Transformer models applied to flow shop scheduling problem

Researchers have developed a novel approach to flow shop scheduling by employing transformer models, a type of machine learning architecture. This method treats scheduling as a next-token prediction task, where tokens represent job-machine-secondary resource assignments. The models are trained on data generated by Mixed Integer Linear Programming (MILP) and then used to complete partial schedules during inference. In computational studies, this transformer-based method demonstrated superior solution quality compared to genetic algorithms and other heuristics, though it was slightly outperformed by MILP and an iterated greedy heuristic. AI

IMPACT This research suggests that transformer models can be effectively applied to complex scheduling problems, potentially improving efficiency in operations research and logistics.

RANK_REASON The item is an academic paper detailing a novel application of machine learning techniques to a specific optimization problem. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.LG →

• arXiv
• Flow Shop Scheduling Problem
• genetic algorithm
• Hugging Face
• Iterated Greedy heuristic
• machine learning
• Mixed Integer Linear Programming
• NEH heuristic
• operations research
• Transformer

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →