Research map / Label-efficient and robust learning / Self-supervised and contrastive
Techniques for Symbol Grounding with SATNet
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 |
|---|---|---|---|
| apply self-supervised clustering and Symbol Grounding Loss to learn without input labels | SATNet · Visual Sudoku ungrounded | total board accuracy (%): 0 $\pm$ 0.0 -> 64.8 $\pm$ 3.0 | Helped |
| apply self-supervised clustering and Symbol Grounding Loss to learn without input labels | SATNet · Visual Sudoku ungrounded | per-cell accuracy (%): 11.2 $\pm$ 0.1 -> 98.4 $\pm$ 0.2 | Helped |
| apply self-supervised clustering and Symbol Grounding Loss to learn without input labels | SATNet · Visual Sudoku ungrounded | visual accuracy (%): 11.6 $\pm$ 0.0 -> 98.9 $\pm$ 0.1 | Helped |
| add a Proofreader layer (ablation) | Original Non-visual SATNet · Non-visual Sudoku | total board accuracy (%): 96.6 $\pm$ 0.3 -> 97.1 $\pm$ 0.3 | Helped |
| add a Proofreader layer (ablation) | Original Visual SATNet · Visual Sudoku grounded | total board accuracy (%): 66.5 $\pm$ 1.0 -> 67.6 $\pm$ 1.2 | Helped |
| add a Proofreader layer (ablation) | Our Method · Visual Sudoku ungrounded | total board accuracy (%): 62.8 $\pm$ 3.2 -> 64.8 $\pm$ 3.0 | Helped |
| add a Proofreader layer (ablation) | Original Non-visual SATNet · Non-visual Sudoku | per-cell accuracy (%): 99.9 $\pm$ 0.0 -> 99.9 $\pm$ 0.0 | No clear effect |
| add a Proofreader layer (ablation) | Original Visual SATNet · Visual Sudoku grounded | per-cell accuracy (%): 98.8 $\pm$ 0.1 -> 98.6 $\pm$ 0.1 | Hurt |
| add a Proofreader layer (ablation) | Our Method · Visual Sudoku ungrounded | per-cell accuracy (%): 98.6 $\pm$ 0.1 -> 98.4 $\pm$ 0.2 | Hurt |
| add a Proofreader layer (ablation) | Original Visual SATNet · Visual Sudoku grounded | visual accuracy (%): 99.0 $\pm$ 0.0 -> 99.0 $\pm$ 0.0 | No clear effect |
| add a Proofreader layer (ablation) | Our Method · Visual Sudoku ungrounded | visual accuracy (%): 98.9 $\pm$ 0.1 -> 98.9 $\pm$ 0.1 | No clear effect |
| distill InfoGAN clusters into a smaller LeNet classifier (ablation) | InfoGAN · MNIST digit clustering | clustering accuracy (%): 89.6 $\pm$ 7.7 -> 86.2 $\pm$ 13.5 | Mixed |
Closest papers by meaning
- Assessing SATNet's Ability to Solve the Symbol Grounding Problem (NeurIPS 2020 · similarity 0.96)
- NeuroBack: Improving CDCL SAT Solving using Graph Neural Networks (ICLR 2024 · similarity 0.84)
- Learning How to Cube (NeurIPS 2026 · similarity 0.84)
- Learning Better Representations From Less Data For Propositional Satisfiability (NeurIPS 2024 · similarity 0.84)
- Constraints-Guided Diffusion Reasoner for Neuro-Symbolic Learning (AAAI 2026 · similarity 0.83)
- Causal language modeling can elicit search and reasoning capabilities on logic puzzles (NeurIPS 2024 · similarity 0.83)
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.