Department of Computer Science
The University of Texas at Austin
yishanlu AT utexas DOT edu
I am interested in making parallel computing more accessible to programmers in different application fields so that more people can enjoy the performance gain from parallel computing. This involves programming languages, compilers, computer architectures and understanding of applications.
I work with Prof. Keshav Pingali in Intelligent Software Systems research group.
I would like to come up with programming models suitable for constrained optimization/constraint satisfaction algorithms. I am exploring such models by using Galois framework to parallelize graph algorithms from the following application fields.
Graph analytics: Graph analytics are important to discover patterns in graph data. For example, k-truss decomposition can cluster nodes in a graph based on their shared links. Since graph analytics can be realized with various combinations of programming models and implementations, the tradeoff between programming productivity and implementation efficiency needs to be addressed.
Electronic Design Automation (EDA): Hardware designers use EDA tools to synthesize circuit descriptions to final layouts for manufacturing. During the synthesis process, functionality should remain the same; timing constraints should be respected; and area, power consumption should be optimized. Since EDA algorithms work on circuits, naturally represented as graphs, and the parallelism is hard to know statically, how to parallelize EDA algorithms becomes challenging.
Intrusion detection: As our daily lives depend heavily on networks, safety for on-line activities is essential now. Traditional signature-based intrusion detection is insufficient, since nowadays advanced attacks tend to leverage social engineering; can lurk for a long time before doing harm; and can be decomposed into seemingly harmless steps. Learning and detecting such advanced intrusion in real-time need to combine information security, machine learning on graph data and high-performance computing.
Chad Voegele, Yi-Shan Lu, Sreepathi Pai, Keshav Pingali. Parallel Triangle Counting and k-Truss Identification using Graph-centric Methods. In IEEE/DARPA/Amazon Graph Challenge 2017 at IEEE High Performance Extreme Computing Conference (HPEC), Waltham, MA, September 12-14, 2017. (Graph Challenge Champion) [pdf][slides]