Modularity in Reinforcement Learning via Algorithmic Independence in Credit Assignment
This work addresses the need for modular learning algorithms in reinforcement learning to improve transfer efficiency, representing an incremental advance by formalizing and testing modularity in credit assignment.
The paper tackles the problem of modular credit assignment in reinforcement learning by defining it as minimizing algorithmic mutual information among feedback signals and introducing a modularity criterion for testing algorithms. They prove that certain temporal difference action-value methods satisfy this criterion and are more sample efficient than policy-gradient methods on transfer tasks requiring sparse changes, with empirical evidence supporting this.
Many transfer problems require re-using previously optimal decisions for solving new tasks, which suggests the need for learning algorithms that can modify the mechanisms for choosing certain actions independently of those for choosing others. However, there is currently no formalism nor theory for how to achieve this kind of modular credit assignment. To answer this question, we define modular credit assignment as a constraint on minimizing the algorithmic mutual information among feedback signals for different decisions. We introduce what we call the modularity criterion for testing whether a learning algorithm satisfies this constraint by performing causal analysis on the algorithm itself. We generalize the recently proposed societal decision-making framework as a more granular formalism than the Markov decision process to prove that for decision sequences that do not contain cycles, certain single-step temporal difference action-value methods meet this criterion while all policy-gradient methods do not. Empirical evidence suggests that such action-value methods are more sample efficient than policy-gradient methods on transfer problems that require only sparse changes to a sequence of previously optimal decisions.