Research Conference Sessions
Join us to explore research that spans all industries and learn from some of the brightest minds in AI, foundation models, robotics, and more.
Featured Sessions
Insights From NVIDIA Research [S73202]
Bill Dally | Chief Scientist and SVP of Research | NVIDIA
The talk will give some highlights from NVIDIA Research for the past year. Detailed topics will be disclosed closer to the event.
Graduate Fellowship Fast Forward Talks [S71166]
Bill Dally | Chief Scientist and SVP of Research | NVIDIA
Bailey Miller | PhD Student | CMU
Nicklas Hansen | PhD Student | UC San Diego
Payman Behnam | PhD Student | Georgia Tech
Reinhard Wiesmayr | PhD Student | ETH Zurich
Songwei Ge | PhD Student | Univ. of Maryland, College Park
Toluwanimi Odemuyiwa | PhD Student | UC Davis
Yiming Li | PhD Student | NY University
Yue Zhao | PhD Student | UT Austin
Zhiqi Li | PhD Student | Nanjing University
Zihao Ye | PhD Student | Univ of Washington
Join a special presentation from our 2024-25 Graduate Fellowship recipients to learn "what's next" from the world of research and academia. Sponsored projects involve a variety of technical challenges, including machine learning, robotics, computer vision, computer graphics, wireless communication systems, high performance computing, autonomous driving, and AI efficiency. We believe that these minds lead the future in our industry and we're proud to support the 2024-25 NVIDIA Graduate Fellows. We'll also announce the 2025-26 Graduate Fellows at this session. For more information on the NVIDIA Graduate Fellowship program, visit www.nvidia.com/en-us/research/graduate-fellowships
NVIDIA Cosmos World Foundation Models for Autonomous Driving Development [S73198]
Sanja Fidler | VP, AI Research | NVIDIA
NVIDIA Cosmos is a platform that aims to accelerate physical AI development. We'll discuss how the Cosmos World Foundation Models, when specialized to the autonomous driving (AV) domain, can enhance AV development workflows.
Igniting the Real Robot Revolution Requires Closing the “Data Gap” [S74739 ]
Ken Goldberg | Co-Founder and Chief Scientist, Ambi Robotics | William S. Floyd Distinguished Chair, College of Engineering, UC Berkeley
Imagine a future where robots handle billions of items with AI to support an aging population that doesn’t have enough human workers. Computation is increasing rapidly but much more data is needed to ignite this robotics revolution. Several unicorn startups emerged in the past year to generate humanoid robot data but I’ll show that there is still a 100,000x “data gap” between large vision-language models and current robot models. Goldberg will present the state-of-the-art in robotics and review the options for generating robot data, including the case study of Ambi Robotics, which has collected 200,000 hours of real robot data from their award-winning robot systems that have sorted 100 million real consumer packages. This is allowing Ambi Robotics to close the data gap for a practical subclass of robot skills to enable a new generation of real industrial robots.
Important: Near capacity, highly suggest arriving early. Attendees are let in on a first-come, first-served basis.
Multi-Modality and Multi-AI Agents for Hardware Design [S72294]
Mark HO | Sr. Research Scientist | NVIDIA
Hardware design presents numerous challenges due to its complexity and rapidly advancing technologies. These challenges result in longer turnaround times (TAT) for optimizing performance, power, area, and cost (PPAC) during synthesis, verification, physical design, and reliability loops. Large language models have demonstrated a remarkable ability to comprehend and generate natural language at scale, leading to numerous potential applications and benefits across various domains. Successful LLM-based agents for hardware design can significantly reduce TAT, enabling faster product cycles, lower costs, improved design reliability, and a reduced risk of costly errors. Cutting-edge multi-AI agents have shown promising results in Hardware Description Language (HDL) generation, layout optimization, design rule code generation, and Multi-Corner Multi-Mode (MCMM) timing report analysis and debugging at both the block and path levels. Notably, the developed multi-AI agents not only achieve a 94.2% functionally correct pass rate on the VerilogEval-Human v2 benchmark, but also deliver a 60X speedup in MCMM timing report analysis compared to experienced human engineers.
Diffusion Models: The Swiss Army Knife for Generative AI in Science [S73269]
Arash Vahdat | Principal Scientist and Research Manager | NVIDIA
Diffusion models have transformed generative AI by enabling breakthroughs in diverse applications, such as images, videos, and speech. We'll show that these versatile frameworks extend beyond traditional domains and hold substantial potential for scientific data modeling. Through case studies spanning molecular generation, protein design, and weather modeling, we'll explore the unique opportunities and challenges involved in leveraging diffusion models to answer complex scientific questions.
AI for Humanoid Robots [S73182]
Pieter Abbeel | Professor | University of California, Berkeley
Humanoid robotics hardware is seeing a major acceleration in progress. It's reasonable to expect commercially viable humanoid hardware to exist within three years. The question is, how do we build the AI that can make these robots helpful? Humanoids have high dimensional action spaces and high dimensional multimodal observation space, posing challenges to existing paradigms. We'll explore these challenges in more depth and highlight some promising paths forward.
Frontiers of AI and Computing: A Conversation With Yann LeCun and Bill Dally [S73208]
Bill Dally | Chief Scientist and SVP of Research | NVIDIA
Yann LeCun | Chief AI Scientist at Meta | Professor at New York University
As artificial intelligence continues to reshape the world, the intersection of deep learning and high performance computing becomes increasingly crucial. This talk brings together Yann LeCun, a pioneer in deep learning and the chief AI scientist at Meta, and Bill Dally, a leading computer architect and chief scientist at NVIDIA, to explore the future of AI models, hardware accelerators, and the evolving computational landscape.
