Autoregressive RL Approach for Mixed-Integer Linear Programs

Paul Mingzheng Tang, Hoong Chuin Lau, Moses Hong De Lee

Confirmation: I have read and agree with the workshop's policy on behalf of myself and my co-authors.

Tracks: Special Session 1: (Deep) Reinforcement Learning in OR Optimization

Keywords: Reinforcement Learning, Graph Neural Networks, ML for OR

TL;DR: An autoregressive RL framework for generating partial solutions for mixed-integer linear programming, directly optimizing for optimality within a fixed time limit.

Abstract: Mixed-Integer Linear Programming (MILP) is a widely used method for modeling combinatorial optimization problems. Due to the NP-hard nature of many of these problems, efforts using machine-learning (ML) have been proposed to generate heuristics to speed up solvers, while maintaining optimality. While there is prior work on using graph neural networks (GNNs) to produce high-quality partial solutions, the methods used are non-auto-regressive which model the prediction of variables as conditionally independent due to concerns with inference speed. In this paper, we propose a novel auto-regressive reinforcement learning (RL) framework using GNNs which directly optimize for optimality and inference speed. Experimental results show our RL method outperforms the benchmark Predict-And-Search (PNS) method on harder real-world problems (55.7% speedup) with time limits and matches performance on easier problems.

Submission Number: 54