Information flow in learning a coin-tossing game

Abstract

Information flow in adaptively interacting stochastic processes is studied. We give an extended form of game dynamics for interacting Markovian processes and compute a measure of causal information flow, which is different from the transfer entropy. In the game theoretic situation, causal information flow can show oscillatory behavior through reward-maximizing adaptation of two players. The adaptive dynamics for the coin-tossing game is exemplified and the causal information flow therein is investigated.