One question central to Reinforcement Learning is how to learn a feature representation that supports algorithm scaling and re-use of learned information from different tasks. Successor Features approach this problem by learning a feature representation that satisfies a temporal constraint. We present an implementation of an approach that decouples the feature representation from the reward fun...

We define a coding of natural numbers – which we will call exponential notations – and interpretations of the less-than-relation, the successor, addition and exponentiation function on exponential notations. We prove that all these interpretations are polynomial time computable. As a corollary we obtain that feasible arithmetic can prove the consistency of the canonical equational theory for th...

We give an exposition in modern language (and using partial orders) of Jech’s method for obtaining models where successor cardinals have large cardinal properties. In such models, the axiom of choice must necessarily fail. In particular, we show how, given any regular cardinal and a large cardinal of the requisite type above it, there is a symmetric extension of the universe in which the axiom ...

Transfer in reinforcement learning refers to the notion that generalization should occur not only within a task but also across tasks. We propose a transfer framework for the scenario where the reward function changes between tasks but the environment’s dynamics remain the same. Our approach rests on two key ideas: successor features, a value function representation that decouples the dynamics ...