Reinforcement Learning and Function Approximation.
Marina Irodova, Robert H. Sloan · 2005
Relational reinforcement learning combines traditional rein-forcement learning with a strong emphasis on a relational (rather than attribute-value) representation. Earlier work used relational reinforcement learning on a learning version of the classic Blocks World planning problem (a version where the learner does not know what the result of taking an action will be). “Structural ” learning results have been obtained, such as learning in a mixed 3–5 block environment and being able to perform in a 3 or 10 block environment. Here we instead take a function approximation approach to reinforcement learning for this same problem. We obtain similar learning accuracies, with much better running times, allowing us to consider much larger problem sizes. For in-stance, we can train on a mix of 3–7 blocks and then per-form well on worlds with 100–800 blocks—using less run-ning time than the relational method required for 3–10 blocks.