On the Value Functions in Reinforcement Learning for Ridesharing
With the quickly growing volume and variety of transportation data, machine learning (ML) and artificial intelligence (AI) technologies present a great potential to arouse improved scientific understandings, transformative informed decisions, and innovative and proactive management solutions for future transportation systems. Meanwhile, transportation systems are complex and dynamic physical systems. They are often associated with very complicated and unique operation problems featuring nonconvexity, nonlinearity, high dimensionality, and prohibitive computational difficulties. The power of AI technology can only be fully unlocked when we well customize it with transportation domain knowledge. This presentation shares several of our research studies, which integrate ML/AI approaches with traffic flow analysis, optimization, game theories, and control technologies to address challenging issues in real-time traffic prediction, event detection, large-scale traffic management and control, involving emerging transportation technologies such as shared mobility and connected and autonomous vehicles.