Yuan Tang Research Statement Motivation and Overview

My research interest is the performance engineering, an interdisciplinary research area at the intersection of parallel and concurrent computing, programming languages, systems and theory. The goal of my research is to obtain scalable and portable performance for real-world applications and algorithms on modern computing systems by exploiting caching and parallelism. The applications and algorithms I have been working on include, but not limited to, dense linear algebra, and dynamic-programming computations, which are widely used in computational biology and image processing, and would like to work on regular and irregular algorithms and data structures such as those used in artificial intelligence, data mining and machine learning. My approach is to design and implement high-performance programming systems based on solid theoretical foundations. My work ranges from traditional high-performance computing [YZ08; YCT07; YFD06; Yua+03; Yua+05; YS05], programming models [Bac+16; DSY16; Yua+15; Yua+14], domain-specific compilation [Bac+16; Yua+11a; Yua+11b], and provably efficient runtime schedulers [Cho+16; DSY16], to general algorithms and data structures [Mic+14; Lui+14]. My recent work on “cache-oblivious wavefront algorithms” and the “Nested Dataflow Model” [Yua+15; Yua+14] provides the possibility to achieve optimal time bounds within the optimal space and cache bounds processor-obliviously (i.e. optimality is achieved independently of processor count) without tuning on shared-memory parallel computing systems. My research suggests that people may have to re-think the way of parallel programming. I plan to extend my research to distributed-memory systems so that wider areas of applications such as traditional highperformance computing, cloud computing, big data, and the Internet of Things can also benefit from the research. I plan to incorporate my research into a strict mathematical model, possibly unified with polyhedral model for algorithm manipulation and derivation, as well as theoretical analyses, reasoning and automatic theorem / property proof. Another direction of future research will be developing a declarative domain-specific language so that my research can be easily used by the community.