MIT News breed · Science

Cathy Wu's Research on Reinforcement Learning for Complex Systems

Cathy Wu, an associate professor at MIT, is developing computational tools, specifically reinforcement learning (RL), to tackle complex societal challenges, with a primary focus on improving transportation systems.

Wu's motivation stems from a desire to improve lives, influenced by her father's commute and childhood computer games. Her research aims to create safe and efficient transportation systems using machine learning and RL.

Traditional methods struggle with the numerous variants in transportation system design. RL offers a potential solution, enabling researchers and practitioners to design better systems.

Wu's work in applying AI to transportation began at MIT, inspired by a lecture on autonomous vehicles. She focused on RL and optimization methodologies.

A breakthrough in 2023 allowed Wu's team to overcome RL's sensitivity. Training RL on a subset of problems that generalize well enables effective performance on related problems, improving training efficiency by up to 30 times.

This advancement restored Wu's confidence in RL's potential for complex optimization problems, particularly in transportation. Her group now focuses on contextual RL.

Recent research shows eco-driving could reduce vehicle emissions by 11-22%, demonstrating RL's utility in informing transportation policy.

Wu advocates for use-inspired research, addressing practical problems to develop fundamental knowledge applicable to various systems. She advises students to be patient and curious.

AI-samenvatting op basis van de bron.