Research map / Probabilistic and scientific modeling / Molecules and proteins
A-NeSI: A Scalable Approximate Method for Probabilistic Neurosymbolic Inference
Emile van Krieken, Thiviyan Thanapalasingam, Jakub M. Tomczak, Frank Van Harmelen, Annette Ten Teije
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| remove the prior p(P) used to sample beliefs for training the inference model (ablation) | A-NeSI (predict) · Multi-digit MNISTAdd N=15 | accuracy: 75.90 -> 0.03 | Hurt |
| remove the prior p(P) (ablation) | A-NeSI (predict) · Multi-digit MNISTAdd N=4 | accuracy: 92.56 -> 44.58 | Hurt |
| remove the prior p(P) (ablation) | A-NeSI (predict) · Multi-digit MNISTAdd N=2 | accuracy: 95.96 -> 95.67 | No clear effect |
| explainable variant instead of prediction-only (ablation) | A-NeSI · Multi-digit MNISTAdd N=15 (symbolic) | accuracy: 75.90 -> 76.84 | Improved |
| explainable variant instead of prediction-only (ablation) | A-NeSI · Multi-digit MNISTAdd N=15 (neural prediction) | accuracy: 54.66 -> 61.77 | Improved |
| add symbolic pruning to prediction-only variant (ablation) | A-NeSI · Multi-digit MNISTAdd N=15 (neural prediction) | accuracy: 54.66 -> 59.88 | Improved |
| add symbolic pruning to prediction-only variant (ablation) | A-NeSI · Multi-digit MNISTAdd N=2 (symbolic) | accuracy: 95.96 -> 95.82 | No clear effect |
| neural prediction instead of symbolic prediction | A-NeSI (predict) · Multi-digit MNISTAdd N=15 | accuracy: 75.90 -> 54.66 | Hurt |
| A-NeSI approximate inference instead of exact inference (DeepProbLog) | A-NeSI (predict) · Multi-digit MNISTAdd N=2 | accuracy: 95.20 -> 95.96 | No clear effect |
| A-NeSI instead of Embed2Sym baseline | A-NeSI (predict) · Multi-digit MNISTAdd N=15 (symbolic) | accuracy: 60.46 -> 75.90 | Improved |
| train neural predictor with A-NeSI loop vs Embed2Sym without background knowledge | A-NeSI (predict) · Multi-digit MNISTAdd N=2 (neural prediction) | accuracy: 84.35 -> 95.95 | Improved |
| A-NeSI instead of exact inference on 4x4 Sudoku | A-NeSI · Visual Sudoku N=4 | accuracy: 86.70 -> 89.70 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- DeepStochLog — DeepStochLog: Neural Stochastic Logic Programming (AAAI 2022) · better on 0 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- On the Hardness of Probabilistic Neurosymbolic Learning (ICML 2024 · similarity 0.90)
- EM-NeSy: Expectation Maximization for Neurosymbolic Learning (NeurIPS 2026 · similarity 0.90)
- Right for the Right Reasons: Avoiding Reasoning Shortcuts via Prototypical Neurosymbolic AI (NeurIPS 2025 · similarity 0.88)
- DOLPHIN: A Programmable Framework for Scalable Neurosymbolic Learning (ICML 2025 · similarity 0.88)
- Relational Neurosymbolic Markov Models (AAAI 2025 · similarity 0.88)
- Verification Learning: Make Unsupervised Neuro-Symbolic System Feasible (ICML 2025 · 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.
