Towards Network-Accelerator Co-Search for Promoting Ubiquitous on-Device Intelligence and Green AI
Add to Google Calendar
Abstract: Deep learning (DL)-powered intelligence embedded into numerous daily-life devices promises to transform the quality of human life. Despite this great promise, there is a vast and increasing gap between the prohibitive complexity of powerful DL algorithms and the constrained resources in daily-life devices. While DL accelerators have the potential to close the aforementioned immense gap and push forward green AI, their power has yet to be unleashed due to the following fundamental challenges: (1) fast DL algorithm advances vs. slow DL accelerator development, and (2) the promise of algorithm and accelerator co-search vs. the lack of such co-search. Therefore, it is imperative to develop innovative techniques that can expedite the development of optimal DL accelerators and unlock the promise of co-searching for optimal DL algorithms and accelerators for maximizing their achievable hardware efficiency.
In this talk, I will present our recently developed techniques towards DL network-accelerator co-search, serving as a timely holistic effort toward addressing the aforementioned challenges. Specifically, I will start by introducing our techniques for designing hardware-aware DL algorithms (i.e., top-down efforts) and algorithm-aware DL accelerators (i.e., bottom-up efforts), which help us to gain important insights for understanding their design space and optimization. Then, I will share our first-of-their-kind techniques that are among the very first generic efforts to enable simultaneous searching for optimal DL algorithms and accelerators (i.e., bridging efforts) to maximize both task accuracy and hardware efficiency. Finally, I will conclude my talk with exciting (1) applications of our co-search framework and (2) pointers to future directions.
Bio: Yingyan (Celine) Lin is an Assistant Professor in the Department of Electrical and Computer Engineering at Rice University. She leads the Efficient and Intelligent Computing (EIC) Lab at Rice, which focuses on developing efficient machine learning techniques towards green AI and ubiquitous machine learning powered intelligence. She received a Ph.D. degree in Electrical and Computer Engineering from the University of Illinois at Urbana-Champaign in 2017.
Prof. Lin is a NSF CAREER Award, IBM Faculty Award, and Facebook Research Award recipient, and recently received the ACM SIGDA Outstanding Young Faculty Award. She was selected as a Rising Star in EECS by the 2017 Academic Career Workshop for Women at Stanford University. She received a Best Student Paper Award at the 2016 IEEE International Workshop on Signal Processing Systems (SiPS 2016), and the 2016 Robert T. Chien Memorial Award for Excellence in Research at UIUC. Prof. Lin is currently the lead PI on multiple multi-university projects (e.g., RTML and 3DML) and her group has been funded by NSF, NIH, DARPA, ONR, Qualcomm, Intel, HP, IBM, and Facebook. She has recently been nominated for the Dean’s Teaching + Research Excellence Award at Rice.