The 2018 GTC opening keynote is delivered by the NVIDIA Founder and CEO, Jensen Huang, speaking on the future of computing.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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?
We''ll present an overview of the StarCraft II machine learning environment, including some basic API examples using C++ and Python.
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.
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.
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.
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.
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.
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.
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.
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.
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.
In this talk I will describe the research and development work on medical imaging, done at PingAn Technology and Google Cloud, covering five different tasks. I'll present the technical details of the deep learning approaches we have developed, and share with the audiences the research direction and scope in the medical fields at PingAn technology and PingAn USA Lab.
Deep learning models give state-of-the-art results on diverse problems, but their lack of interpretability is a major problem. Consider a model trained to predict which DNA mutations cause disease: if the model performs well, it has likely identified patterns that biologists would like to understand. However, this is difficult if the model is a black box. We present algorithms that provide detailed explanations for individual predictions made by a deep learning model and discover recurring patterns across the entire dataset. Our algorithms address significant limitations of existing interpretability methods. We show examples from genomics where the use of deep learning in conjunction with our interpretability algorithms leads to novel biological insights.
Learn how to apply recent advances in GPU and open data to unravel the mysteries of biology and etiology of disease. Our team has built data driven simulated neurons using CUDA and open data, and are using this platform to identify new therapeutics for Parkinson's disease with funding from the Michael J. Fox Foundation. In this session I'll discuss the open data which enables our approach, and how we are using Nvidia Tesla cards on Microsoft Azure to dynamically scale to more than 100,000 GPU cores while managing technology costs.
Radiological diagnosis and interpretation should not take place in a vacuum -- but today, it does. One of the greatest challenges the radiologist faces when interpreting studies is understanding the individual patient in the context of the millions of patients who have come previously. Without access to historical data, radiologists must make clinical decisions based only on their memory of recent cases and literature. Arterys is working to empower the radiologist with an intelligent lung nodule reference library that automatically retrieves historical cases that are relevant to the current case. The intelligent lung nodule reference library is built on top of our state-of-the-art deep learning-based lung nodule detection, segmentation and characterization system.
As deep learning techniques have been applied to the field of healthcare, more and more AI-based medical systems continue to come forth, which are accompanied by new heterogeneity, complexity and security risks. In the real-world we've seen this sort of situation lead to demand constraints, hindering AI applications development in China's hospitals. First, we'll share our experience in building a unified GPU accelerated AI engine system to feed component-based functionality into the existing workflow of clinical routine and medical imaging. Then, we'll demonstrate in a pipeline of integrating the different types of AI applications (detecting lung cancer, predicting childhood respiratory disease and estimating bone age) as microservice to medical station, CDSS, PACS and HIS system to support medical decision-making of local clinicians. On this basis, we'll describe the purpose of establishing an open and unified, standardized, legal cooperation framework to help AI participants to enter the market in China to build collaborative ecology.
This talk will overview the fields of Personalised Computational Medicine and In Silico Clinical Trials, which are revolutionizing Medicine and Medical Product Development. This talk will introduce these concepts, provide examples of how they can transform healthcare, and emphasize why artificial intelligence and machine learning are relevant to them. We will also explain the limitations of these approaches and why it is paramout to engage in both phenomenological (data-driven) and mechanistic (principle-driven) modelling. Both areas are in desperate need for better infrastructures -sofrware and hardaware- giving access to computational and storage resources. The talk will be thought-provoking and eye-opening as to opportunities in this space for researchers and industries alike.
The transformation towards value-based healthcare needs inventive ways to lower cost and increase patient health outcomes. Artificial intelligence is vital for realizing value-based care. Turning medical images into biomarkers helps to increase effectiveness of care.
