[R] [1906.04493] Unsupervised Minimax: Adversarial Curiosity, Generative Adversarial Networks, and Predictability Minimization (Schmidhuber)[reddit]https://arxiv.org/abs/1906.04493/r/MachineLearningPM uses gradient-based minimax or adversarial training to learn an encoder of the data, such that the codes are distributed like the data, and the probability of a given pattern can be read off its code as the product of the predictor-modeled probabilities of the code components (Sec

Abstract: Generative Adversarial Networks (GANs) learn to model data distributions
through two unsupervised neural networks, each minimizing the objective
function maximized by the other. We relate this game theoretic strategy to
earlier neural networks playing unsupervised minimax games. (i) GANs can be
formulated as a special case of Adversarial Curiosity (1990) based on a minimax
duel between two networks, one generating data through its probabilistic
actions, the other predicting consequences thereof. (ii) We correct a
previously published claim that Predictability Minimization (PM, 1990s) is not
based on a minimax game. PM models data distributions through a neural encoder
that maximizes the objective function minimized by a neural predictor of the
code components.