Explore featured sessions for deep learning.

Watch sessions from GTC on-demand to learn about the latest breakthroughs in deep learning and artificial intelligence (AI) applications. We’ll feature speakers from various leading industries and domains, along with demos, related content, and more. Stay tuned for more information.


Jensen Huang | NVIDIA | Founder and CEO

Watch the keynote replay to hear Jensen Huang's insight into how NVIDIA is driving the rapid pace of technology advancements to help solve the world's toughest challenges.

Featured Speakers

Deep Learning: What the Future Might Hold

Ilya Sutskever 
Co-founder and Chief Scientist, OpenAI

Applying Conversational AI for Business - Transcription, Virtual Assistants and Chatbots

Heather Nolis
Principal Machine Learning Engineer, T-Mobile

Ambient Intelligence: Illuminating the Dark Spaces of Health Care

Fei-Fei Li
Denning Co-Director of Stanford’s Human-Centered AI Institute, Stanford University

Sessions By Topics

Conversational AI

  • Technical Leaders in Natural Language Processing

    • Adam Henryk Grzywaczewski, Senior Deep Learning Data Scientist, NVIDIA
    • Mohammad Shoeybi, Research Manager, NVIDIA
    • Magnus Sahlgren, Head of Natural Language Processing, Research Institutes of Sweden (RISE)

    Large language models seem to be the future of NLP, but building them is far from trivial. This panel gathers leaders in NLP to discuss the key technical challenges that had to be overcome to make large language models a reality.

  • How to Quickly Build Working ASR Systems as a Startup

    • Marie Granier, CEO, Lexistems
    • Thomas Pellissier-Tanon, VO Research, Lexistems

    Lexistems is a French startup focusing on conversational AI to build better search systems. Learn how the NVIDIA platform, including NeMo and Triton, helped us quickly build a state-of-the-art automatic speech recognition (ASR) system in multiple languages.

  • Your First Steps to Design an Intelligent Assistant for Hands-Free Applications

    • David Taubenheim, Senior Data Scientist, NVIDIA

    Practical conversational AI systems like intelligent assistants (IA) now provide spoken responses to users’ requests in a fraction of a second thanks to accelerated computing. Learn the initial steps to create an example IA for a hands-free application: performing an upgrade to an aircraft while keeping maintenance notes.

  • How Hugging Face Delivers 1 Millisecond Inference Latency for Transformers in Infinity

    • Jeff Boudier, Head of Product, Hugging Face

    Learn how Hugging Face achieved 1 millisecond Transformers inference for customers of its new Infinity solution. Transformers models conquered natural language processing with breakthrough accuracy. But these large models and complex architectures are too challenging for most companies to put in production with enough performance to power real-time experiences like semantic search, and large workloads like frequent text classification over large datasets. Infinity helps achieve unprecedented single-digit millisecond latency on models like BERT.

Recommender Systems

  • Building and Deploying Recommender Systems Quickly and Easily with NVIDIA Merlin

    • Even Oldridge, Senior Manager, Recommender Systems Framework Team, NVIDIA

    In this session, we'll outline the open source NVIDIA Merlin framework, go over the library, and deep dive into the new features that have been added in the past year. NVIDIA Merlin now supports CPU implementations, as well as simpler ML models including collaborative filtering through the Implicit library. Join us to learn how you can leverage Merlin in your own RecSys pipeline.

  • Applying GPUs to Very Large Scale Recommender Machine Learning Inference

    • Xiang Wu, Ad Ranking Tech Lead, Snap
    • Dmitry Sigaev, Ml Engineer, Snap

    At Snap, we apply the latest ML technology to find engaging content and relevant ads. ML plays a central role in delivering long-term value to Snapchatters, creators, and our advertisers. In this talk, we’ll share our experiences and insights in applying the GPU technology to accelerate ML model inference.

Computer Vision

  • Train. Adapt. Optimize. Supercharge your AI Development Worklflow and Application Development with NVIDIA TAO Toolkit.

    • Chintan Shah, Sr. Product Manager Intelligent Video Analytics, NVIDIA
    • Akhil Docca, Senior Product Marketing Manager, NVIDIA

    In this talk we will show you the power and ease of leveraging NVIDIA’s latest tools including NVIDIA TAO Toolkit to supercharge your AI development workflow. This one platform does the Training, Adaptation, and Optimization (TAO) and integrates with NVIDIA’s latest application frameworks. We will explore customer examples, demos, and sample applications to get you up and running in minutes.

