Speaker
Title
Combining Game Theory and Machine Learning to Refine Interaction Strategies.
Abstract
Games and game theory have been foundational to AI from its early days.
Classical algorithms typically search through the space of possible moves, actions or policies in order to analyze potential courses of action and opponent responses.
The recent shift towards machine learning uncovered methods, such as deep reinforcement learning leading to superhuman performance in many settings.
Using examples from various games I will explore what such techniques offer, and the trade-offs between task performance, exploitability and explainability and transparency.
Finally, I will show how a combination of game theory and machine learning can offer a good foundation for cooperative AI systems.