Saved in:
Bibliographic Details
Main Authors: Eaton, Kenneth, Balloch, Jonathan, Kim, Julia, Riedl, Mark
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.19532
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913450255450112
author Eaton, Kenneth
Balloch, Jonathan
Kim, Julia
Riedl, Mark
author_facet Eaton, Kenneth
Balloch, Jonathan
Kim, Julia
Riedl, Mark
contents Interpretability of deep reinforcement learning systems could assist operators with understanding how they interact with their environment. Vector quantization methods -- also called codebook methods -- discretize a neural network's latent space that is often suggested to yield emergent interpretability. We investigate whether vector quantization in fact provides interpretability in model-based reinforcement learning. Our experiments, conducted in the reinforcement learning environment Crafter, show that the codes of vector quantization models are inconsistent, have no guarantee of uniqueness, and have a limited impact on concept disentanglement, all of which are necessary traits for interpretability. We share insights on why vector quantization may be fundamentally insufficient for model interpretability.
format Preprint
id arxiv_https___arxiv_org_abs_2407_19532
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle The Interpretability of Codebooks in Model-Based Reinforcement Learning is Limited
Eaton, Kenneth
Balloch, Jonathan
Kim, Julia
Riedl, Mark
Artificial Intelligence
Machine Learning
Interpretability of deep reinforcement learning systems could assist operators with understanding how they interact with their environment. Vector quantization methods -- also called codebook methods -- discretize a neural network's latent space that is often suggested to yield emergent interpretability. We investigate whether vector quantization in fact provides interpretability in model-based reinforcement learning. Our experiments, conducted in the reinforcement learning environment Crafter, show that the codes of vector quantization models are inconsistent, have no guarantee of uniqueness, and have a limited impact on concept disentanglement, all of which are necessary traits for interpretability. We share insights on why vector quantization may be fundamentally insufficient for model interpretability.
title The Interpretability of Codebooks in Model-Based Reinforcement Learning is Limited
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2407.19532