How to Build a Custom Synthetic Data Pipeline to Train AI Perception Models [A41329]
Omniverse Replicator SDK is highly extensible framework build on a a scalable Omniverse platform that enables physically accurate 3D synthetic data generation to accelerate the training and boost the performance of AI perception networks. In this session we are building upon core concepts of Omniverse Replicator APIs we introduced during Spring GTC 2022 by discussing a typical process of creating a custom workflow for generating synthetic data for training a perception network. Training a perception model with a synthetic dataset is a multi-step process that requires considerations around simulation-ready assets, contextual scene generation, plausible randomizations, different annotators, photo-realistic rendering, and writing data that is usable by downstream machine learning frameworks. We walk through this process with an example use case and along the way introduce enhancements we are bringing to the product. We are joined by our partners who have been building with us using Replicator SDK on the Omnivese platform and are accelerating what is possible.
Bhumin Pathak, Senior Product Manager, Omniverse Replicator, NVIDIA Alessandro Festa, Sr. Technical Product Manager, SmartCow AI Aman Kishore, CTO, Mirage Zachi Mann, CTO, Simulation Business Unit, Digital Manufacturing, Siemens
Primary Topic: Deep Learning - Training Industry Segment: Retail Monday, Sep 198:00 AM - 8:45 AM PDT
How to Build Simulation-Ready USD 3D Assets [A41339]
The next wave of industries and AI requires us to build physically accurate virtual worlds indistinguishable from reality. Building virtual worlds is hard, and today’s existing universe of 3D assets is inadequate, representing just the visual representation of an object. Whether building digital twins or virtual worlds for training and testing autonomous vehicles or robots, 3D assets require many more technical properties, requiring a need to develop and adopt novel processes, techniques, and tools. NVIDIA is introducing a new class of 3D assets called “SimReady” assets — the building blocks of virtual worlds. SimReady assets are more than just 3D objects — they encompass accurate physical properties, behavior, and connected data streams built on Universal Scene Description (USD). We’ll show you how you can get started with SimReady USD assets, and present the tools and techniques required to develop and test these assets.
Renato Gasoto, Robotics & AI Engineer, NVIDIA Beau Perschall, Director, Omniverse Sim Data Ops, NVIDIA
Primary Topic: World Simulation & Digital Twins Industry Segment: All Industries Monday, Sep 199:00 AM - 9:50 AM PDT
How CUDA Programming Works [A41101]
Come for an introduction to programming the GPU by the lead architect of CUDA. CUDA’s unique in being a programming language designed and built hand-in-hand with the hardware that it runs on. Stepping up from last year’s “How GPU Computing Works” deep dive into the architecture of the GPU, we’ll look at how hardware design motivates the CUDA language and how the CUDA language motivates the hardware design. This is not a course on CUDA programming. It’s a foundation on what works, what doesn’t work, and why. We’ll tell you how to think about a problem in a way that will run well on the GPU, and you’ll see how the CUDA programming model is built to run that way. If you’re new to CUDA, we’ll give you the core background knowledge you need — getting started begins with understanding. If you’re an expert, hopefully you’ll face your next optimization problem with a new perspective on what might work, and why.
, CUDA Architect, NVIDIA
Primary Topic: Accelerated Computing & Dev Tools - Programming Languages / Compilers Industry Segment: All Industries Monday, Sep 1910:00 AM - 10:50 AM PDT
How to Build a Digital Twin: Full-Design Fidelity Visualization and Aggregation of 3D Data [A41383]
Start learning how to build a digital twin in NVIDIA Omniverse. Building a digital twin is a complex process and there are multiple ways to get started. We’ll walk through a generalized example of one of the most common challenges in large-scale design and digital twin projects — being able to aggregate disparate 3D datasets (i.e., CAD, design, animation, or simulation) from many software ecosystems, and visualize them in full-design fidelity with no data loss or model decimation. Learn how NVIDIA Omniverse’s connected and collaborative Universal Scene Description-based workflows can help unite and unlock your data, and see examples of Omniverse Connectors and understand what questions to ask to determine which are best for your workflow. Be sure to check out “How to Build a Digital Twin: Bringing in Robotics” as a follow-up session.
Brian Harrison, Director, Software Product Management, NVIDIA
Primary Topic: World Simulation & Digital Twins Industry Segment: All Industries Monday, Sep 1911:00 AM - 12:00 PM PDT A41383- HOW TO BUILD A DIGITAL TWIN: FULL-DESIGN FIDELITY VISUALIZATION AND AGGREGATION OF 3D DATA.PDF
All You Need to Know About Implementing Parallel Pipelines with DeepStream [A41173]
NVIDIA DeepStream is the fastest way to create and deploy vision AI applications for all NVIDIA platforms, from edge to cloud. In this session, developers of all technical levels will learn how DeepStream can help them accelerate their development and accelerate time to market. We’ll demonstrate how to create and optimize ready-made reference applications for specific markets leveraging the latest models and plugins from DeepStream. We’ll show you how to extend the reference applications for your specific needs and share best practices for maximizing application performance. Finally, we’ll share the latest updates to the DeepStream SDK, developer assets, and more.
