Report Title: Verified NN4Sys: The What, Why, and How

Report time: June 27th, 10:00-11:00

Offline report location: B1002, Science Building

Host: Professor Zhang Min

Report Summary:

Neural networks are powerful tools. The application of them to computer systems (operating systems, databases, and network systems) has attracted widespread attention. However, neural networks are complex black boxes that may produce unexpected results. Our vision is to build neural networks (NN4Sys) suitable for computer systems that meet predefined correctness attributes. We refer to these validated NN4Sys as verified NN4Sys. In this lecture, I will introduce our recent attempts to pursue this vision, including building the NN4Sys benchmark, training validated NN4Sys, and applying NN4Sys to multiple systems.

Reported by:

Cheng Tan is an assistant professor at the Curry School of Computer Science at Northeastern University in the United States. His research interests include computer systems, verifiable systems, and neural networks within systems. He has won the SOSP'17 Best Paper Award Janet Fabri Doctoral Dissertation Award and NSF CAREER Award.