We will introduce deep learning applications in clinical neuroimaging (using MRI, CT, PET, etc.) and recent breakthrough results from Stanford and Subtle Medical. Perspectives and feedbacks of applying AI technologies in neuroimaging are shared, from expert radiologists and deep learning experts. How Deep Learning/AI is changing clinical neuroimaging practice * How will deep learning be applied in radiology workflow right now and in the future * Practical concerns and perspectives from radiologists How Deep Learning assists smarter neuroimaging decision making * Multi-scale 3D network enables lesion outcome prediction for stroke * More accurate lesion segmentation in neuroimaging How Deep Learning enables safer and cheaper neuroimaging screening * Deep Learning and GAN enables >95% reduction in radiation for functional medical imaging * Deep Learning enables 90% reduction in chemical (Gadolinium) contrast agent usage in contrast enhanced MRI How Deep Learning accelerate neuroimaging * Further acceleration and improved MRI reconstruction using deep learning * Deep Generative Adversarial Network for Compressed Sensing
iFLYTEK Health's mission is to use the most advanced artificial intelligence technologies to revolutionize healthcare industry to help doctors provide quality care to more patients with higher efficiency. Developed upon iFLYTEK's world class hardware/software technologies in voice recognition and voice synthesization, iFLYTEK's products can help reduce doctors' burden in writing medical records and free their time to focus more on caring patients. These technologies can also reduce errors and improve completeness and accuracy of medical records, therefore support advanced intelligence applications based on complete patient data. Automated image analysis tools can help doctors find abnormalities in images with confidence, especially for the inexperienced doctors from lower tier hospitals. Clinical Decision Support (CDS) system is based on authoritative medical literature, large amount of expert knowledge, and real cases to improve primary doctors' ability of accurate diagnosis using complete and accurate patient information.
Discuss the difficulties in digital mammography, and the computational challenges we encountered while adapting deep learning algorithms, including GAN, to digital mammography. Learn how we address those computational issues, and get the information of our benchmarking results using both consumer and enterprise grade GPUs.
There is large promise in machine learning methods for the automated analysis of medical imaging data for supporting disease detection, diagnosis and prognosis. These examples include the extraction of quantitative imaging biomarkers that are related to presence and stage of disease, radiomics approaches for tumor classification and therapy selection, and deep learning methods for directly linking imaging data to clinically relevant outcomes. However, the translation of such approaches requires methods for objective validation in clinically realistic settings or clinical practice. In this talk, I will discuss the role of next generation challenges for this domain.
Learn about the key types of clinical use cases for AI methods in medical imaging beyond simple image classification that will ultimately improve medical practice, as well as the critical challenges and progress in applying AI to these applications. We''ll first describe the types of medical imaging and the key clinical applications for deep learning for improving image interpretation. Next, we''ll describe recent developments of word-embedding methods to leverage narrative radiology reports associated with images to generate automatically rich labels for training deep learning models and a recent AI project that pushes beyond image classification and tackles the challenging problem of clinical prediction. We''ll also describe emerging methods to leverage multi-institutional data for creating AI models that do not require data sharing and recent innovative approaches of providing explanation about AI model predictions to improve clinician acceptance.
Dive in to recent work in medical imaging, where TensorFlow is used to spot cancerous cells in gigapixel images, and helps physicians to diagnose disease. During this talk, we''ll introduce concepts in Deep Learning, and show concrete code examples you can use to train your own models. In addition to the technology, we''ll cover problem solving process of thoughtfully applying it to solve a meaningful problem. We''ll close with our favorite educational resources you can use to learn more about TensorFlow.
Protecting crew health is a critical concern for NASA in preparation of long duration, deep-space missions like Mars. Spaceflight is known to affect immune cells. Splenic B-cells decrease during spaceflight and in ground-based physiological models. The key technical innovation presented by our work is end-to-end computation on the GPU with the GPU Data Frame (GDF), running on the DGXStation, to accelerate the integration of immunoglobulin gene-segments, junctional regions, and modifications that contribute to cellular specificity and diversity. Study results are applicable to understanding processes that induce immunosuppressionlike cancer therapy, AIDS, and stressful environments here on earth.
