SIMULTANEOUS GENERATION AND IMPROVEMENT: A UNIFIED RL PARADIGM FOR FJSP OPTIMIZATION

Hongyi Zhou, Lei Song

Primary Area: reinforcement learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Reinforcement Learning, Flexible Job Shop Schedule Problem, FJSP

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: A Unified RL Paradigm to solve Flexible Job Shop Optimization

Abstract: We present an end-to-end reinforcement learning framework designed to address the Flexible Job Shop Problem (FJSP). Our approach consists of two primary components: a generative model that produces problem solutions stepwise, and a secondary model that continually refines these (partial) solutions. Importantly, we train both models concurrently, enabling each to be cognizant of the other's policy and make informed decisions. Extensive experimentation demonstrates that our model delivers better performance in shorter time on several public datasets comparing to baseline algorithms. Furthermore, we highlight the superior generalizability of our approach, as it maintains strong performance on large-scale instances even when trained on small-scale instances. It is worth noting that this training paradigm can be readily adapted to other combinatorial optimization problems, such as the traveling salesman problemand beyond.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9463