Abstract
While supervised learning has enabled great progress in many applications,unsupervised learning has not seen such widespread adoption, and remains animportant and challenging endeavor for artificial intelligence. In this work,we propose a universal unsupervised learning approach to extract usefulrepresentations from high-dimensional data, which we call ContrastivePredictive Coding. The key insight of our model is to learn suchrepresentations by predicting the future in latent space by using powerfulautoregressive models. We use a probabilistic contrastive loss which inducesthe latent space to capture information that is maximally useful to predictfuture samples. It also makes the model tractable by using negative sampling.While most prior work has focused on evaluating representations for aparticular modality, we demonstrate that our approach is able to learn usefulrepresentations achieving strong performance on four distinct domains: speech,images, text and reinforcement learning in 3D environments.