Research map / Architectures and efficiency / RNNs and normalization
SATNet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver
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 |
|---|---|---|---|
| Use a differentiable MAXSAT SDP layer to learn Sudoku constraints from examples instead of a convolutional model with a completion mask | 9×9 Sudoku · Original Sudoku test | board-wise accuracy: 15.1% -> 98.3% | Helped |
| Use a differentiable MAXSAT SDP layer to learn Sudoku constraints from examples instead of a convolutional model with a completion mask | 9×9 Sudoku · Permuted Sudoku test | board-wise accuracy: 0% -> 98.3% | Helped |
| Connect a visual digit classifier to a SATNet constraint-solving layer for end-to-end training instead of using convolutional layers for both stages | Visual Sudoku · Visual Sudoku test | board-wise accuracy: 0.1% -> 63.2% | Helped |
| Use SATNet rather than an LSTM sequence classifier | Parity (chained XOR) · Input lengths 20 and 40 | test error rate: 0.476 | Helped |
| Use a differentiable MAXSAT solver rather than a convolutional model to solve 4×4 Sudoku | 4×4 Sudoku · 4×4 Sudoku test | board-wise accuracy: 93% -> 100% | Helped |
| Use low-rank coordinate descent instead of the corresponding OpenMP CPU implementation | SATNet GPU implementation · Preliminary runtime benchmark | speedup: 18-30x | Helped |
Closest papers by meaning
- Learning Reliable Logical Rules with SATNet (NeurIPS 2023 · similarity 0.94)
- Assessing SATNet's Ability to Solve the Symbol Grounding Problem (NeurIPS 2020 · similarity 0.91)
- Learning Better Representations From Less Data For Propositional Satisfiability (NeurIPS 2024 · similarity 0.86)
- Learning How to Cube (NeurIPS 2026 · similarity 0.86)
- Teaching Transformers to Solve Combinatorial Problems through Efficient Trial & Error (NeurIPS 2025 · similarity 0.86)
- Causal language modeling can elicit search and reasoning capabilities on logic puzzles (NeurIPS 2024 · similarity 0.85)
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.