EnerGNN learns energy functions for exact constrained combinatorial inference
Original title: EnerGNN: Learning Optimization-Compatible Energy Functions for Exact Constrained Combinatorial Inference
On a cell-therapy supply-chain task it reports a 0.05% mean primal gap and a 17.4× MIP speedup.
EnerGNN learns energy functions over complete binary configurations and minimizes them under explicit hard constraints with mixed integer programming. A graph neural network encoder, optimization-compatible readout, and nonlinear energy head support exact constrained inference. An optimality-aware contrastive loss trains the energy for solution quality and constraint violations. Across Quadratic Knapsack, Combinatorial Auctions, and a cell-therapy supply-chain problem, the authors report near-optimal solutions with 100% feasibility.
- Supply-chain result is a 0.05% mean primal gap and 17.4× speedup over monolithic MIP.
- Learned energy generalizes to larger instances, the authors report.
- Zero-shot constraint injection adapts to new constraints without retraining.
- Feasibility stayed at 100% across the reported tasks.
- AI models
- EnerGNN
From the original article
1 min read at arxiv.orgLearning-based approaches to combinatorial optimization often predict solutions directly or use predictions to restrict search, while expressive energy-based models typically rely on approximate inference. We introduce EnerGNN, a framework that learns energy functions over complete binary configurations and minimizes them under explicit hard constraints through mixed integer programming (MIP).
By Niki Triantafyllou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.