  • Accelerating the Development of Next-Generation AI applications with DeepStream 6.0

    • Alvin Clark, Product Marketing Manager, NVIDIA

    Edge AI and distributed processing applications are at the forefront of the deep learning revolution. Learn how DeepStream can help you create the next big thing for retail, manufacturing, healthcare, smart cities, and beyond. Harnessing the power of DeepStream 6.0 has never been easier with the introduction of Graph Composer, while the addition of temporal video processing adds a number of new use cases.

  • Addressing Generalization and Scalability Challenges in Satellite Imagery Analysis Using NVIDIA GPUs and Deep Learning

    • Philipe Ambrozio Dias, Research Associate, Oak Ridge National Laboratory (ORNL)
    • Lexie Yang, Research Scientist, Oak Ridge National Laboratory (ORNL)

    We’ll share our experiences on extracting building footprint and roads from satellite imagery datasets, using multi-GPU and multi-node HPC platforms to leverage NVIDIA’s DGX machines and ORNL’s Summit supercomputer. We'll also discuss our research using Jetson edge computing devices and few-shot learning for unmanned aerial survey utility pole inspection and damage assessment.

  • Perceive, Reason, Act: Closing the AI Loop

    • Gal Chechik, Director of AI, NVIDIA

    AI aims to build systems that interact with their environment, with people, and with other agents in the real world. This vision requires combining perception with reasoning and decision-making. It poses hard algorithmic challenges: from generalizing effectively from few or no samples to adapting to new domains to communicating in ways that are natural to people. I'll discuss our recent research thrusts for facing these challenges. These will include approaches to model the high-level structure of a visual scene; leveraging compositional structures in attribute space to learn from descriptions without any visual samples; and teaching agents new concepts without labels, by using elimination to reason about their environment.

Deep Learning Inference

  • Using Apache TVM for Automatic, Machine Learning-powered TensorCore Code Generation

    • Jason Knight, Co-founder and Chief Product Officer, OctoML
    • Junru Shao, Principal Engineer, OctoML

    Learn how the next generation of improvements from Apache TVM can automatically generate efficient CUDA for anything from matrix multiplication to arbitrary deep learning kernels written in Python. The TVM project last year acquired the ability to fully automatically generate computational schedules through a process called auto-scheduling. While auto-scheduling delivers high performance across a variety of deep learning models, it is currently unable to leverage complex hardware instructions such as NVIDIA’s WMMA for TensorCore acceleration. We'll describe how we removed this limitation and demonstrate its effectiveness by enabling cuBLAS and CUTLASS levels of performance. See how this next generation of auto-scheduling unifies the treatment of custom schedules, parameterized schedule templates for semi-automatic scheduling, and full template-free auto-scheduling.

  • AI Inference Workloads: Solving Challenges Beyond Training Models (Presented by Run.ai)

    • Ronen Dar, CTO, Run:AI

    AI/machine learning (ML) teams are under pressure to optimize and manage AI inference workloads in production and deliver a return on investment. Ronen Dar, CTO and co-founder of Run:AI, will give an overview of the challenges in moving ML prototypes to production, and how best-in-class ML teams are successfully overcoming these hurdles.

  • Accelerate Deep Learning Inference in Production with TensorRT

    • Joohoon Lee, Group Product Manager, NVIDIA
    • Jay Rodge, Product Marketing Manager, NVIDIA

    TensorRT is an SDK for high-performance deep learning inference used in production to minimize latency and maximize throughput. The latest generation of TensorRT provides a new compiler to accelerate specific workloads optimized for NVIDIA GPUs. Deep learning compilers need to have a robust method to import, optimize, and deploy models. We'll show a workflow to accelerate frameworks including PyTorch, TensorFlow, and ONNX. New users can learn about the standard workflow, while experienced users can pick up tips and tricks to optimize specific use-cases.

  • Accelerate PyTorch Inference with TensorRT

    • Naren Sivagnanadasan, Automotive Deep Learning Solution Architect, NVIDIA
    • Nick Comly, Product Manager for Deep Learning Inference - TensorRT, NVIDIA

    Learn how to accelerate PyTorch inference without leaving the framework with Torch-TensorRT. Torch-TensorRT makes the performance of NVIDIA’s TensorRT GPU optimizations available in PyTorch for any model. You'll learn about the key capabilities of Torch-TensorRT, how to use them, and the performance benefits you can expect. We'll walk you through how to easily transition from a trained model to an inference deployment fine-tuned for your specific hardware, all with just a few lines of familiar code. If you want more technical details, the second half of the talk will give you a chance to deep dive into how Torch-TensorRT operates, the mechanics of key features, and a few in-depth examples.

