Iterative Policy-Space Expansion in Reinforcement Learning
This work addresses the problem of slow learning rates in reinforcement learning for agents, presenting an incremental improvement by adapting curriculum learning internally.
The paper tackles the challenge of improving learning efficiency in reinforcement learning by introducing an algorithm that gradually expands the policy space from simple to complex, rather than using an external curriculum. Experimental results in Tetris show a superior learning rate compared to existing algorithms, though specific numerical gains are not provided.
Humans and animals solve a difficult problem much more easily when they are presented with a sequence of problems that starts simple and slowly increases in difficulty. We explore this idea in the context of reinforcement learning. Rather than providing the agent with an externally provided curriculum of progressively more difficult tasks, the agent solves a single task utilizing a decreasingly constrained policy space. The algorithm we propose first learns to categorize features into positive and negative before gradually learning a more refined policy. Experimental results in Tetris demonstrate superior learning rate of our approach when compared to existing algorithms.