Planning and Decision Making for Intelligent Agents. 3 credits

Catalog description

This course will introduce the definition of planning domains, including representations for world states and actions. We will cover optimal search-based planning strategies in domains with finite states as well as methods for effective planning and acting in the real world, where both sensing and actions are uncertain. We introduce the framework or Markov Decision Processes for decision making under uncertainty and introduce basic algorithms for finding optimal policies; mapping from states to actions. In the second part of the course we will focus on reinforcement learning problems, covering model-free and model-based reinforcement learning methods, temporal difference learning and policy gradient algorithms. The course will also cover algorithms used to solve multi-arm bandit problems, each with its own trade-offs between exploration and exploitation. Reinforcement learning is an essential part of fields ranging from modern robotics to game-playing (e.g. Poker, Go, and Starcraft).Offered by Engineering & Computing. May not be repeated for credit.