Carlos Garcia-Sierra, Product Manager, NVIDIA DeepStream, NVIDIA Alvin Clark, Sr. Solution Architect, NVIDIA
Primary Topic: Computer Vision - Intelligent Video Analytics Industry Segment: All Industries Monday, Sep 1912:00 PM - 1:00 PM PDT
Design, Train, and Evaluate Domain-Specialized Healthcare Imaging AI Models with MONAI [DLIT41275]
Learn about designing, training, and evaluating domain-specialized health-care imaging AI models using MONAI. Researchers and data scientists need a common foundation to perform training experiments and compare against the state of the art. MONAI provides domain-specific implementations to help kick-start development and research, including new features like self-supervised learning, Transformer-based Networks for Medical Imaging (UNETR), and DiNTS, a new neural architecture search method. We’ll introduce MONAI Core and then dive deep into the more technical features of MONAI, with a hands-on walkthrough of Self-Supervised Learning, AutoML/DiNTS, and researcher best practices.
Python Deep Learning Basic familiarity with medical imaging
Michael Zephyr, Developer Technical Lead, NVIDIA Dong Yang, Senior Applied Research Scientist, NVIDIA Yufan He, Applied Research Scientist in DLMED, NVIDIA
Primary Topic: Healthcare - Medical Devices Industry Segment: Healthcare & Life Sciences Monday, Sep 191:00 PM - 3:00 PM PDT
GTC 2022 Keynote - September [A41312]
Don’t miss this keynote from NVIDIA Founder and CEO, Jensen Huang. He will share how NVIDIA’s new computing platforms, cloud technologies, and advances in virtual collaboration are transforming every industry.
Jensen Huang, Founder and CEO, NVIDIA
Primary Topic: AI Strategy for Business Leaders Industry Segment: All Industries Tuesday, Sep 208:00 AM - 9:30 AM PDT
Future of AI: Fireside Chat with Turing Award Winners [A41191]
The goal of Artificial Intelligence (AI) is not only to push forward the frontier of machine intelligence but also to transform how this intelligence can become the engine to drive real world applications. The advancement made in the world of computing has enabled both industry and academia to develop advanced algorithms and complex models that are capable to imitate human brains. In this fireside conversation, we will be joined by 2018 ACM Turing Award winners, Yoshua Bengio, Geoffrey Hinton, and Yann LeCun who are the thought leaders and visionaries in AI, often referred to as godfathers of AI. Their contribution to the field of AI now underpins the proliferation of AI technologies being adopted around the world in various applications from natural language processing, medical imaging to autonomous machines. They will provide their perspective on how AI will transform various aspects of our life and how applications driven by AI will help us in solving the most pressing issues of our time.
Sanja Fidler, VP of AI Research, NVIDIA Yoshua Bengio, Founder & Scientific Director at Mila - Quebec Artificial Intelligence Institute, Full Professor at University of Montreal Yann LeCun, VP & Chief AI Scientist at Meta (FAIR), Silver Professor at New York University Geoffrey Hinton, VP and Engineering Fellow at Google, Chief Scientific Advisor at Vector Institute, Emeritus Professor at University of Toronto
Primary Topic: Deep Learning - Inference Industry Segment: Higher Education / Academia Tuesday, Sep 2010:00 AM - 11:00 AM PDT
The Rise of Transformer AI and Digital Twins in Healthcare [A41228]
The healthcare industry is generating about one-third of the world’s data. Breakthroughs in AI, accelerated computing, and real-time sensing have created new opportunities for drug discovery and healthcare delivery. Transformer AI models are powering a new era of life sciences, helping researchers encode the structure and function of biology and chemistry, making sense of unstructured patient data, and improving detection and diagnosis in medical imaging. At the same time, advances in digital twin technology are powering researchers and clinical teams to simulate cells, organs, and surgeries to better understand workflows and improve patient outcomes. Learn how NVIDIA’s Clara computing platform is helping scientists, researchers, and engineers develop AI from data to model to device to software service.
