Game Playing 1 - Minimax, Alpha-beta Pruning | Stanford CS221: AI (Autumn 2019)

For more information about Stanford’s Artificial Intelligence professional and graduate programs, visit: stanford.io/3Cke8v4
Topics: Minimax, expectimax, Evaluation functions, Alpha-beta pruning
Percy Liang, Associate Professor & Dorsa Sadigh, Assistant Professor - Stanford University
onlinehub.stanford.edu/
Associate Professor Percy Liang
Associate Professor of Computer Science and Statistics (courtesy)
profiles.stanford.edu/percy-l...
Assistant Professor Dorsa Sadigh
Assistant Professor in the Computer Science Department & Electrical Engineering Department
profiles.stanford.edu/dorsa-s...
To follow along with the course schedule and syllabus, visit:
stanford-cs221.github.io/autu...
0:00 Introduction
0:43 Course plan
2:09 A simple game
3:29 Roadmap
4:01 Game tree
5:05 Two-player zero-sum games
8:55 Example: chess
11:43 Characteristics of games
22:33 Game evaluation example
29:01 Expectimax example
33:51 Extracting minimax policies
34:21 The halving game
38:44 Face off
45:41 Minimax property 2
48:18 Minimax property 3
53:02 A modified game
53:49 Expectiminimax example
55:26 Expectiminimax recurrence
57:19 Computation

[@suchalooser1175]

Really good lecture series on reinforcement learning, good balance of math, theory, and actual implementation details!!!

[@suchalooser1175]

Not sure why this is having less view count, lectures are high quality and detailed.

[@saicharanmarrivada5077]

​@@parmoksha Reinforcement learning also quite popular bro

[@black-sci]

Nice Lecture.

[@leventaksakal5]

these algorithms looks cool in theory

[@AkshitSharma0]

47:09 Why is V(pimax,pi7)=2 and not 5, assuming agent will try to maximize his value while the opponent will act stochastically (ie. 0,2,5 as distributions)

[@paladin1410]

Hi, I believe the agent try to maximize his value with the assumption that the opponent is a minimizer. It is like you do not know what your opponent next move but you will imagine your opponent is a minimizer and calculate the value for your opponent under that assumption. In that scenario, if my policy is pi_max, I always choose the second branch.