The paper integrates the theory of reinforcement learning with work on cognition and neuroscience. It pays special attention to the successor representation (SR) and its generalizations, which have been widely applied as engineering tools and models of brain function. The convergence of these areas suggests that certain types of predictive representations may serve as versatile building blocks of intelligence. Practical learning algorithms and their challenges are discussed, as well as the application of these theories in artificial intelligence.
Publication date: 12 Feb 2024
Project Page: https://arxiv.org/abs/2402.06590
Paper: https://arxiv.org/pdf/2402.06590
