I am a PhD student at the School of Computing at the University of Leeds.
My PhD aims to accelerate deep learning workloads by optimising distributed communications and computation kernels. There are two strands of my research. The first is to optimise sparse matrix multiplication that dominates the execution time for most of the state-of-the-art deep neural networks (DNNs), including graph neural networks and Transformer-based models. The second theme of my PhD is to reduce the communication latency across GPUs and machines when training DNNs in a distributed computing environment. I am currently interested in Multi-Level Intermediate Representation (MLIR) project and would like to build an auto-scheduling accelerator above MLIR architecture. My other interests lie in Machine Learning/Deep Learning, Graph Neural Networks, Machine Learning Compilers and a little bit of explainable AI.