Kimberly Powell, General Manager and VP, Healthcare and Life Sciences, NVIDIA
Primary Topic: Healthcare - Drug Discovery Industry Segment: Healthcare & Life Sciences Tuesday, Sep 2011:00 AM - 11:25 AM PDT
Maximizing GPU utilization in Large Scale Machine Learning Infrastructure [A41246]
The products of ByteDance heavily rely on machine learning (ML) and deep learning (DL). Large-scale clusters are built to support these workloads including for both model training and real time online inferencing. In this talk, I will share how we make best use of our complex infrastructure to simultaneously run ML training and inference workloads in our extensive GPU clusters. Our objectives are to maximize GPU resource utilization while providing our users service level guarantees.
Yibo Zhu, Director of Machine Learning Systems, ByteDance
Primary Topic: Deep Learning - Inference Industry Segment: Consumer Internet Tuesday, Sep 2012:00 PM - 1:00 PM PDT
5G Killer App: Making Augmented and Virtual Reality a Reality [A41234]
The killer apps for 5G must be able to tap into the new network capabilities and perform solutions that weren’t previously possible in early-generation networks. Extended reality (XR, comprising augmented, virtual, and mixed realities) is consistently envisioned as one of the key killer apps for 5G, as XR requires ultra-low latency and large bandwidths to deliver wired-equivalent experiences for users. Verizon, AWS, and Ericsson will share how they’re collaborating to combine 5G and XR technology with NVIDIA GPUs, RTX vWS, and CloudXR to build the infrastructure for commercial XR services across a variety of industries.
Veronica Yip, Product Manager & Product Marketing Manager, NVIDIA Peter Linder, Head of 5G Marketing, North America, Ericsson David Randle, Global Head of GTM for Spatial Computing, Amazon Web Services Balaji Raghavachari, Executive Director of Device Technology, Verizon
Primary Topic: XR (Virtual and Augmented Reality) Industry Segment: Telecommunications Tuesday, Sep 201:00 PM - 1:50 PM PDT
NVIDIA’S Earth -2: Digital Twins For Weather and Climate [A41326]
NVIDIA’s recently launched Earth-2 initiative aims to build digital twins of the Earth to address one of the most pressing challenges of our time, climate change. Earth-2 aims to improve our predictions of extreme weather, projections of climate change, and accelerate the development of effective mitigation and adaptation strategies — all using the most advanced and scientifically principled machine learning methods at unprecedented scale. Combining accelerated computing with physics-informed machine learning at scale, on the largest supercomputing systems today, Earth-2 will provide actionable weather and climate information at regional scales. Here we review progress and challenges. Highlights include data-driven global weather prediction at unprecedented resolution, speed and scale, calibrated ensemble forecasting to achieve realistic atmospheric chaos dynamics, engineering innovations to address massive scale, and updates to our strategy to expand beyond weather towards future climate emulation
Anima Anandkumar, Senior Director of ML Research, NVIDIA Karthik Kashinath, Principal Engineer and Scientist, AI-HPC and Engineering Lead, Earth-2, NVIDIA Mike Pritchard, Director for Climate Simulation Research, NVIDIA
Primary Topic: HPC - Climate / Weather / Ocean Modeling Industry Segment: HPC / Supercomputing Tuesday, Sep 202:00 PM - 3:00 PM PDT A41326 NVIDIA’S EARTH -2 DIGITAL TWINS FOR WEATHER AND CLIMATE.PDF
Ray Tracing: How NVIDIA Solved the Impossible! [A41171]
Light transport simulations are the industry-standard way of creating convincing photorealistic imagery and are widely used in creating animation movies, computer animations, and medical and architectural visualizations, among many other notable applications. These techniques simulate how millions of rays of light interact with a virtual scene. However, they are extremely expensive to compute — a photorealistic image can take anywhere from several minutes to weeks to compute. We’ll describe how recent research works from NVIDIA tackle this problem and create beautiful, photorealistic images in real time.
, Independent Scientist, Two Minute Papers
Primary Topic: Graphics - Real-Time Rendering and Ray Tracing Industry Segment: Media & Entertainment Wednesday, Sep 217:00 AM - 7:25 AM PDT
Deep Learning Demystified [A41165]
Artificial intelligence has evolved and improved methods for data analysis and complex computations, solving problems that seemed well beyond our reach only a few years ago. Today, deep learning is transforming every industry, from health care and retail to automotive and financial services. Join us to understand the fundamentals of accelerated data analytics, high-level use cases, and problem-solving methods. We’ll cover: • Demystifying artificial intelligence, machine learning, and deep learning; • Understanding the key challenges organizations face in adopting this new approach and how to address them; and • Learning about the latest tools and technologies, along with training resources, that can help deliver breakthrough results.
