Computer Science Presentation: Accelerating Sparse and Irregular Computations for Machine Learning Applications, Dr. Hao Zheng, the Department of Electrical and Computer Engineering at the University of Central Florida

Abstract:

The use of Artificial Intelligence and Machine Learning (ML) is surging right now in many applications.  This surge is mainly driven by advancements in powerful computing hardware such as GPUs and data centers. ML’s continually growing model and dataset size call for unprecedented computing power. Specialized hardware architectures in the form of accelerators have emerged as the prevailing solution to meet such growing demands in the post-Moore era. However, the recent shift toward more complex ML models and non-Euclidean data is posing significant challenges to sustaining the performance scaling of current specialized architectures. This talk will cover our recent efforts in designing versatile and scalable architectures to address the major challenges in modern ML accelerator design from both architecture and algorithm perspectives. Specifically, I will first describe the necessity of enabling flexibility in dataflow and architecture for Graph Neural Network (GNN) applications. Next, I will present our recent work exploring sparse-dense matrix multiplication from a graph perspective, with significant improvements in performance, energy efficiency, and scalability. I will conclude this talk by describing my future research directions in this area. 

Bio:

Hao Zheng is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Central Florida. He received his Ph.D. from George Washington University in 2021. His research interests lie in the broad areas of computer architecture and machine learning, with an emphasis on hardware accelerators for machine learning, hardware/software co-design, AI-assisted chip design, and on/off-chip communication. His work has been published in top-tier venues in fields such as ISCA, MICRO, HPCA, DAC, and ICCAD, including a Best Paper Nomination at DAC 2020. He is a recipient of the National Science Foundation CAREER Award. He is serving as an Associate Editor for IEEE Transactions on Computers and IEEE Transactions on Sustainable Computing.