Learn how researchers at Stanford University are leveraging the power of GPUs to improve medical ultrasound imaging. Ultrasound imaging is a powerful diagnostic tool that can provide clinicians with feedback in real time. Until recently, ultrasound beamforming and image reconstruction has been performed using dedicated hardware in order to achieve the high frame rates necessary for real-time diagnostic imaging. Though many sophisticated techniques have been proposed to further enhance the diagnostic utility of ultrasound images, computational and hardware constraints have made translation to the clinic difficult. We have developed a GPU-accelerated software beamforming toolbox that enables researchers to implement custom real-time beamforming on any computer with a CUDA-capable GPU, including commercial ultrasound scanners. In this session, we will: 1) briefly introduce the basics of ultrasound beamforming, 2) present our software beamforming toolbox, and 3) show videos demonstrating its capabilities from a clinical study of echocardiography, as well as an implementation of a novel speckle removing beamformer that utilizes deep fully convolutional neural networks.
Learn CAIDE Systems'' unique diagnosis system with highly accurate prediction and delineation of brain stroke lesion. We''ll present how we increase sensitivity in medical diagnosis system and how we develop a state-of-the-art generative deep learning model for acquiring segmented stroke lesion CT images, and demonstrate our market-ready product: a diagnostic tool as well as a medical deep learning platform. We trained our diagnostic system using CT image data from thousands of patients with brain stroke and tested to see commercial feasibility of use for hospitals and mobile ambulances.
In medical imaging, acquisition procedures and imaging signals vary across different modalities and, thus, researchers often treat them independently, introducing different models for each imaging modality. To mitigate the number of modality-specific designs, we introduced a simple yet powerful pipeline for medical image segmentation that combines fully convolutional networks (FCNs) with fully convolutional residual networks (FC-ResNets). FCNs are used to obtain normalized images, which are then iteratively refined by means of a FC-ResNet to generate a segmentation prediction. We''ll show results that highlight the potential of the proposed pipeline, by matching state-of-the-art performance on a variety of medical imaging modalities, including electron microscopy, computed tomography, and magnetic resonance imaging.
The NVIDIA Genomics Group has developed a deep learning platform to transform noisy, low-quality DNA sequencing data into clean, high-quality data. Hundreds of DNA sequencing protocols are used to profile phenomena such as protein-DNA binding and DNA accessibility. For example, the ATAC-seq protocol identifies open genomic sites by sequencing open DNA fragments; genome-wide fragment counts provide a profile of DNA accessibility. Recent advances enable profiling from smaller patient samples than previously possible. To reduce sequencing cost, we developed a convolutional neural network that denoises data from a small number of DNA fragments, making the data suitable for various downstream tasks. Our platform aims to accelerate adoption of DNA sequencers by minimizing data requirements.
Nanopore sequencing is a breakthrough technology that marries cutting edge semiconductor processes together with biochemistry, achieving fast, scalable, single molecule DNA sequencing. The challenge is real-time processing of gigabytes of data per second in a compact benchtop instrument. GPUDirect, together with the cuDNN library, enables Roche to maximize the effectiveness of Tesla V100 GPUs in their next generation sequencing instrument. Attendees will learn how these pieces come together to build a streaming AI inference engine to solve a signal processing workflow. Analysis and performance comparisons of the new TensorCore units, available on Volta hardware, will be included.cal cuDNN API
Learn how to use (multi) GPU and CUDA to speed up the process of stitching very large images (up to TeraBytes in size). Image stitching is the process of combining multiple photographic images with overlapping fields of view to produce a segmented panorama or high-resolution image. Image stitching is widely used in many important fields, like high resolution photo mosaics in digital maps and satellite photos or medical images. Motivated by the need to combine images produced in the study of the brain, we developed and released for free the TeraStitcher tool that we recently enhanced with a CUDA plugin that allows an astonishing speedup of the most computing intensive part of the procedure. The code can be easily adapted to compute different kinds of convolution. We describe how we leverage shuffle operations to guarantee an optimal load balancing among the threads and CUDA streams to hide the overhead of moving back and forth images from the CPU to the GPU when their size exceeds the amount of available memory. The speedup we obtain is such that jobs that took several hours are now completed in a few minutes.
