2M) technologies to improve municipal services, create additional sources of revenue, and enable city management in new and creative ways.
In this session, we'll explore some of the common challenges with scaling-out deep learning training and inference deployment on data centers and public cloud using Kubernetes on NVIDIA GPUs. Through examples, we'll review a typical workflow for AI deployments on Kubernetes. We'll discuss advanced deployment options such as deploying to heterogenous GPU clusters, specifying GPU memory requirements, and analyzing and monitoring GPU utilizations using NVIDIA DCGM, Prometheus and Grafana.
This session will cover the basics of deploying an AI application and includes a summary of the latest updates to the NVIDIA Deep Learning SDK. For inference deployment we will introduce the TensorRT Hyperscale Inference Server and how it fits into cloud or on prem AI software application ecosystem.
The Turing Tensor Cores deliver great performance for networks that can be executed in FP16 or INT8. Most models are developed in FP32. This talk will show how models can be converted to use TensorCores and the performance and accuracy implications of that conversion. We'll show results achieved with Tensor Cores in Volta, Xavier and Turing. We'll discuss how this works with networks created in TensorFlow or in an ONNX compatible framework like PyTorch. The higher throughput delivered allows organizations to deliver services at lower cost.
AI's potential cuts across all industries, from agriculture to healthcare to oil and gas and more. But it also can help government agencies be more efficient, better at identifying waste and fraud, and more responsive and convenient for all Americans. This panel will discuss the various AI applications that can make government smarter, and how we get there.
Artificial intelligence has opened a new class of efficient technologies to help the American farmer from AI-assisted thinning, weeding and spraying for row crops to automated soft fruit picking. This panel will discuss the latest AI innovations for the farm and the policies that will continue to advance U.S. agriculture through the 21st Century.
The U.S. is the world leader in developing AI technologies, but other countries are catching up. What must the U.S. do to sustain and strengthen its global leadership in AI research and development? What are the challenges and what more can and should be done by industry and the government to advance AI?
As a new computing paradigm, Virtual Reality (VR) is changing workflows and redefining how we interact with computers. Deep Learning (DL) is revolutionizing business processes, defining how autonomous machines interact with us and with the world, and demanding application developers learn new ways of working in every field touching compute. In this panel we explore the intersection of these two revolutions with VR industry innovators who are leveraging deep learning using NVIDIA GPU compute systems to bring depth to the VR experience. This discussion will focus on the use of Artificial Intelligence (AI) in both building rich VR environments and enhancing the user's interaction with the VR environment. Panelists will share their vision on how AI will shape the near future of VR and give the audience a view of potential challenges to that future.
In this session, we will explore:
o Pain points in creating VR experiences, which are driving adoption of AI in the VR space
o Challenges encountered in using DL to bring rich content to life in a VR environment
o Challenges of implementing DL-enhanced VR environment interaction within the latency-critical VR space
o How DL/AI will continue to fundamentally change the VR space
Holodeck is NVIDIA's advanced technology platform for Virtual Reality. We will begin this session by reviewing the motivation and goals of Holodeck. We will then highlight the new features from our most recent Holodeck release. The latest version adds support for architectural design review, including model review at scale, teleporting to different floors, and specifying points of interest. Two new technologies are featured in this release: VR Audio and the new Holodeck Web Browser. The Audio SDK from VRWorks provides real-time, physically-based acoustic simulation, accurately capturing how a scene's geometry and materials affect sounds. And VRWorks Audio is now hardware accelerated on NVIDIA's new RTX GPU's! The new Holodeck Web Browser opens up a world of new functionality, as VR experiences can now include all kinds of 2D, responsive content, including: videos, product manuals, and even Google Hangout sessions (with participants who are not in VR). Come hear this talk to learn all about the technology, and also visit the VR Village to experience the new Holodeck and VRWorks features firsthand!
Winslow Porter, Co-founder of New Reality Company, an award-winning and critically acclaimed VR studio with offices in NYC and LA, has been using Nvidia's Quadro Series GPUs to help bring their immersive, cinematic VR stories to life. He will detail how they use Houdini, Maya and Adobe tools in conjunction with the Unreal Engine with sensory elements such as scents, heat, wind and vibrations to create transformative, first person experiences that change a viewer's perspective on how they engage with media, and see the world. NVIDIA's involvement has not only been instrumental to the production of cutting-edge pieces Giant and Tree (both premiered at Sundance and the latter most recently shown at the World Economic Forum in Tianjin, China). Winslow will also detail their newest project, Breathe. In this piece, we'll be using a custom breath sensor to help place the viewer into the shoes of a 6-year-old war survivor, following her throughout key moments of her life on a journey to empowerment until her final breath. In order to maintain the highest quality look and feel inside the headset at 90 frames per second, NVIDIA's tools have been an essential partner, helping us create unforgettable experiences that in turn generate real world empathy and impact.
This three part presentation will explore how Radiant Solutions is making it possible to see and understand our changing world by applying computer vision to satellite imagery, enabling interactive terrain analytics, and powering immersive analytics in virtual reality.
Kevin McGee will talk first about machine learning/computer vision, then Ryan Smith will talk about terrain analytics, and Nick Deliman will close with VR.
Discussion and demonstration of the potential with running HPC, and VDI workloads on common clusters for a modern a datacenter Dr. Jekyll and Mr. Hyde scenario. Explore the coexistence of CUDA based HPC or Deep Learning job engines in conjunction with both Linux and Windows machines used for virtual desktop infrastructure. The demonstration will focus on a very minimal VMware vSphere cluster deployment using VSAN storage or RedHat RHVM cluster deployment to host both the Linux HPC multi node cluster for CUDA workloads and a VMware Horizon view or Citrix XenDesktop deployment for Linux and Windows Virtual Desktops performing DirectX, OpenGL, OpenCL, and CUDA based visualization workloads as used by engineering and analysis power users.
This session presents some of the challenges facing the military simulation and training community as we extend the operational life of existing simulators thru technology insertion and modernization. Solutions to these challenges involve embracing the application of virtualization to existing system architectures, and extending into cloud based services as part of future architectures. We specifically look at unique requirements for real-time 3D image generation and explore approaches to using NVIDA virtual GPU technologies to provide the required performance while capitalizing on the reduced footprint and enhanced sustainability afforded by virtualize GPUs.