What Planning Problems Can A Relational Neural Network Solve?

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mao, Jiayuan, Lozano-Pérez, Tomás, Tenenbaum, Joshua B., Kaelbling, Leslie Pack
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910432821772288
author Mao, Jiayuan
Lozano-Pérez, Tomás
Tenenbaum, Joshua B.
Kaelbling, Leslie Pack
author_facet Mao, Jiayuan
Lozano-Pérez, Tomás
Tenenbaum, Joshua B.
Kaelbling, Leslie Pack
contents Goal-conditioned policies are generally understood to be "feed-forward" circuits, in the form of neural networks that map from the current state and the goal specification to the next action to take. However, under what circumstances such a policy can be learned and how efficient the policy will be are not well understood. In this paper, we present a circuit complexity analysis for relational neural networks (such as graph neural networks and transformers) representing policies for planning problems, by drawing connections with serialized goal regression search (S-GRS). We show that there are three general classes of planning problems, in terms of the growth of circuit width and depth as a function of the number of objects and planning horizon, providing constructive proofs. We also illustrate the utility of this analysis for designing neural networks for policy learning.
format Preprint
id arxiv_https___arxiv_org_abs_2312_03682
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle What Planning Problems Can A Relational Neural Network Solve?
Mao, Jiayuan
Lozano-Pérez, Tomás
Tenenbaum, Joshua B.
Kaelbling, Leslie Pack
Machine Learning
Artificial Intelligence
Neural and Evolutionary Computing
Goal-conditioned policies are generally understood to be "feed-forward" circuits, in the form of neural networks that map from the current state and the goal specification to the next action to take. However, under what circumstances such a policy can be learned and how efficient the policy will be are not well understood. In this paper, we present a circuit complexity analysis for relational neural networks (such as graph neural networks and transformers) representing policies for planning problems, by drawing connections with serialized goal regression search (S-GRS). We show that there are three general classes of planning problems, in terms of the growth of circuit width and depth as a function of the number of objects and planning horizon, providing constructive proofs. We also illustrate the utility of this analysis for designing neural networks for policy learning.
title What Planning Problems Can A Relational Neural Network Solve?
topic Machine Learning
Artificial Intelligence
Neural and Evolutionary Computing
url https://arxiv.org/abs/2312.03682