Abstract:The Max-k-Cut problem is a fundamental combinatorial optimization challenge that generalizes the classic NP-complete Max-Cut problem. While relaxation techniques are commonly employed to tackle Max-k-Cut, they often lack guarantees of equivalence between the solutions of the original problem and its relaxation. To address this issue, we introduce the Relax-Optimize-and-Sample (ROS) framework. In particular, we begin by relaxing the discrete constraints to the continuous probability simplex form. Next, we pre-train and fine-tune a graph neural network model to efficiently optimize the relaxed problem. Subsequently, we propose a sampling-based construction algorithm to map the continuous solution back to a high-quality Max-k-Cut solution. By integrating geometric landscape analysis with statistical theory, we establish the consistency of function values between the continuous solution and its mapped counterpart. Extensive experimental results on random regular graphs and the Gset benchmark demonstrate that the proposed ROS framework effectively scales to large instances with up to 20000 nodes in just a few seconds, outperforming state-of-the-art algorithms. Furthermore, ROS exhibits strong generalization capabilities across both in-distribution and out-of-distribution instances, underscoring its effectiveness for large-scale optimization tasks.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the **Max - k - Cut problem**, which is a classic combinatorial optimization difficult problem. Specifically, the Max - k - Cut problem is to partition the vertex set of a graph into k disjoint subsets so as to maximize the total weight of the edges between different subsets. This problem is classified as an NP - complete problem because it generalizes the well - known Max - Cut problem. #### Challenges of the Max - k - Cut problem 1. **Discrete constraints**: Traditional solution methods usually transform discrete constraints into a continuous probability simplex form through relaxation techniques, but this may lead to inconsistent solutions between the original problem and the relaxed problem. 2. **Computational complexity**: For large - scale instances, existing exact algorithms and approximation algorithms are very computationally expensive and it is difficult to find high - quality solutions within a reasonable time. 3. **Generalization ability**: Existing methods often cannot guarantee good generalization performance when dealing with data of different distributions. #### Solutions proposed in the paper To address the above challenges, the authors propose a **Relax - Optimize - and - Sample (ROS) framework** based on graph neural networks (GNN). The main steps of this framework are as follows: 1. **Continuous relaxation**: Relax the discrete constraints to a continuous probability simplex form, changing the problem from discrete optimization to continuous optimization. 2. **Optimization**: Use pre - trained and fine - tuned GNN models to efficiently optimize the relaxed continuous problem. 3. **Sampling**: Propose a sampling - based construction algorithm to map continuous solutions back to high - quality Max - k - Cut solutions. #### Key contributions 1. **Novel framework**: Propose a scalable ROS framework specifically for the weighted Max - k - Cut problem with arbitrary signed weights, using efficient learning - based techniques to solve the continuous relaxation problem. 2. **Theoretical basis**: Conduct a strict theoretical analysis of the relaxation and sampling steps, combining geometric landscape analysis and statistical theory to prove the function value consistency between continuous solutions and their sampled discrete solutions. 3. **Superior performance**: Extensive experimental results show that this framework can generate high - quality solutions for Max - k - Cut instances with up to 20,000 nodes within a few seconds, and significantly outperforms existing algorithms while showing strong generalization ability on various instance types. Through these innovations, the ROS framework not only improves the solution efficiency of the Max - k - Cut problem but also theoretically provides a more reliable solution quality guarantee.

ROS: A GNN-based Relax-Optimize-and-Sample Framework for Max-k-Cut Problems

Decision-focused Graph Neural Networks for Combinatorial Optimization

Distributed Constrained Combinatorial Optimization leveraging Hypergraph Neural Networks

A New Relaxation Approach to Normalized Hypergraph Cut

Beyond smoothness: A general optimization framework for graph neural networks with negative Laplacian regularization

Enhancing GNNs Performance on Combinatorial Optimization by Recurrent Feature Update

On relaxations of the max $k$-cut problem formulations

Learning to Solve Combinatorial Optimization Problems on Real-World Graphs in Linear Time

Max‐flow min‐cut algorithm with application to road networks

Deep Graph Reinforcement Learning for Solving Multicut Problem

Recursive Quantum Relaxation for Combinatorial Optimization Problems

ROCO: A General Framework for Evaluating Robustness of Combinatorial Optimization Solvers on Graphs

Towards a General Recipe for Combinatorial Optimization with Multi-Filter GNNs

A unified pre-training and adaptation framework for combinatorial optimization on graphs

An Unsupervised Learning Framework Combined with Heuristics for the Maximum Minimal Cut Problem

QROSS: QUBO Relaxation Parameter Optimisation via Learning Solver Surrogates

Assessing and Enhancing Graph Neural Networks for Combinatorial Optimization: Novel Approaches and Application in Maximum Independent Set Problems

Graph Neural Networks for Maximum Constraint Satisfaction

A restart local search algorithm with relaxed configuration checking strategy for the minimum k-dominating set problem