Generative Adversarial Networks are Special Cases of Artificial Curiosity (1990) and also Closely Related to Predictability Minimization (1991)

I review unsupervised or self-supervised neural networks playing minimax\ngames in game-theoretic settings: (i) Artificial Curiosity (AC, 1990) is based\non two such networks. One network learns to generate a probability distribution\nover outputs, the other learns to predict effects of the outputs. Each network\nminimizes the objective function maximized by the other. (ii) Generative\nAdversarial Networks (GANs, 2010-2014) are an application of AC where the\neffect of an output is 1 if the output is in a given set, and 0 otherwise.\n(iii) Predictability Minimization (PM, 1990s) models data distributions through\na neural encoder that maximizes the objective function minimized by a neural\npredictor of the code components. I correct a previously published claim that\nPM is not based on a minimax game.\n