Action and Planning in AI: Learning, Models, and Algorithms

Course type

Lecture

Study programs

• Master Computer Science
• Master Data Science
• Master Software Systems Engineering

Content

• Models and Solvers in AI
• State models and heuristic search
• Logic, SAT solving, and ASP solving
• Classical planning: language, model, basic algorithms (heuristic search and SAT)
• Markov Decision Processes (MDPs): basic models and algorithms (VI, PI, RTDP, MCTS)
• Deep learning (DL) as model and solver; supervised learning, Approx VI, PI, MPI with DL
• Reinforcement Learning (RL): Value-based and policy gradient methods
• Current research in planning and RL

Objectives

Knowledge: on completion, the students have acquired detailed knowledge on

• models and algorithms for planning,
• models and algorithms for reinforcement learning, and
• models and algorithms for reinforcement learning that use deep neural networks.

Skills: on completion, the students know

• how these various algorithms work
• when they can be applied, and
• how they can be applied

Competences: on completion, the students are able to

• model and solve sequential decision problems,
• formulate and solve sequential decision problems by trial and error when model is not known
• formulate and solve sequential decision problems over large state states using deep nets.

Recommended prior knowledge

Basic knowledge of probability theory and logic, basic AI and ML course.

Recommended reading

• S. Russell and P. Norvig. AI: A Modern approach, 4th edition, 2021
• R. Sutton and A. Barto. Reinforcement learning: An introduction. 2nd Edition, 2018
• H. Geffner, B. Bonet. A Concise Introduction to Models and Methods for Automated Planning. 2013