Enabling Ubiquitous 3D Intelligence via Multi-Granular Algorithm-Hardware Synergy

Speaker: Chaojian Li
Georgia Institute of Technology

Title: Enabling Ubiquitous 3D Intelligence via Multi-Granular Algorithm-Hardware Synergy

Date: Wednesday, 26 March 2025

Time: 9:30am - 10:30am

Join Zoom Meeting: https://hkust.zoom.us/j/96688516988?pwd=qfj1PQIjEi0I75lwVGfY7PurdPDRBW.1
Meeting ID: 966 8851 6988
Passcode: 202526

Abstract:

3D intelligence is emerging as one of the next frontiers of artificial intelligence, extending beyond text and image processing to enable richer and more immersive experiences. However, realizing this promise comes with significant computational and memory challenges, particularly for real-time applications on resource-constrained edge devices. Achieving ubiquitous 3D intelligence requires overcoming challenges related to efficiency, accessibility, and adaptability to enable "every application on every device all at once."

In this talk, I will discuss how the unified insight of multi-granular algorithm-hardware synergy, combined with the development of research infrastructure, can help alleviate the aforementioned challenges of efficiency, accessibility, and adaptability. First, I will introduce Instant-3D, which is designed to tackle the efficiency challenge. Instant-3D is a hardware-algorithm co-design that optimizes both memory usage and access regularity for bottleneck operators, enabling instant on-device 3D reconstruction. Next, I will present MixRT, which addresses the accessibility challenge. MixRT exploits heterogeneity across different operators to fully utilize commonly available hardware resources on modern GPUs, enabling real-time rendering across edge devices, from mobile phones to laptops. Then, I will introduce Uni-Render, which is designed to tackle the adaptability challenge. Uni-Render is a unified neural rendering accelerator that dynamically adjusts dataflows to align with specific rendering metric requirements, achieving real-time rendering speed across five different models using a single accelerator consuming approximately five watts. Finally, after briefly discussing my contributions to building the corresponding research infrastructure, I will conclude with my future research vision for ubiquitous 3D intelligence and explore how these research innovations and infrastructure can be extended beyond 3D intelligence to advance more efficient, accessible, and adaptable AI.

Biography:

Chaojian Li is a final-year Ph.D. candidate at the Georgia Institute of Technology, advised by Professor Yingyan (Celine) Lin. His research lies at the intersection of deep learning, computer architecture, and 3D vision, where he co-designs systems, architectures, and algorithms to advance neural rendering, 3D reconstruction, and related areas (e.g., healthcare, AI for chemistry, and video understanding), driving innovations in 3D intelligence applications. He has received various recognitions, including the Best Paper Award (as Algorithm Lead) at MICRO 2024, recognition as an ML&Systems Rising Star by MLCommons in 2024, and 1st Place in the Ph.D. Forum at DAC 2024. Additionally, he led a team to win 1st place among 150 teams in the TinyML Contest at ICCAD 2022.