This talk will present the challenges and opportunities in developing a deep learning program for use in medical imaging. It will present a hands on approach to the challenges that need to be overcome and the need for a multidisciplinary approach to help define the problems and potential solutions. The role of highly curated data for training the algorithms and the challenges in creating such datasets is addressed. The annotation of data becomes a key point in training and testing the algorithms. The role of experts in computer vision, and radiology will be addressed and how this project can prove to be a roadmap for others planning collaborative efforts will be addressed Finally I will discuss the early results of the Felix project whose goal is nothing short of the early detection of pancreatic cancer to help improve detection and ultimately improve patient outcomes.
Motion tracking with motion compensation is an important component of modern advanced diagnostic ultrasonic medical imaging with microbubble contrast agents. Search-based on sum of absolute differences a well-known technique for motion estimation is very amenable to efficient implementations, which exploit the fine grained parallelism inherent in GPUs. We''ll demonstrate a real-world application for motion estimation and compensation in the generation of real-time maximum intensity projections over time to create vascular roadmaps in medical images of organs, such as the liver with ultrasound contrast agents. We''ll provide CUDA kernel code examples which make this application possible as well as performance measurements demonstrating the value of instruction-level parallelism and careful control of memory access patterns for kernel performance improvement. We hope to provide insight to CUDA developers interested in motion estimation and compensation as well as general insight into kernel performance optimization relevant for any CUDA developer.
Clinical laboratories play a crucial role in healthcare ecosystem - the laboratories are pivotal and act as a screening sub-system by providing early inference in disease and abnormality diagnosis. An estimated 70% of clinical decisions regarding prevention, diagnosis and treatment involve lab tests. Surprisingly, 60% of the inferencing done at a clinical laboratory can be performed by one "wonder-tool" - microscope. Microscopy has helped pathologists assess and analyse the patients for over several centuries. The key hurdles in the microscopic examination are the amount of time that the pathologists have to spend in manual analysis and the need for the pathologists to be co-located with the specimen. In this talk, we introduce SigTuple's AI powered smart microscope that can automatically learn, analyse and summarize the inferences of several hundred abnormalities across different biological specimen (blood, urine and semen). It also utilizes the power of GPU computing on cloud to provide higher order analysis of the samples and acts as a tele-pathology enabler by providing pathologists the power to view or review any analysis or report from any part of the world.
For more than a decade, GE has partnered with Nvidia in Healthcare to power our most advanced modality equipment, from CT to Ultrasound. Part 1 of this session will offer an introduction to the deep learning efforts at GEHC, the platform we're building on top of NGC to accelerate new algorithm development, and then a deep dive into a case study of the evolution of our cardiovascular ultrasound scanner and the underlying extensible software stack. It will contain 3 main parts as follows: (a) Cardiovascular ultrasound imaging from a user perspective. Which problems we need to solve for our customers. Impact of Cardiovascular disease in a global perspective (b) An introduction to the Vivid E95 and the cSound platform , GPU based real time image reconstruction & visualization. How GPU performance can be translated to customer value and outcomes and how this has evolved the platform during the last 2 ½ years. (c) Role of deep learning in cardiovascular ultrasound imaging, how we are integrating deep learning inference into our imaging system and preliminary results from automatic cardiac view detection.
Hear about how GPU technology is disrupting the way your eye doctor works and how ophthalmic research is performed today. The rise of Electronic Medical Records in medicine has created mountains of Big Data particularly in ophthalmology where many discrete quantitative clinical elements like visual acuity can be tied to rich imaging datasets. In this session, we will explore the transformative nature that GPU acceleration has played in accelerating clinical research and show real-life examples of deep learning applications to ophthalmology in creating new steps forward in automated diagnoses, image segmentation, and computer aided diagnoses.
