Research map / Learning theory and algorithms / Combinatorial and search optimization
Learning Logic Programs Though Divide, Constrain, and Conquer
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 DCC instead of Popper | ILP learning on trains4 · trains4 | predictive accuracy: 42 $\pm$ 5 -> 100 $\pm$ 0 | Helped |
| use DCC instead of Popper | ILP learning on buttons · buttons | predictive accuracy: 19 $\pm$ 0 -> 98 $\pm$ 0 | Helped |
| use DCC instead of Popper | ILP learning on coins · coins | predictive accuracy: 17 $\pm$ 0 -> 86 $\pm$ 0 | Helped |
| use DCC instead of Popper | ILP learning on sorted · sorted | predictive accuracy: 96 $\pm$ 1 -> 94 $\pm$ 2 | Hurt |
| reuse constraints learned from failures (ablation) | DCC on last · last | predictive accuracy: 82 $\pm$ 5 -> 100 $\pm$ 0 | Helped |
| reuse constraints learned from failures (ablation) | DCC on last · last | learning time (s): 107 $\pm$ 32 -> 2 $\pm$ 0.4 | Helped |
| remove constraint reuse (ablation) | DCC on trains1 · trains1 | predictive accuracy: 100 $\pm$ 0 -> 95 $\pm$ 3 | Hurt |
| use laziness to reuse a hypothesis that already covers a chunk (ablation) | DCC on trains1 · trains1 | learning time (s): 300 $\pm$ 0 -> 8 $\pm$ 2 | Helped |
| remove laziness (ablation) | DCC on sorted · sorted | learning time (s): 29 $\pm$ 3 -> 74 $\pm$ 7 | Hurt |
| use compression to merge chunks covered by the same hypothesis (ablation) | DCC on sorted · sorted | learning time (s): 109 $\pm$ 21 -> 29 $\pm$ 3 | Helped |
| remove compression (ablation) | DCC on trains1 · trains1 | learning time (s): 8 $\pm$ 2 -> 12 $\pm$ 3 | Hurt |
| divide positive examples into progressively larger chunks and reuse constraints across searches | DCC versus Popper · IGGP games | predictive accuracy | Mixed |
Closest papers by meaning
- Learning Logic Programs by Discovering Where Not to Search (AAAI 2023 · similarity 0.93)
- Learning MDL Logic Programs from Noisy Data (AAAI 2024 · similarity 0.88)
- Self-Supervised Inductive Logic Programming (AAAI 2026 · similarity 0.87)
- Inductive Learning of Logical Theories with LLMs: A Expressivity-graded Analysis (AAAI 2025 · similarity 0.84)
- Logic.py: Bridging the Gap between LLMs and Constraint Solvers (NeurIPS 2025 · similarity 0.84)
- Enhancing Reasoning Capabilities of LLMs via Principled Synthetic Logic Corpus (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.