Saved in:
Bibliographic Details
Main Authors: Gross, Dennis, Spieker, Helge
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2409.10218
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916396385959936
author Gross, Dennis
Spieker, Helge
author_facet Gross, Dennis
Spieker, Helge
contents Pruning neural networks (NNs) can streamline them but risks removing vital parameters from safe reinforcement learning (RL) policies. We introduce an interpretable RL method called VERINTER, which combines NN pruning with model checking to ensure interpretable RL safety. VERINTER exactly quantifies the effects of pruning and the impact of neural connections on complex safety properties by analyzing changes in safety measurements. This method maintains safety in pruned RL policies and enhances understanding of their safety dynamics, which has proven effective in multiple RL settings.
format Preprint
id arxiv_https___arxiv_org_abs_2409_10218
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Safety-Oriented Pruning and Interpretation of Reinforcement Learning Policies
Gross, Dennis
Spieker, Helge
Machine Learning
Pruning neural networks (NNs) can streamline them but risks removing vital parameters from safe reinforcement learning (RL) policies. We introduce an interpretable RL method called VERINTER, which combines NN pruning with model checking to ensure interpretable RL safety. VERINTER exactly quantifies the effects of pruning and the impact of neural connections on complex safety properties by analyzing changes in safety measurements. This method maintains safety in pruned RL policies and enhances understanding of their safety dynamics, which has proven effective in multiple RL settings.
title Safety-Oriented Pruning and Interpretation of Reinforcement Learning Policies
topic Machine Learning
url https://arxiv.org/abs/2409.10218