We'll show how recent advances in 3D fully convolutional networks (FCN) have made it feasible to produce dense voxel-wise predictions of volumetric images. FCNs can be trained to automatically segment 3D medical images, such as computed tomography (CT) scans based on manually annotated anatomies like organs and vessels. The presented methods achieve competitive segmentation results while avoiding the need for handcrafting features or training class-specific models, in a clinical setting. We'll explain a two-stage, coarse-to-fine approach that will first use a 3D FCN based on the 3D U-Net architecture to roughly define a candidate region. This candidate region will then serve as input to a second 3D FCN to do a fine prediction. This cascaded approach reduces the number of voxels the second FCN has to classify to around 10 percent of the original 3D medical image, and therefore allows it to focus on more detailed segmentation of the organs and vessels. Our experiments will illustrate the promise and robustness of current 3D FCN based semantic segmentation of medical images, achieving state-of-the-art results on many datasets. Code and trained models will be made available.
Explore how parallelized programming and DL can radically impact medical ultrasound imaging. In this session, we will describe how the processing of ultrasound signals can be implemented not only providing real-time capabilities, but also a flexible environment for research and innovative new products. In this view, we will i) demonstrate 2D and 3D real-time imaging using open hardware platforms, and ii) provide an overview, how both radical parallelization and DL can be integrated within processing pipelines, providing new applications and improved image quality at unprecedented speed.
In this session, attendees will learn how to develop an AI Learning Platform for healthcare, develop initial(imaging) AI applications in specific care areas, and embed AI into devices creating "intelligent imaging systems".
Learn about the importance of clinical domain expertise in AI algorithm/model development and incorporation into clinical workflow, specifically in medical imaging, from a radiologist. With growing media attention, there is much fear, hype, and hope when it comes to using DL in radiology. We will present through examples why it is essential to incorporate clinical domain expertise when developing DL models. We will demonstrate various ways AI can augment the radiologists both in image interpretation as well as beyond within the overall workflow. In the second portion of this talk, we will present the gap between developing a great AI model in isolation and having it become part of daily medical practice. From integration and hospital connectivity to algorithm serving at scale to meet growing demand, we will show how an AI Marketplace can create the ecosystem that allows AI to flourish.
The Role of Data in Achieving Precision and Value in Healthcare The goal of healthcare is to provide the most effective treatment to every patient in the most efficient way. Data plays a key role in every aspect of this process from decision support systems that provide a clinician with the right information at the right time, to scheduling algorithms that predict patient flow and schedule accordingly, to analytics to coach and support patients in achieving or maintaining a healthy lifestyle. Achieving the vision of a data-informed healthcare system will require fundamental advances in many areas including causal inference, inference on complex, high-dimensional and heterogeneous data, missing data, process modeling, bias reduction, statistical validation, and model adaptation, to name a few. In this talk, I will illustrate some of these challenges through concrete examples within the Malone Center.
Diabetic retinopathy, also known as diabetic eye disease, is a major complication of diabetes, which damage occurs to the retina due to diabetes mellitus and is a leading cause of blindness. AirDoc's product Dirctor, Emma Xu and Professor You Li of Shanghai Changzheng Hospital, will share how AirDoc, the leading Intelligent Medical startup in China, leverages Nvidia's GPU and Deep Learning to improve the DR diagnose with Automatic left/right eye recognition, Automatic detection of the location and numbers, Automatic DR staging, Fast recognition speed, Patient Information Management for real-time screening statistics and usage management.
Apollo Computing Unit (ACU), a mass production-oriented autonomous driving computing platform launched by Baidu, mainly features Apollo Pilot system and Intelligent Map service. As an important part of the Apollo platform, ACU is launched for mass production by the Baidu''s partners. Based on the different computing capabilities required by different scenarios, it is divided into three series of products: ACU-Basic, ACU-Advanced, and ACU-Professional.
We''ll introduce the latest advances on topics such as learning-to-learn, meta-learning, deep learning for robotics, deep reinforcement learning, and AI for manufacturing and logistics.
