Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Blocked Gibbs meets Diffusion Transformers: Unsupervised Learning for Constraint Optimization

About

Diffusion models have shown promise in learning to solve constraint optimization problems. However, they are mostly restricted to problems with binary variables and rely on graph neural networks, hindering their application to a broader range of problems such as those with general discrete variables or constraint structures that necessitate global rather than local reasoning. We investigate the use of Diffusion Transformers to address the aforementioned limitations. A naive implementation performs poorly due to a fundamental mismatch between the standard diffusion process and constraint solving: while the former applies small, incremental denoising across all variables, the latter requires substantially altering specific subsets of variables to attain feasibility or optimality. Our method, Blocked Gibbs Diffusion Transformer (BloGDiT), is the first to address this limitation by replacing standard joint Gaussian denoising with blocked Gaussian denoising. BloGDiT uses iterative block resampling and anneals the block size over time to facilitate large, targeted edits within a block of variables. Across Sudoku, Graph Coloring, Maximum Independent Set, and MaxCut, BloGDiT matches or outperforms existing methods, demonstrating that blocked Gibbs-style diffusion provides a highly effective inductive bias for Transformer-based constraint satisfaction and optimization.

Yudong W. Xu, Wenhao Li, Xiaoyu Wang, Scott Sanner, Elias B. Khalil• 2026

Related benchmarks

TaskDatasetResultRank
Maximum Independent SetMIS Large-scale
Objective Value37.05
21
MaxCutGSET (|V|=800)
Average Gap to Best Known Cut0.11
19
MaxCutGSET (|V|=1K)
Average Gap to Best Known Cut1.33
19
MaxCutGSET |V|=2K
Average Gap to Best Known Cut7.88
18
MaxCutGSET (|V|>=3K)
Average Gap to Best Known Cut32.75
18
Maximum Independent SetMIS RB-small
Average Objective Value19.52
10
Sudoku SolvingSudoku Hard
Success Rate94.1
9
Sudoku SolvingSudoku Easy
Accuracy100
9
Graph ColoringGraph Coloring 10 Colors, n=100
Success Rate53.92
8
Graph ColoringGraph Coloring 10 Colors, n=200
Success Rate18.58
8
Showing 10 of 12 rows

Other info

Follow for update