Speaker: Xiangyu Zhang
Time: 2015-06-26 15:00
Place: Room 632, EE-3 Building, West Campus
Errors pose a serious threat to output validity for modern data processing, which is often performed by programs. Raw inputs are acquired by physical instruments that have precision limitations, leading to input errors. Parameters in data processing may be provided by human experts based on their experience, leading to uncertainty. Data may not be precisely represented due to the limited precision of the machine used, leading to representation errors. Once these errors get into computation, they may get propagated and magnified, producing unreliable output. This is called the instability problem. Instability problems have substantial impact in many areas such as scientific data processing, financial decision making, and machine learning. In this talk, I will present our recent findings in addressing instability problems through runtime techniques.
Xiangyu Zhang is an associate professor at Purdue University. He received his PhD degree in the University of Arizona in 2006, and his MS and BS degrees from USTC. His research interest lies in dynamic and static program analysis and their applications in debugging, forensic analysis, and data processing. He is currently the Purdue University Scholar. He has received the 2006 ACM SIGPLAN Distinguished Doctoral Dissertation Award, NSF Career Award, and a few best paper awards in top conferences.