Deep learning algorithms have benefited greatly from the recent performance gains of GPUs. However, it has been unclear whether GPUs can speed up machine learning algorithms such as generalized linear modeling, random forests, gradient boosting machines, and clustering. H2O.ai, the leading open source AI company, is bringing the best-of-breed data science and machine learning algorithms to GPUs. We introduce H2O4GPU, a fully featured machine learning library that is optimized for GPUs with a robust python API that is a drop dead replacement for scikit-learn. We'll demonstrate benchmarks for the most common algorithms relevant to enterprise AI and showcase performance gains as compared to running on CPUs.
Road identification and route prediction in near real time remains a challenging problem for many geographic regions, particularly in the case of natural disasters or crisis situations. Existing methods such as manual road labeling or aggregation of mobile GPS track data are currently insufficient in dynamic scenarios. The frequent revisits of satellite imaging constellations may accelerate efforts to rapidly update road network and optimal path prediction, provided routing information can be extracted from imaging pixels. We'll demonstrate deep learning segmentation methods for identifying road center lines and intersections from satellite imagery, and inferring networks from these road segments. We'll also explore data quality requirements by comparing open source labels with-high precision labels created as part of the SpaceNet Roads challenge.
We'll showcase the MPC Virtual Production Platform called Genesis. While we won't be able to show any datasets currently in production, we'll showcase the technology and have some MPC original content to share.
We''ll present a summary of ongoing work that targets the use of newer GPU architecture (Pascal and Volta) features in real-time signal processing applications in radio astronomy telescopes, and outline the future growth path for this exciting new application of GPUs. With Pascal and Volta architectures, we''ll discuss the advantage of using higher memory bandwidth, half-single precision, and integer arithmetic in existing GPU-based correlator pipeline code. This is an ongoing effort between the National Centre for Radio Astrophysics and NVIDIA. We''ll look at various processing stages involved in the pipeline for exploring optimization possibilities, and highlight interesting results that were achieved. We''ll address in detail the effect of using half precision with respect to accuracy of performance and required library changes.
We'll present a complete open-source software stack for self-driving vehicles, called Autoware, and its open integration with the NVIDIA DRIVE platform. Autoware implements working modules of localization and 3D mapping with LiDAR and GNSS, object detection and traffic light recognition with deep learning, path planning with lattice and search methods, and vehicle dynamics control. Compute-intensive tasks of these modules are accelerated by using CUDA, and timing-aware tasks are protected by RTOS capabilities. We'll discuss the impact of CUDA acceleration on self-driving vehicles and its performance evaluation. Learn how Autoware enables any by-wire vehicles to become high-quality self-driving vehicles that can operate in real-world environments.
Pony.ai will share the key technological milestones it has achieved in the past several months of road testing in China, including the company's soft launch of China's first-ever autonomous vehicle robotaxi service. CEO James Peng will share the unique challenges posed by a Chinese road environment and how we leveraged deep learning and computational models to conquer those challenges. Pony.ai's mission is to build the safest and most reliable L4 autonomous driving technology. The startup was founded at the end of 2016 and is co-located in the heart of Silicon Valley and China.
DRIVE PX is an open platform for Autonomous Driving Ecosystem. Its been adopted by over 300 partners in the automotive ecosystem to develop solutions for vehicles that are intelligent and autonomous. This talk will outline the technical challenges facing development of autonomous intelligent vehicles and provide details of how the next generation of DRIVE AI car computer i.e. DRIVE Xavier and DRIVE Pegasus address these challenges.
We'll show how deep neural networks can ingest raw data from multiple types of sensors to generate improved perception results in real time, using processors fit for automotive mass production. Today's mass-produced driver-assistance systems are typically implemented with a late-fusion paradigm. This approach has a number of limitations in terms of accuracy, portability, and robustness to sensor failure. We'll propose an earlier stage of fusion, called Deep Sensor Fusion, where sensors transmit raw data over higher bandwidth in-vehicle networking, which is already used in mass production today.
