Critic-Driven Voronoi-Quantization for Distilling Deep RL Policies to Explainable Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Deproost, Senne, Steckelmacher, Denis, Nowé, Ann
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909043884294144
author Deproost, Senne
Steckelmacher, Denis
Nowé, Ann
author_facet Deproost, Senne
Steckelmacher, Denis
Nowé, Ann
contents Despite many successful attempts at explaining Deep Reinforcement Learning policies using distillation, it remains difficult to balance the performance-interpretability trade-off and select a fitting surrogate model. In addition to this, traditional distillation only minimizes the distance between the behavior of the original and the surrogate policy while other RL-specific components such as action value are disregarded. To solve this, we introduce a new model-agnostic method called Critic-Driven Voronoi State Partitioning, which partitions a black box control policy into regions where a simple class of model can be optimized using gradient descent. By exploiting the critic value network of the original policy, we iteratively introduce new subpolicies in regions with insufficient value, standing in for a measure of policy complexity. The partitioning, a Voronoi quantizer, uses nearest neighbor lookups to assign a linear function to each point in the state space resulting in a cell-like diagram. We validate our approach on several well known benchmarks and proof that this distillation approaches the original policy using a reasonable sized set of linear functions.
format Preprint
id arxiv_https___arxiv_org_abs_2605_14897
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Critic-Driven Voronoi-Quantization for Distilling Deep RL Policies to Explainable Models
Deproost, Senne
Steckelmacher, Denis
Nowé, Ann
Machine Learning
Artificial Intelligence
Despite many successful attempts at explaining Deep Reinforcement Learning policies using distillation, it remains difficult to balance the performance-interpretability trade-off and select a fitting surrogate model. In addition to this, traditional distillation only minimizes the distance between the behavior of the original and the surrogate policy while other RL-specific components such as action value are disregarded. To solve this, we introduce a new model-agnostic method called Critic-Driven Voronoi State Partitioning, which partitions a black box control policy into regions where a simple class of model can be optimized using gradient descent. By exploiting the critic value network of the original policy, we iteratively introduce new subpolicies in regions with insufficient value, standing in for a measure of policy complexity. The partitioning, a Voronoi quantizer, uses nearest neighbor lookups to assign a linear function to each point in the state space resulting in a cell-like diagram. We validate our approach on several well known benchmarks and proof that this distillation approaches the original policy using a reasonable sized set of linear functions.
title Critic-Driven Voronoi-Quantization for Distilling Deep RL Policies to Explainable Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2605.14897