Diagnosing Bottlenecks in Deep Q-learning Algorithms
This work addresses understanding and improving deep Q-learning methods for reinforcement learning practitioners, though it appears incremental in nature.
The researchers investigated performance bottlenecks in deep Q-learning algorithms by creating a unit testing framework with oracles to isolate error sources, finding that large neural network architectures improve learning stability and developing a novel sampling method that yields fair improvement on high-dimensional continuous control domains.
Q-learning methods represent a commonly used class of algorithms in reinforcement learning: they are generally efficient and simple, and can be combined readily with function approximators for deep reinforcement learning (RL). However, the behavior of Q-learning methods with function approximation is poorly understood, both theoretically and empirically. In this work, we aim to experimentally investigate potential issues in Q-learning, by means of a "unit testing" framework where we can utilize oracles to disentangle sources of error. Specifically, we investigate questions related to function approximation, sampling error and nonstationarity, and where available, verify if trends found in oracle settings hold true with modern deep RL methods. We find that large neural network architectures have many benefits with regards to learning stability; offer several practical compensations for overfitting; and develop a novel sampling method based on explicitly compensating for function approximation error that yields fair improvement on high-dimensional continuous control domains.