The development of self-driving cars requires a strong relationships between partners in a different way as we know it today. This might be the only way to successfully bring self-driving vehicles on the road. ZF, Virtual Vehicle, and NVIDIA have joined forces to develop an AI-based L4 vehicle for urban scenarios in only six months; the so-called dream car. Learning while sleeping is the groundbreaking idea of the dream car which was realized in the second half of 2017. Without driving around, the car constantly learns and adapts itself based on data acquired from other cars driving around somewhere else in the world. The key is AI and ZF''s ProAI which was developed with NVIDIA in the past year. ProAI interprets the data in real-time, learns from it, validates the data, checks the plausibility, and adjusts the vehicle behavior. We''ll summarizes the implementation steps, HW and SW architecture, relevant driving/testing scenarios, our AI approach, and the challenges met in order to realize the dream car.
We''ll discuss the important emerging field of connected automated driving, including technical and policy topics in this area. We''ll provide background on vehicular safety communications and current deployments in various parts of the world. Vehicular communication will enable sensor data sharing between vehicles, which could be the key for achieving higher levels of automation. Novel artificial intelligence techniques exploiting sensor data (camera, radar, GPS etc.) from neighboring cars can be used for designing perception and mapping functionalities for automated vehicles. We''ll discuss results from field testing and show advantages of connected automated driving.
Learn about our application of deep learning techniques for perception systems in autonomous driving, reinforcement learning for autonomous systems, label detection in warehouse inventory management, and undergraduate engagement in this research. In collaboration with Clemson University''s International Center for Automotive Research, we''ve developed a perception module that processes camera inputs to provide environmental information for use by a planning module to actively control the autonomous vehicle. We''re extending this work to include an unsupervised planning module for navigation with reinforcement learning. We''ve also applied these techniques to automate the job of warehouse inventory management using a deep neural network running on a mobile, embedded platform to automatically detect and scan labels and report inventory, including its location in the warehouse. Finally, we''ll discuss how we involve undergraduate students in this research.
Autonomous vehicles require highly accurate, up-to-date maps for a safe, comfortable and optimized experience. TomTom's multi-source, multi-sensor approach leads to HD Maps that have greater coverage, are more richly attributed, and have higher quality than single-source, single-sensor maps. Autonomous vehicles also need to be able to access the latest, most up-to-date HD Maps with minimal latency. Learn how TomTom is taking on this challenge.
Deep learning/artificial intelligence methods are increasingly being deployed to enable new avenues of big-data-driven discovery in key scientific application areas such as the quest to deliver Fusion Energy identified by the 2015 CNN "Moonshots for the 21st Century" series as one of 5 prominent modern grand challenges. Princeton University''s associated R&D methods have been successfully applied to accelerate progress in reliably predicting and avoiding large-scale losses (called "disruptions") of the thermonuclear plasma fuel in magnetically-confined devices the largest of which is the $25B international ITER device a burning plasma experiment under construction with the potential to exceed "breakeven" fusion power (i.e., "power out = power in") by a factor of 10 or more.
Learn about the requirements gathering, solution analysis, benchmarking, and user testing techniques we implemented to decide on a virtual workstation configuration that met all the requirements for our CAD users, while meeting the density requirements to remain cost-effective. DENSO International America began investigating the feasibility of implementing CAD on VDI to simplify its workstation ecosystem in 2013. Several options were explored, but it wasn't until NVIDIA Quadro vDWS was released that we found a solution that met our performance and density requirements. We'll discuss our journey from vSGA/vDGA to vGPU. Starting with Kepler, through the latest Pascal architecture, we've performed benchmarking and user testing of each to demonstrate the benefit of each.
Miovision presents a video-based traffic analytics system, capable of tracking and classifying vehicles in real time throughout cities. The system leverages Jetson TX2 modules and inferencing to accurately classify vehicles at over 50 frames per second using single-shot multibox detection and DAC, a VGG-based network. We'll cover many of the issues our teams went through to design and implement the system, including data collection, annotation, training, incorporating continuous training, and deep learning iteration. We'll also illustrate how the measured traffic trends were used to reduce congestion and evaluate the health of traffic corridors.
