Stochastic Games with Neural Perception Mechanisms

Explore a lecture on strategic reasoning in multi-agent coordination within complex environments, focusing on partially-observable stochastic games (POSGs) with neural perception mechanisms. Delve into the development of a model class and algorithms for one-sided POSGs that utilize neural networks to observe continuous environments. Examine how ReLU neural-network classifiers create finite decompositions of continuous environments into polyhedral for each classification label. Learn about the derivation of an abstract piecewise constant representation of value functions and a point-based solution method that computes lower and upper bounds on the value function to determine (approximately) optimal strategies. Gain insights into the progress made in addressing the challenges of computing or approximating optimal values and strategies in real-world settings involving multiple agents, uncertainty, and partial information.

Stochastic games with neural perception mechanisms