English

CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems

Machine Learning 2024-03-28 v4 Networking and Internet Architecture Systems and Control Systems and Control

Abstract

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.

Keywords

Cite

@article{arxiv.2302.13483,
  title  = {CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems},
  author = {Sagar Patel and Sangeetha Abdu Jyothi and Nina Narodytska},
  journal= {arXiv preprint arXiv:2302.13483},
  year   = {2024}
}
R2 v1 2026-06-28T08:50:05.975Z