Research map / Large language models / Agents, code and math
FACT: Learning Governing Abstractions Behind Integer Sequences
Peter Belcak, Ard Kastrati, Flavio Schenker, Roger Wattenhofer
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 |
|---|---|---|---|
| train and test on synthetic instead of OEIS data | RNN · sequence classification | F1 score: 0.37 -> 0.53 | Improved |
| train and test on synthetic instead of OEIS data | Transformer · sequence continuation | RMSLE: 0.578 -> 0.395 | Improved |
| use RNN instead of MLP for next sequence-part prediction | RNN · next sequence-part prediction, OEIS | binary accuracy: 0.733 -> 0.869 | Improved |
| use RNN instead of CNN for next sequence-part prediction | RNN · next sequence-part prediction, OEIS | binary accuracy: 0.551 -> 0.869 | Improved |
| use Transformer instead of RNN for continuation | Transformer · sequence continuation, OEIS | RMSLE: 0.603 -> 0.578 | Improved |
| use Transformer instead of CNN for similarity | Transformer · sequence similarity, OEIS | top-5 RMSE: 0.428 -> 0.267 | Improved |
| use CNN instead of Transformer for unmasking | CNN · sequence element unmasking, OEIS | top-5 RMSE: 2.811 -> 2.440 | Improved |
| use random forest instead of dummy classifier | Random Forest · sequence classification, synthetic | F1 score: 0.50 -> 0.51 | Improved |
| use RNN instead of random forest for classification | RNN · sequence classification, synthetic | F1 score: 0.51 -> 0.53 | Improved |
| use XGBoost instead of dummy classifier | XGBoost · sequence classification, OEIS | F1 score: 0.50 -> 0.37 | Hurt |
| use linear regressor instead of dummy for continuation | Linear Regressor · sequence continuation, OEIS | RMSLE: 0.923 -> 0.797 | Improved |
| use Transformer instead of RNN for similarity on synthetic | Transformer · sequence similarity, synthetic | top-5 RMSE: 0.438 -> 0.270 | Improved |
Closest papers by meaning
- Benchmarking Abstract and Reasoning Abilities Through A Theoretical Perspective (ICML 2025 · similarity 0.88)
- Benchmarking Large Language Models with Integer Sequence Generation Tasks (NeurIPS 2025 · similarity 0.87)
- Emergent Symbolic Mechanisms Support Abstract Reasoning in Large Language Models (ICML 2025 · similarity 0.87)
- Meaningful Learning: Enhancing Abstract Reasoning in Large Language Models via Generic Fact Guidance (NeurIPS 2024 · similarity 0.86)
- CORE: Concept-Oriented Reinforcement for Bridging the Definition–Application Gap in Mathematical Reasoning (ICLR 2026 · similarity 0.86)
- Turning Dust into Gold: Distilling Complex Reasoning Capabilities from LLMs by Leveraging Negative Data (AAAI 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.
