Research map / Large language models / Agents, code and math
RLIE: Rule Generation with Logistic Regression, Iterative Refinement, and Evaluation for Large Language Models
Yang Yang, Hua XU, Zhangyi Hu, Yutao Yue
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 |
|---|---|---|---|
| use logistic combiner instead of prompting with rules (E2) (ablation) | RLIE (DeepSeek-V3.2) · Citations | accuracy: 60.4 -> 64.6 | Improved |
| use logistic combiner instead of prompting with rules (E2) (ablation) | RLIE (DeepSeek-V3.2) · Reviews | accuracy: 65.7 -> 70.9 | Improved |
| add rule weights to the LLM prompt (E3 vs E2) (ablation) | RLIE rules · 6 datasets, DeepSeek-V3.2 | accuracy: LLM Detect 89.6 -> 85.2; Citations 60.4 -> 55.2 | Hurt |
| add combiner prediction to the prompt (E4 vs E3) (ablation) | RLIE rules · Reviews, DeepSeek-V3.2 | accuracy: 65.4 -> 69.5 | Improved |
| add combiner prediction to the prompt (E4 vs E3) (ablation) | RLIE rules · Citations, DeepSeek-V3.2 | accuracy: 55.2 -> 57.3 (still below combiner 64.6) | Mixed |
| use stronger backbone (DeepSeek-V3.2 vs Qwen3-Next-80B) | RLIE · LLM Detect | accuracy: 87.6 -> 90.7 | Improved |
| iterative multi-rule weighted RLIE vs single-rule IO Refinement | RLIE · Headlines | accuracy: 62.0 -> 67.0 | Improved |
| add multiple rules without global weights (HypoGeniC) vs single-rule IO Refinement | HypoGeniC · Headlines | accuracy: 62.0 -> 59.9 | No clear effect |
| few-shot examples vs zero-shot | DeepSeek-V3.2 · Retweets | accuracy: 63.5 -> 57.7 | Hurt |
| hard-example driven refinement vs maintaining large rule bank | RLIE vs HypoGeniC · token cost | avg tokens (M): 13.24 -> 12.23 | Improved |
| L1 regularization prunes uninformative rules | RLIE · Headlines rule set | rule weight: Urgency rule weight 0.000 | Improved |
Closest papers by meaning
- Logically Consistent Language Models via Neuro-Symbolic Integration (ICLR 2025 · similarity 0.89)
- Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes (ICLR 2026 · similarity 0.89)
- Phenomenal Yet Puzzling: Testing Inductive Reasoning Capabilities of Language Models with Hypothesis Refinement (ICLR 2024 · similarity 0.88)
- Large Language Models are Interpretable Learners (ICLR 2025 · similarity 0.88)
- Doing Experiments and Revising Rules with Natural Language and Probabilistic Reasoning (NeurIPS 2024 · similarity 0.88)
- Learning to Disentangle Latent Reasoning Rules with Language VAEs: A Systematic Study (AAAI 2026 · similarity 0.88)
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.