A successful autonomous system needs to not only understand the visual world but also communicate its understanding with humans. To make this possible, language can serve as a natural link between high level semantic concepts and low level visual perception. We'll discuss recent work in the domain of vision and language, covering topics such as image/video captioning and retrieval, and question-answering. We'll also talk about our recent work on task execution via language instructions.
We''ll discuss the GPU accelerated Monte Carlo compute at JP Morgan which was architected for C1060 cards and revamped a few times as new architectures were released. The key features of the code are exclusive use of double precision, data caching, and code structure where significant amount of CPU pre-compute is followed by running multiple GPU kernels. On the latest devices, memory per flop is a throughput limiting factor for a class of our GPU-accelerated models. As byte/flop ratio is continuing to fall from one generation of GPU to the next, we are exploring the ways to re-architecture Monte Carlo simulation code to decrease memory requirements and improve TCO of the GPU-enabled compute. Obvious next steps are store less, re-calculate more, and unified memory.
Rent The Runway gets millions of visitors every day. We serve personalized recommendations to them based on browser behavior and some explicit feedback. To add to the complexity, we have multiple membership programs. Fashion has some unique challenges regarding seasonality, fit and feedback. We also have a unique business model where order fulfillment and reservations are tied together in a unique way. We have moved to a GPU first infrastructure to scale instead of Spark clusters. We will discuss how we are moving to power all our algorithms this way.
GPU virtualization in the Cloud has ushered in a new era for architects, builders, designers and engineers. In this case study session you will learn how TBI personnel are now using Autodesk applications including BIM 360, Stingray, Revit and Navisworks, through a digital workspace hosted on Citrix XenDesktop HDX 3D Pro running on Microsoft Azure NV-series virtual machines with NVIDIA Quadro Workstation technology. This technology stack enables TBI employees to work together in real time, from any location, while enjoying a highly optimized 3D user experience on any device, even the low-cost Raspberry Pi. In their technology journey, TBI progressed from an age of 2D flatland, to the more advanced age of optimization of 3D digital data, to the present-day era of interoperability and collaboration in a new age where connectivity is key.This session will also include a Citrix customer panel discussion. Hear from customers who have implemented virtualized 3D workloads to solve complex business challenges. Bring your questions along and join in on the knowledge sharing in an interactive setting.
Watch leading Healthcare AI Startups compete for cash prizes at the NVIDIA GTC 2018 Inception Awards Finale.
Watch leading Enterprise AI Startups compete for cash prizes at the NVIDIA GTC 2018 Inception Awards Finale
Watch leading Autonomous Systems AI Startups compete for cash prizes at the NVIDIA GTC 2018 Inception Awards Finale
Facebook's strength in AI innovation comes from the ability to quickly bring cutting-edge research into large scale production using a multi-faceted toolset. We'll discuss how Facebook leverages open source software to perform truly iterative AI research, scale it seamlessly for inference, and deploy it across the data center and mobile environments with ONNX.
We'll cover recent features and performance improvement in the NVIDIA collective communication library (NCCL). NCCL is designed to make computing on multiple GPUs easy and is integrated in most deep learning frameworks to accelerate training times. NCCL supports communication over Shared memory, PCI, NVLink, Sockets, and InfiniBand Verbs, to support both multi-GPU machines and multi-node clusters.
We'll present a deep learning system able to decide if two people are similar or not. This system use the global appearance of a person, not just the face, to perform the re-identification. Our system also provides attributes (top color, bottom color, genre, length of the clothes, and the hair). We'll describe how to train a system with tensorflow on a GPU cluster and how to use it on a global video analysis system running on GPU devices.
We''ll discuss how to get started with PyTorch from the creator of the project, Soumith Chintala. PyTorch is a fast and flexible deep learning framework that has been called a ''breath of fresh air'' by researchers and developers alike for its ease of use, flexibility, and similarity to python programming. It consists of an ndarray library that natively supports GPU execution (automatic differentiation engine that is flexible and fast), and an optimization package for gradient based optimization methods.