Limitations of Codebook Interpretability in Model-Based Reinforcement Learning

The Interpretability of Codebooks in Model-Based Reinforcement Learning is Limited

Kenneth Eaton, Jonathan Balloch, Julia Kim, Mark Riedl

Original Paper: https://arxiv.org/abs/2407.19532

InteOriginal paper: https://arxiv.org/abs/2407.19532

Title: The Interpretability of Codebooks in Model-Based Reinforcement Learning is Limited

Authors: Kenneth Eaton, Jonathan Balloch, Julia Kim, Mark Riedl

Abstract:
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.

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.