Research map / Learning theory and algorithms / Combinatorial and search optimization
LinSATNet: The Positive Linear Satisfiability Neural Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace single-set Sinkhorn with LinSAT to enforce partial matching constraints | NGMv2 · Pascal VOC Keypoint (unfiltered) | Mean F1 (%): 60.7 -> 61.2 | Helped |
| Enforce expert preference constraints using LinSAT instead of Softmax | LSTM portfolio allocator · Portfolio allocation | Mean Sharpe: 2.15 -> 2.27 | Helped |
| Enforce expert preference constraints using LinSAT instead of Gurobi Opt | LSTM portfolio allocator · Portfolio allocation | Mean Sharpe: 2.08 -> 2.27 | Helped |
| Enforce expert preference constraints using LinSAT instead of Softmax | StemGNN portfolio allocator · Portfolio allocation | Mean Sharpe: 2.11 -> 2.42 | Helped |
| Enforce expert preference constraints using LinSAT instead of Gurobi Opt | StemGNN portfolio allocator · Portfolio allocation | Mean Sharpe: 2.00 -> 2.42 | Helped |
| Use Transformer features rather than random pre-projected values (ablation) | LinSAT TSP solver · TSP-SE | average tour length: 7.414 -> 3.811 | Helped |
| Use Transformer features rather than random pre-projected values (ablation) | LinSAT TSP solver · TSP-PRI | average tour length: 7.426 -> 3.943 | Helped |
| Use Transformer features rather than directly trained pre-projected values (ablation) | LinSAT TSP solver · TSP-SE | average tour length: 5.546 -> 3.811 | Helped |
| Use Transformer features rather than directly trained pre-projected values (ablation) | LinSAT TSP solver · TSP-PRI | average tour length: 5.646 -> 3.943 | Helped |
| Reduce temperature and use direct rounding (ablation) | LinSAT TSP solver · TSP-SE | Feasible Ratio: 1.19% -> 20.95% -> 86.69% -> 89.35% | Helped |
| Reduce temperature and use direct rounding (ablation) | Partial GM network · Partial graph matching | Mean F1: 0.4756 -> 0.5547 -> 0.5748 -> 0.5673 | Mixed |
Closest papers by meaning
- GLinSAT: The General Linear Satisfiability Neural Network Layer By Accelerated Gradient Descent (NeurIPS 2024 · similarity 0.87)
- Are Graph Neural Networks Optimal Approximation Algorithms? (NeurIPS 2024 · similarity 0.85)
- NeuroBack: Improving CDCL SAT Solving using Graph Neural Networks (ICLR 2024 · similarity 0.84)
- Learning How to Cube (NeurIPS 2026 · similarity 0.84)
- SAT-Based Tree Decomposition with Iterative Cascading Policy Selection (AAAI 2024 · similarity 0.84)
- HardCore Generation: Generating Hard UNSAT Problems for Data Augmentation (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.