Deep Learning Training

  • Torch-ort Can Accelerate PyTorch Experiments on NVIDIA GPUs Using ORTModule

    • Manash Goswami, Principal Program Manager, Microsoft
    • Timothy Harris, Principal Architect, Microsoft

    Learn how ORTModule can accelerate deep learning PyTorch training by up to 40% with only a three-line change in the training script. Understand how ORTModule uses graph level optimizations, memory management optimizations, and composability with DeepSpeed to improve training throughput for large scale distributed training experiments. We'll provide an in-depth review of the improvements in ORTModule and share the performance data for training well-known reference models on NVIDIA GPUs.

  • Constructing Cross-sectional Systematic Strategies by Learning to Rank

    • Daniel Poh, Ph.D. Researcher, Oxford-Man Institute of Quantitative Finance, University of Oxford
    • Stefan Zohren, Associate Professor, Oxford-Man Institute of Quantitative Finance, University of Oxford

    Recently there has been increased interest in applying deep learning techniques in finance. After reviewing some of the latest advances in this field, we focus on a case study that uses techniques from information retrieval in the context of cross-sectional systematic trading strategies. In this session, we describe a new framework to enhance cross-sectional portfolios with Learning to Rank algorithms as well as an extension where we explore incorporating a high-frequency market fragility measure to rankers.

  • Improving Fraud Detection with Auto-encoders

    • Benny Lifschitz, Director of Data Science , Riskified

    Today we are witnessing a growing number of fraud attempts in e-commerce, while their sophistication is increasing. Meanwhile, merchants expect a very high accuracy and approval rate. In this session, we'll describe (1) the main frauds we see, and how they're changing; (2) the difficulties of regular models; and (3) how auto-encoders can detect fraud. We utilize NVIDIA GPU clusters on AWS to accelerate training to enable fast train, test, validate, and deploy.

Deep Learning Frameworks

  • Ambient Intelligence: Illuminating the Dark Spaces of Health Care

    • Fei-Fei Li, Denning Co-Director of Stanford’s Human-Centered AI Institute, Stanford University

    Among the greatest challenges in health care is the vulnerability of patients to accidents and the progression of symptoms that often go unnoticed until it’s too late. In this talk, Fei-Fei Li will discuss her research into “ambient intelligence” — smart, sensor-based solutions that augment the awareness and capabilities of human clinicians to close the gaps of patient safety and care.

  • Deep Learning: What the Future Might Hold

    • Ilya Sutskever, Co-founder and Chief Scientist, OpenAI

    What is deep learning, and how does it work? OpenAI’s Chief Scientist Ilya Sutskever will discuss the history of deep learning, what it can do today, and what the future might hold.

  • Can neural networks learn to reason? Insights, plus a long term vision for Apple’s new machine learning research group

    • Samy Bengio, Senior Director of Machine Learning Research, Apple

    The successes of deep learning critically rely on the ability of neural networks to output meaningful predictions on unseen data --- generalization. Yet despite its central importance, there remain fundamental open questions on how neural networks generalize. In particular, how much do neural networks rely on memorization --- seeing highly similar training examples --- and how much are they capable of human-intelligence styled reasoning --- identifying abstract rules underlying the training data? In this presentation, I will propose an approach to explore the limits of neural network generalization which reveals both subtle failures and surprising successes.

See deep learning session highlights from the previous GTC. Get ready for what’s to come.

Find the complete GTC On-Demand playlist here.

Explore More Conference Topics

Explore All Session Topics

NVIDIA Developer Program

Get the advanced tools and training you need to successfully build applications on all NVIDIA technology platforms.

Accelerate your Startup

Explore the startup track at GTC to learn how NVIDIA Inception can fuel your growth through go-to-market support, world-class training, and technology assistance.

Get Hands-On Training

Interested in developing key skills in AI, accelerated data science, or accelerated computing? Get hands-on instructor-led training from the NVIDIA Deep Learning Institute (DLI) and earn a certificate demonstrating subject matter competency.