Game Theory in Machine Learning, part 2 - Costantinos Daskalakis - MLSS 2020, Tübingen

0:00:00 Game Theory and Machine Learning
0:03:13 Future of AI: learning + strategic reasoning
0:04:20 A dictum
0:07:06 Our focus: min-max optimization
0:11:11 Training Oscillations of GDA (and its variance)
0:12:35 Negative Momentum OGDA and EO
0:14:49 Part 1: convex-concave objectives
0:16:05 Negative Momentum, in the Wild
0:17:48 Optimistic Adam, on CIFAR10
0:22:57 Decreasing Momentum Trend
0:25:12 Part 2: nonconvex-nonconcave objectives
0:50:56 Computational Complexity
0:52:01 The Complexity of GDA Fixed Points
0:57:34 The PPAD Complexity Class
1:00:48 Sperner's Lemma
1:15:47 The Sperner problem (precisely)
1:17:02 Solving SPERNER
1:19:00 The Complexity of GDA Fixed Points
1:22:02 Rough Proof Idea: Edge Triangle Game
1:25:17 Reinforcement Learning: Single vs. Multiple
1:29:17 Reinforcement Learning: Algorithms
1:30:34 2-Player Zero-Sum RL
1:32:14 Conclusions
1:33:41 Thank You