Planning in a recurrent neural network that plays Sokoban

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Taufeeque, Mohammad, Quirke, Philip, Li, Maximilian, Cundy, Chris, Tucker, Aaron David, Gleave, Adam, Garriga-Alonso, Adrià
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912402839175168
author Taufeeque, Mohammad
Quirke, Philip
Li, Maximilian
Cundy, Chris
Tucker, Aaron David
Gleave, Adam
Garriga-Alonso, Adrià
author_facet Taufeeque, Mohammad
Quirke, Philip
Li, Maximilian
Cundy, Chris
Tucker, Aaron David
Gleave, Adam
Garriga-Alonso, Adrià
contents Planning is essential for solving complex tasks, yet the internal mechanisms underlying planning in neural networks remain poorly understood. Building on prior work, we analyze a recurrent neural network (RNN) trained on Sokoban, a challenging puzzle requiring sequential, irreversible decisions. We find that the RNN has a causal plan representation which predicts its future actions about 50 steps in advance. The quality and length of the represented plan increases over the first few steps. We uncover a surprising behavior: the RNN "paces" in cycles to give itself extra computation at the start of a level, and show that this behavior is incentivized by training. Leveraging these insights, we extend the trained RNN to significantly larger, out-of-distribution Sokoban puzzles, demonstrating robust representations beyond the training regime. We open-source our model and code, and believe the neural network's interesting behavior makes it an excellent model organism to deepen our understanding of learned planning.
format Preprint
id arxiv_https___arxiv_org_abs_2407_15421
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Planning in a recurrent neural network that plays Sokoban
Taufeeque, Mohammad
Quirke, Philip
Li, Maximilian
Cundy, Chris
Tucker, Aaron David
Gleave, Adam
Garriga-Alonso, Adrià
Machine Learning
Artificial Intelligence
Planning is essential for solving complex tasks, yet the internal mechanisms underlying planning in neural networks remain poorly understood. Building on prior work, we analyze a recurrent neural network (RNN) trained on Sokoban, a challenging puzzle requiring sequential, irreversible decisions. We find that the RNN has a causal plan representation which predicts its future actions about 50 steps in advance. The quality and length of the represented plan increases over the first few steps. We uncover a surprising behavior: the RNN "paces" in cycles to give itself extra computation at the start of a level, and show that this behavior is incentivized by training. Leveraging these insights, we extend the trained RNN to significantly larger, out-of-distribution Sokoban puzzles, demonstrating robust representations beyond the training regime. We open-source our model and code, and believe the neural network's interesting behavior makes it an excellent model organism to deepen our understanding of learned planning.
title Planning in a recurrent neural network that plays Sokoban
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2407.15421