Professor Xiaowen Chu

About

Professor Chu Xiaowen  has focused on improving the performance of deep learning systems, which are the key techniques behind many modern artificial intelligence (AI) applications. In 2016, his research team developed one of the earliest open-source benchmarking suites named DLBench that can evaluate, trace and compare the performance of different deep learning platforms. This pioneering work attracted significant attention from the deep learning community, including Microsoft’s CNTK team, MXNet team, Nvidia, Intel, Tencent, and Inspur. Similar benchmarks have been launched by academia (such as Stanford’s DAWNBench in 2017) and industry (MLPerf in 2018). Since 2018, Professor Chu’s team has proposed many novel methods to reduce the training time of AI models on graphics processing unit (GPU) clusters. 

As a pioneer in the field of GPU computing, Professor Chu’s research group has designed and implemented many GPU-based parallel algorithms and software to significantly reduce the running time of real-world problems, including DNA sequence alignment, protein identification, data mining, data security, linear algebra routines and deep learning. His team has not only published many highly cited papers, but also developed several open-source software. For example, the work of G-BLASTN was not only published in the top-tier journal Bioinformatics, but also received more than 1,000 downloads from Sourceforge and Github. His recent work, “Dissecting GPU Memory Hierarchy through Microbenchmarking”, has proposed a novel fine-grained P-Chase algorithm to expose many previously unknown properties of GPU memory systems. It is the first paper to successfully analyse the cache properties of Keper and Maxwell GPUs, and the proposed method has been widely used by the research community to (1) analyse the memory system of recent GPUs (such as Volta and Turing); (2) accurately predict the performance of GPU applications such as parallel convolution operations in deep learning; and (3) design cache-friendly optimisation techniques for matrix operations and convolutions. 

Professor Chu has published more than 170 research papers in the field of parallel and distributed computing, deep learning systems, cloud computing, wireless networks, etc. Many of his works have been published in prestigious international journals, including IEEE Transactions on Parallel and Distributed Systems, IEEE Journal on Selected Areas in Communications, IEEE Transactions on Mobile Computing, IEEE/ACM Transactions on Networking, IEEE Transactions on Smart Grid, IEEE Transactions on Computers and Bioinformatics, and also top-ranked international conference proceedings including IEEE INFOCOM, ACM PPoPP, IEEE ICDCS, IEEE IPDPS, IJCAI, IEEE ICRA and ACM e-Energy. Professor Chu is dedicated to impactful research that can influence not only the research community but also industry practitioners. He has received many external research grants and donations from the Research Grants Council, the Innovation and Technology Fund and industry, totalling more than HK$18 million.

Achievements

Research Outputs