Research map / Architectures and efficiency / RNNs and normalization
Predicting Propositional Satisfiability via End-to-End Learning
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 |
|---|---|---|---|
| predict satisfying assignments jointly with satisfiability (ablation) | Exchangeable network · Random 3-SAT 100-600 vars | accuracy: 0.712 -> 0.726 (100 vars) | Helped |
| predict satisfying assignments jointly with satisfiability (ablation) | Message passing (NeuroSAT) · Random 3-SAT 100-600 vars | accuracy: +1.6% on average | Helped |
| use permutation-invariant architectures instead of CNN on dense matrix (ablation) | SAT prediction · Random 3-SAT | accuracy: 50.5% -> 71-84% | Helped |
| replace hand-engineered features with end-to-end exchangeable network | SAT satisfiability prediction · Random 3-SAT 100-600 vars | accuracy: 73% -> 78.9% average | Helped |
| use feed-forward NN instead of random forest on hand-engineered features | hand-engineered baseline · Random 3-SAT | accuracy: 0.744 -> 0.725 (300 vars) | No clear effect |
| exchangeable vs message passing when trained on 100 vars and tested on larger sizes | SAT prediction generalization · Random 3-SAT 550-600 vars | accuracy: 0.683 -> 0.809 (600 vars) | Helped |
| add hidden layer between pooled vector and output (ablation) | Exchangeable network · Random 3-SAT | accuracy | No clear effect |
| exchangeable vs message passing when trained and tested on same size | SAT prediction · Random 3-SAT | accuracy: 78.9% vs 79.1% average | No clear effect |
Closest papers by meaning
- NSNet: A General Neural Probabilistic Framework for Satisfiability Problems (NeurIPS 2022 · similarity 0.91)
- SATNet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver (ICML 2019 · similarity 0.90)
- HardCore Generation: Generating Hard UNSAT Problems for Data Augmentation (NeurIPS 2024 · similarity 0.88)
- Learning Better Representations From Less Data For Propositional Satisfiability (NeurIPS 2024 · similarity 0.88)
- Learning How to Cube (NeurIPS 2026 · similarity 0.88)
- NeuroBack: Improving CDCL SAT Solving using Graph Neural Networks (ICLR 2024 · similarity 0.87)
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.