Optimal Action
Optimal action selection, aiming to maximize cumulative rewards or minimize regret in dynamic environments, is a central problem across diverse fields like robotics, control systems, and economics. Current research focuses on developing algorithms that handle non-stationarity, uncertainty, and high-dimensional action spaces, employing techniques like model predictive control, reinforcement learning (with architectures such as transformers and neural networks), and optimal transport. These advancements are improving decision-making in complex systems, with applications ranging from autonomous driving and energy management to scientific discovery and multi-agent coordination.
Papers
CEM-GD: Cross-Entropy Method with Gradient Descent Planner for Model-Based Reinforcement Learning
Kevin Huang, Sahin Lale, Ugo Rosolia, Yuanyuan Shi, Anima Anandkumar
Scientific Discovery and the Cost of Measurement -- Balancing Information and Cost in Reinforcement Learning
Colin Bellinger, Andriy Drozdyuk, Mark Crowley, Isaac Tamblyn