Publication Details
Overview
 
 
Senne Deproost, Ann Nowe, Mehrdad Asadi
 

Chapter in Book/ Report/ Conference proceeding

Abstract 

We introduce State Vector Space Partitioning (SVSP), a novel method to mimic a black-box reinforcement learning policy using a set of human-interpretable sub-policies. By partitioning a distillation dataset of state–action pairs with linear support vector machine splits, SVSP constructs a compact and structured represen- tation of the original policy where linear models can be interpreted as a measure of feature importance. Our method improves mean return by +7.4\% over previous critic-driven state partitioning attempts such as Voronoi State Partitioning (VSP) and +2.8\% over the original TD3 policy, while reducing the number of required sub-policies against VSP by 82.1\%. Our results pave the path towards a more flexi- ble form of distillation where both the decision boundary and surrogate models can be chosen within a margin of the original black box behavior.

Reference