ScheduleNet: learning to solve multi-worker scheduling problems using Graph Neural Networks and Reinforcement Learning.

Abstract

In this paper, we present a general framework for learning deep, parameterized scheduling policies in a online, end-to-end fashion, using Graph Neural Networks and Reinforcement Learning. Specifically, we introduce a novel heteregeneous worker-task graph representation that is suitable for solving both single and multi-worker scheduling problems. We empirically show effectiveness of our approach by solving MinMax Multiple Travelling Salesman Problem, achieving comparable performance to leading metaheuristic algorithm Google OR-Tools on real-world mTSP instances.