I'm a Ph.D. student at the Department of Computer Science and Engineering, The Chinese University of Hong Kong (CUHK), with Prof. Bei Yu. My research interests lie in the general area of computer architecture, compilers, and systems with a focus on the system-level and programming challenges of machine learning systems.
[Aug 2020 - Now] The Chinese University of Hong Kong
Ph.D, Major in Computer Science.
[Sep 2017 - Jun 2020] Chinese Academy of Sciences
M.Eng, Major in Computer Science.
[Sep 2013 - Jun 2017] Xidian University
B.Eng, Major in Telecommunication Engineering.
[Jun 2019 - Oct 2019] Computer Systems Lab, Cornell Univeristy
Work with Prof. Zhiru Zhang.
Do research in Multi-Paradigm Programming Infrastructure for Software-Defined Heterogeneous Computing.
AI Lab - ML Systems, ByteDance Beijing
: Full Stack Deep Learning Compiler [slides].
Computer Systems Lab - Cornell Univeristy, Ithaca
: Summary about OpenCL [slides].
ACPNet: Anchor-Center Based Person Network for Human Pose Estimation and Instance Segmentation
Yang Bai, Weiqiang Wang, IEEE International Conference on Multimedia and Expo (ICME) 2019 [paper]
We present an effective approach to tackle the multi-person pose estimation and person instance segmentation jointly. The approach based on Mask R-CNN uses a set of well-designed labels, called anchor-center based label, to learn keypoints localization in complex and crowded multi-person scenes.
OpenCL Backend Development for HeteroCL
I develop the Xilinx & Intel OpenCL backend for TVM-inspired HeteroCL and implement critical compiler backend optimization, e.g., loop unrolling, loop pipelining and partition for Xilinx OpenCL backend and implemented arbitrary precision integers for Intel OpenCL backend. I Implement the whole pipeline from Python-based domain-speciﬁc language to FPGA-targeted compilation ﬂow
AWS-F1 Tutorial for KNN-DigitRec:
I write tutorials for running KNN-DigitRec example on AWS-F1 using HeteroCL. I design a new target backend for AWS development, combined HeteroCL Vivado HLS C++ code and host ﬁle based on Rosetta automatically generate host and wrapper ﬁles for design automation.
Anchor-Center Based Multi-Person 2D Pose Estimation
I develop a top-down multi-person 2D pose estimation network with the anchor-center based label for the keypoints learning.
[Fall 2020] CMSC 5743 Efficient Computing of Deep Neural Networks