Ozzy Johnson, Director, Solutions Engineering, NVIDIA
Primary Topic: Deep Learning - Training Industry Segment: All Industries Wednesday, Sep 218:00 AM - 8:50 AM PDT
Lowe’s: Fully Digitizing the World of Home Improvement [A41319]
Join this session to learn how FORTUNE 50 retailer Lowe’s is building the future of home improvement. From our products to our stores to our customers’ homes, Lowe’s is using digital twins to create new possibilities. Whether helping to visualize the result of a room redesign or running simulations to optimize the layout of each store, Lowe’s is removing friction for our customers and associates.
Cheryl Friendman, Vice President, Lowe’s Innovation Labs, Lowe’s Mason Sheffield, Director of Creative Technology, Lowe’s Innovation Labs, Lowe’s
Primary Topic: World Simulation & Digital Twins Industry Segment: Retail Wednesday, Sep 2110:00 AM - 10:50 AM PDT
ILM: Leveraging AI in Visual Effects and StageCraft Virtual Production [A41351]
When it comes to modern visual effects and cutting-edge virtual production, creating photorealistic digital sets and environments that can be manipulated in real time is the name of the game. One key element in any exterior environment is a sky dome often utilized for both lighting the virtual scene and in-camera effects. Working with NVIDIA’s powerful AI-enabled DeepSearch, the team from Industrial Light & Magic will showcase how they leverage their incomparable asset library and Omniverse Enterprise to provide filmmakers the ultimate flexibility when developing the right look and ideal lighting for a scene.
TJ Galda, Director of Product Management, NVIDIA Landis Fields, Real-Time Principal Creative, Industrial Light & Magic
Primary Topic: Graphics - Animation / VFX / Virtual Production Industry Segment: Media & Entertainment Wednesday, Sep 2111:00 AM - 12:00 PM PDT
Getting Started with Ray Tracing Graphics Tools [A41104]
NVIDIA GPU architectures provide immense processing power for creating jaw dropping computer graphics. To fully harness this power and bring real-time ray-traced photorealistic graphics into reality, developers need the best tools. We’ll go through what tools are available to allow you to debug, profile, and optimize your modern graphics application. We’ll cover how you can use the newest features in Nsight Graphics, Nsight Systems, and Nsight Aftermath to create a fast, high-fidelity and stable graphics application that utilizes ray tracing to maximum potential.
Avinash Baliga, Software Engineering Manager, NVIDIA Aurelio Reis, Director, Graphics Developer Tools, NVIDIA
Primary Topic: Graphics - Real-Time Rendering and Ray Tracing Industry Segment: Game Development Wednesday, Sep 2112:00 PM - 12:50 PM PDT A41104 - GETTING STARTED WITH RAY TRACING GRAPHICS TOOLS.PDF
Training and Deploying Multi-Stage Recommender Systems [DLIT41349]
This tutorial introduces the Merlin framework which aims to make the development and deployment of recommender systems easier, providing methods for evaluating existing approaches, developing new ideas and deploying them to production. There are many techniques, such as different model architectures (e.g. MF, DLRM, DCN, etc), negative sampling strategies, loss functions or prediction tasks (binary, multi-class, multi-task) that are commonly used in these pipelines. Merlin provides building blocks that allow RecSys practitioners to focus on the “what” question in designing their model pipeline instead of “how”. Supporting research into new ideas within the RecSys spaces is equally important and Merlin supports the addition of custom components and the extension of existing ones to address gaps.
In this tutorial, participants will learn: How to easily implement common recommender system techniques for comparison Deploying recommender systems - using an open source framework Merlin and its libraries
Prerequisite(s): Basic familiarity with Python Deep learning and deep learning frameworks
Radek Osmulski, Senior System Software Engineer, NVIDIA Ronay Ak, Senior Data Scientist, NVIDIA Benedikt Schifferer, Deep Learning Engineer, NVIDIA
Primary Topic: Recommenders / Personalization Industry Segment: Consumer Internet Thursday, Sep 227:00 AM - 9:00 AM PDT
Technical Breakout: How to Build a Digital Twin with Omniverse [SE41415]
Do you want to learn how to get started building a digital twin in Omniverse? After watching “A41384 How to Build a Digital Twin: Full Design Fidelity Visualization & Aggregation of 3D Data” and “A4138 How to Build a Digital Twin: Bringing in Robotics”, join this interactive support session with Omniverse technical product managers, engineers, and solutions architects where you can bring your questions, issues, and ideas specific to your use case. Our technical leads are here to help, whether you are looking to validate your use case, understand where to start in the process, or trying to solve technical issues.
Brian Harrison, Director, Software Product Management, NVIDIA Mike Geyer Sean Young Raphael Boehm Teresa Conceicao
Primary Topic: World Simulation & Digital Twins Industry Segment: All Industries Thursday, Sep 2210:00 AM - 11:30 AM PDT