Learning for Dynamic Combinatorial Optimization without Training Data

2025-05-26

AI Summary

Addressing the challenge of scarce labeled training data in dynamic combinatorial optimization (DCO), this paper proposes DyCO-GNN—the first fully training-data-free, unsupervised learning framework. DyCO-GNN leverages structural similarity across temporal graph snapshots, modeling dynamic topologies via graph neural networks and enabling cross-timestep knowledge transfer through unsupervised contrastive learning. It further incorporates a dynamic solution-space guidance mechanism to enhance computational efficiency. Evaluated on dynamic Max-Cut, Maximum Independent Set (MIS), and Traveling Salesman Problem (TSP) benchmarks, DyCO-GNN achieves speedups of 3–60× over state-of-the-art baselines while preserving solution quality. This significantly improves real-time responsiveness and resource efficiency, establishing a novel paradigm for DCO in zero-shot, label-free scenarios—where no historical instances or ground-truth labels are available.

Technology Category

• Search and Optimization: Combinatorial Optimization
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We introduce DyCO-GNN, a novel unsupervised learning framework for Dynamic Combinatorial Optimization that requires no training data beyond the problem instance itself. DyCO-GNN leverages structural similarities across time-evolving graph snapshots to accelerate optimization while maintaining solution quality. We evaluate DyCO-GNN on dynamic maximum cut, maximum independent set, and the traveling salesman problem across diverse datasets of varying sizes, demonstrating its superior performance under tight and moderate time budgets. DyCO-GNN consistently outperforms the baseline methods, achieving high-quality solutions up to 3-60x faster, highlighting its practical effectiveness in rapidly evolving resource-constrained settings.

Problem

• Unsupervised learning for dynamic combinatorial optimization without training data
• Leveraging structural similarities in time-evolving graphs for faster optimization
• Superior performance in dynamic problems under tight time constraints

Innovation

• Unsupervised learning without training data
• Leverages time-evolving graph similarities
• Achieves faster high-quality solutions