Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2302.13483 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911816960966656 |
|---|---|
| author | Patel, Sagar Jyothi, Sangeetha Abdu Narodytska, Nina |
| author_facet | Patel, Sagar Jyothi, Sangeetha Abdu Narodytska, Nina |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2302_13483 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems Patel, Sagar Jyothi, Sangeetha Abdu Narodytska, Nina Machine Learning Networking and Internet Architecture Systems and 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. |
| title | CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems |
| topic | Machine Learning Networking and Internet Architecture Systems and Control |
| url | https://arxiv.org/abs/2302.13483 |