CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems
This provides interpretability for DRL systems in domains like computer systems, addressing a known bottleneck but with incremental improvements over prior explainability techniques.
The authors tackled the problem of explaining Deep Reinforcement Learning (DRL) controllers in input-driven environments by introducing CrystalBox, a model-agnostic framework that generates future-based explanations, demonstrating high-fidelity results in applications like adaptive bitrate streaming and congestion control.
We present CrystalBox, a novel, model-agnostic, posthoc explainability framework for Deep Reinforcement Learning (DRL) controllers in the large family of input-driven environments which includes computer systems. We combine the natural decomposability of reward functions in input-driven environments with the explanatory power of decomposed returns. We propose an efficient algorithm to generate future-based explanations across both discrete and continuous control environments. Using applications such as adaptive bitrate streaming and congestion control, we demonstrate CrystalBox's capability to generate high-fidelity explanations. We further illustrate its higher utility across three practical use cases: contrastive explanations, network observability, and guided reward design, as opposed to prior explainability techniques that identify salient features.