Applying Classification Learning on Linear System Solver Selection


Chih-Wei Chang

10:30:00 - 12:00:00

R430 , Astronomy and Mathematics Building

Using modern linear system solvers usually involve multiple parameter settings and algorithm choices. We will discuss how we can apply machine learning techniques to tackle this problem. After a brief introduction to machine learning (ML), we will first model the linear system solver selection problem into a typical multi-class classification problem in ML. Second, we will talk about the features extraction and the ML algorithm we use. Finally, we will demonstrate some empirical experimental results.