Universal learning theory

Abstract

Universal (machine) learning is concerned with the development and study of algorithms that are able to learn from data in a very large range of environments with as few assumptions as possible. The class of environments typically considered includes all computable stochastic processes. The investigated learning tasks range from inductive inference, sequence prediction, sequential decisions, to (re)active problems like reinforcement learning (Hutter, 2005), but also include clustering, regression, and others (Li & Vitányi, 2008).