Accelerating Irregular Data Movement in AI Computing via Algorithm–Hardware Co-Design

Speaker: Dr. Chaojian LI
Assistant Professor
Department of Computer Science and Engineering
Hong Kong University of Science and Technology

Title: Accelerating Irregular Data Movement in AI Computing via Algorithm–Hardware Co-Design

Date: Monday, 27 October 2025

Time: 4:00pm - 5:00pm

Venue: Lecture Theater F (Leung Yat Sing Lecture Theater, near lift 25/26), HKUST

Abstract:

Modern AI workloads, from 3D neural reconstruction and rendering to large language model (LLM) reasoning, are increasingly dominated by irregular data movement rather than arithmetic computation. Such irregularity arises from unstructured data dependencies, sparse memory access patterns, and dynamic control flows, all of which limit the efficiency of current hardware accelerators. Addressing these challenges requires a cross-layer algorithm–hardware co-design that jointly optimizes data representation, scheduling, and memory hierarchy. This talk presents recent advances in bridging algorithmic irregularity and hardware efficiency, using representative examples from 3D vision and LLM hardware accelerators to illustrate the design principles for efficient AI computing.

Biography:

Chaojian Li is an Assistant Professor in the Department of Computer Science and Engineering (CSE) at the Hong Kong University of Science and Technology (HKUST). Prior to joining HKUST, he received his Ph.D. in Computer Science from the Georgia Institute of Technology in 2025, advised by Prof. Yingyan (Celine) Lin. His research focuses on algorithm-hardware-infrastructure co-design for developing efficient, adaptable, and scalable AI systems, with applications in large language models, 3D intelligence, and AI-driven hardware design. His contributions have been recognized with several prestigious honors, including the Best Paper Award at MICRO 2024, selection as an MLCommons ML & Systems Rising Star, 1st Place in the DAC 2024 Ph.D. Forum, and 1st Place in the ICCAD 2022 TinyML Design Contest.