[05-26]Verification-in-the-loop: A Safe Reinforcement Learning Framework

文章来源:  |  发布时间:2023-05-17  |  【打印】 【关闭

Title:   Verification-in-the-loop: A Safe Reinforcement Learning Framework
Speaker: 黄超 (University of Liverpool,Assistant Professor)
Time:  5月26日(周五),10:00-12:00
Venue: 线上:腾讯会议 831-251-884
Abstract:   Reinforcement learning (RL) is one of the main areas in machine learning and is ideal for decision making in complex environment. In a typical RL framework, an agent will interact with the environment and use the feedback to update its control policy. Compared with the classical control algorithms, RL can be used in a complex or even unknown environment. However, due to the insufficient consideration of safety, existing RL algorithms suffer from the safety issue and thus can hardly be deployed in safety-critical scenarios. In this presentation, I will introduce our high-level safe RL framework by involving verification in the learning loop, and our latest work for unknown environment. The empirical results show that involving verification in the loop can significantly outperforms existing techniques in terms of safety.
Bio:  Dr. Huang is current a lecturer (assistant professor) in Computer Science at the University of Liverpool. He is also an adjunct assistant professor at Northwestern University. Prior to Liverpool, he was a postdoc fellow at Northwestern University. He received the Bachelor degree and the PhD degree at Nanjing University. His research interests include design and verification of machine-learning-based autonomous systems.