The classification game: Complexity regularization through interaction

Abstract

We show that if a population of neural network agents is allowed to interact during learning, so as to arrive at a consensus solution to the learning problem, then they can implicitly achieve complexity regularization. We call this learning paradigm, the classification game. We characterize the game-theoretic equilibria of this system, and show how low-complexity equilibria get selected. The benefit of finding a low-complexity solution is better expected generalization. We demonstrate this benefit through experiments. © Springer-Verlag Berlin Heidelberg 2010.