Decision-Theoretic and Multi-Agent Planning

Professors:Ronen I. Brafman
Credits:3
Range:B
Note:

Full description

Title: Decision-Theoretic and Multi-Agent Planning

Lecturer: Prof. Ronen I. Brafman, Ben-Gurion University, Beer-Sheva, Israel

Schedule:

day 1: September 26
day 2: September 27
day 3: September 28
day 4: October 5
day 5: October 6
day 6: October 10
day 7: October 13

each day lectures will be in Room B203, from 5 p.m. to 7:30 p.m.  

Program:

Lecture 1: Markov Decision Processes (MDPs) — model, basic solution concepts, and the three basic solution algorithms: value iteration, policy iteration, linear programming!

Lecture 2: Advanced Methods for MDP solution: prioritized sweeping, backwards VI, search-based method such as AO* and LAO*, RTDP, etc.

Lecture 3: Partially observable Markov Decision Processes (POMDPs) — model, belief states, belief state update, transformation to belief-space MDP, structure of the value function

Lecture 4: Value iteration for POMDPs and basic pruning methods

Lecture 5: Point-based Value iteration for POMDPs

Lecture 6: Sparse sampling and Monte-Carlo tree search for MDPs

Lecture 7: Sparse sampling and Monte-Carlo tree search for POMDPs

Lecture 8: Belief state representation and update using Particle filters

Lecture 9: Classical Multi-Agent planning: model, private and public aspects, complexity

Lecture 10: Classical Multi-Agent planning: distributed forward search, privacy, and uncertainty