The discussion will cover:
• The next breakthroughs in deep learning and AI architectures
• How hardware innovation drives AI efficiency and scalability
• Challenges in training large-scale models and real-time AI inference
Bending Scaling Laws with Brighter Algorithms [S73204]
Yejin Choi | Senior Research Director | NVIDIA
Scaling laws tell us "more is more" — that brute-force scaling of data and compute smoothly enhances AI capabilities. Yet, in reality, this path is becoming increasingly unsustainable, which calls for algorithmic innovations to bend the scaling laws and accelerate progress. I will present our recent efforts in this spirit, including gradient-based methods for enhancing synthetic data, symbolic search algorithms for supercharging test-time reasoning, and test-time training that enables continuous model learning and adaptation.
Introduction to NVIDIA Halos – Strategy for AV Safety [S74743]
Marco Pavone | Director, Autonomous Vehicle Research | NVIDIA
Jonas Nilsson | Director of Safety Engineering, Head of AV Safety | NVIDIA
Karl Greb | Sr. Director Safety Engineering | NVIDIA
We'll introduce NVIDIA Halos, a full-stack comprehensive safety system that unifies vehicle architecture, AI models, chips, software, tools, and services to ensure the safe development of autonomous vehicles (AVs) from cloud to car. After the introduction, we'll dive deep into design-time safety guardrails, from platform to algorithmic safety.
From Research to Production: Transforming AV Technology With AI [S72707]
Jose Alvarez | Director, Research | NVIDIA
Boris Ivanovic | Senior Research Scientist and Manager, Autonomous Vehicle Research | NVIDIA
Learn how AI is transforming the development of autonomous vehicle (AV) technology, both in the data center and in the vehicle. We’ll focus on three key areas: (1) self-supervised training pipelines unlocking data scaling at unprecedented levels; (2) composing ingredients for universal and controllable end-to-end simulation; and (3) architecting end-to-end autonomy stacks that leverage foundation models (such as Cosmos world foundation models and vision language models) to seamlessly generalize to long-tail events.
We'll demonstrate how applying the latest AI advancements across the three main computers of a car (training, simulation, and on-vehicle) can significantly accelerate AV development and large-scale productization.
From Research to Production: Transforming AV Technology With AI [S72707]
Jose Alvarez | Director, Research | NVIDIA
Boris Ivanovic | Senior Research Scientist and Manager, Autonomous Vehicle Research | NVIDIA
Learn how AI is transforming the development of autonomous vehicle (AV) technology, both in the data center and in the vehicle. We’ll focus on three key areas: (1) self-supervised training pipelines unlocking data scaling at unprecedented levels; (2) composing ingredients for universal and controllable end-to-end simulation; and (3) architecting end-to-end autonomy stacks that leverage foundation models (such as Cosmos world foundation models and vision language models) to seamlessly generalize to long-tail events.
We'll demonstrate how applying the latest AI advancements across the three main computers of a car (training, simulation, and on-vehicle) can significantly accelerate AV development and large-scale productization.
Enable Human Telepresence with Generative AI [S72755]
Shalini De Mello | Director of Research, New Experiences and Distinguished Research Scientist | NVIDIA
We present the key research challenges in enabling highly photorealistic digital human synthesis. Departing from the industry-wide de-facto approach of using expensive multi-view human data captures, we present our innovative techniques that successfully harness the power of generative AI, large internet-scale data, and state-of-the-art neural 3D scene representations to create AI-based foundation models for highly photorealistic 3D digital human synthesis. We present the numerous downstream applications enabled by these foundational pieces of technology for digital humans, including, AI-mediated telepresence, content creation for film and gaming, and digital agents. We deep dive into AI-mediated telepresence and highlight many practical considerations encountered while building real-time cloud-enabled streaming systems for such telepresence solutions. Finally, we highlight open frontiers, which are avenues for further research in digital human understanding and creation.
An Introduction to NVIDIA Cosmos World Foundation Models [S72431]
Ming-Yu Liu | VP Generative AI Research | NVIDIA
NVIDIA Cosmos is transforming how robots and autonomous vehicles learn, accelerating physics-aware synthetic video generation with world foundation models. Join Ming-Yu, NVIDIA’s VP of Generative AI Research, to see how Cosmos is democratizing physical AI development—giving developers open models and tools to build custom world models faster than ever.
Important: Near capacity, highly suggest arriving early. Attendees are let in on a first-come, first-served basis.
Advancing AI Reasoning: From Games to Complex Problem Solving [S73862]
Vartika Singh | Strategic AI Lead | NVIDIA
Noam Brown | Research Scientist | OpenAI
Bryan Catanzaro | VP, Applied Deep Learning Research | NVIDIA
The evolution of artificial intelligence has seen remarkable milestones, particularly in developing systems capable of advanced reasoning. This panel will explore the journey from AI mastering strategic games to tackling complex, real-world problems. We will delve into the significance of integrating deliberate, System 2 thinking into AI models, enabling them to process information more thoroughly and solve intricate tasks. The discussion will highlight the challenges and breakthroughs in enhancing AI’s reasoning capabilities, the role of reinforcement learning and self-play in training, and the implications of these advancements for future applications. By examining case studies and their performance in complex reasoning tasks, we aim to shed light on the current state and future directions of AI reasoning research.