Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems
(Submitted on 27 Feb 2023 (v1), last revised 27 Mar 2024 (this version, v4))
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.
Submission history
From: Sagar Patel [view email][v1] Mon, 27 Feb 2023 02:42:27 GMT (2052kb,D)
[v2] Thu, 8 Jun 2023 02:20:09 GMT (2052kb,D)
[v3] Mon, 18 Dec 2023 12:50:14 GMT (2022kb,D)
[v4] Wed, 27 Mar 2024 17:38:27 GMT (2015kb,D)
Link back to: arXiv, form interface, contact.