arXiv:2510.19244v2 Announce Type: replace 
Abstract: Deep reinforcement learning (RL) achieves remarkable performance but lacks interpretability, limiting trust in policy behavior. The existing SILVER framework (Li, Siddique, and Cao 2025) explains RL policy via Shapley-based regression but remains restricted to low-dimensional, binary-action domains. We propose SILVER with RL-guided labeling, an enhanced variant that extends SILVER to multi-action and high-dimensional environments by incorporating the RL policy's own action outputs into the boundary points identification. Our method first extracts compact feature representations from image observations, performs SHAP-based feature attribution, and then employs RL-guided labeling to generate behaviorally consistent boundary datasets. Surrogate models, such as decision trees and regression-based functions, are subsequently trained to interpret RL policy's decision structure. We evaluate the proposed framework on two Atari environments using three deep RL algorithms and conduct human-subject study to assess the clarity and trustworthiness of the derived interpretable policy. Results show that our approach maintains competitive task performance while substantially improving transparency and human understanding of agent behavior. This work advances explainable RL by transforming SILVER into a scalable and behavior-aware framework for interpreting deep RL agents in high-dimensional, multi-action settings.

نهج جديد في التعلم المعزز (RL) يسمى SILVER مع التسمية الموجهة بواسطة RL يثير ضجة في هذا المجال من خلال تحسين القابلية للتفسير في البيئات المعقدة. كانت الطرق التقليدية تكافح مع السيناريوهات متعددة الإجراءات وعالية الأبعاد، لكن هذا الإطار المبتكر يعد بسد هذه الفجوة. من خلال استخدام الانحدار القائم على شابلي، فإنه يسمح بفهم أفضل لسلوك السياسات، وهو أمر حاسم لبناء الثقة في أنظمة الذكاء الاصطناعي. لا تعزز هذه الخطوة أداء RL فحسب، بل تفتح أيضًا أبوابًا لتطبيقه في مواقف العالم الحقيقي الأكثر تعقيدًا.

Un nuevo enfoque para el aprendizaje por refuerzo (RL) llamado SILVER con etiquetado guiado por RL está causando revuelo en el campo al mejorar la interpretabilidad en entornos complejos. Los métodos tradicionales luchaban con escenarios de múltiples acciones y de alta dimensión, pero este marco innovador promete cerrar esa brecha. Al utilizar la regresión basada en Shapley, permite una mejor comprensión del comportamiento de las políticas, lo cual es crucial para construir confianza en los sistemas de IA. Este avance no solo mejora el rendimiento del RL, sino que también abre puertas para su aplicación en situaciones del mundo real más intrincadas.

Une nouvelle approche de l'apprentissage par renforcement (RL) appelée SILVER avec étiquetage guidé par RL fait sensation dans le domaine en améliorant l'interprétabilité dans des environnements complexes. Les méthodes traditionnelles avaient du mal avec des scénarios à actions multiples et de haute dimension, mais ce cadre innovant promet de combler cette lacune. En utilisant la régression basée sur Shapley, il permet une meilleure compréhension du comportement des politiques, ce qui est crucial pour instaurer la confiance dans les systèmes d'IA. Cette avancée non seulement améliore les performances du RL, mais ouvre également des portes à son application dans des situations réelles plus complexes.

A new approach to deep reinforcement learning (RL) called SILVER with RL-guided labeling is making waves in the field by enhancing interpretability in complex environments. Traditional methods struggled with multi-action and high-dimensional scenarios, but this innovative framework promises to bridge that gap. By utilizing Shapley-based regression, it allows for better understanding of policy behavior, which is crucial for building trust in AI systems. This advancement not only boosts the performance of RL but also opens doors for its application in more intricate real-world situations.

Interpret Policies in Deep Reinforcement Learning using SILVER with RL-Guided Labeling: A Model-level Approach to High-dimensional and Multi-action Environments

Interpret Policies in Deep Reinforcement Learning using SILVER with RL-Guided Labeling: A Model-level Approach to High-dimensional and Multi-action Environments

Was this article worth reading? Share it

ISMS Policy Generator

ClassX

HubRE AI