Research map / Large language models / Language, safety and interpretability
Dynamic Prompt Learning via Policy Gradient for Semi-structured Mathematical Reasoning
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 |
|---|---|---|---|
| Learn to select in-context examples with policy gradient instead of selecting them randomly | Few-shot-CoT GPT-3 · TabMWP test | accuracy: 5.31% | Helped |
| Learn to select examples dynamically with policy gradient rather than random selection | Few-shot-CoT GPT-3 · TabMWP development (1,000 examples) | accuracy: 65.2 $\pm$ 4.01 -> 70.9 $\pm$ 1.27 | Helped |
| Select examples with the same question type instead of random examples | Few-shot-CoT GPT-3 · TabMWP development (1,000 examples) | accuracy: 65.2 $\pm$ 4.01 -> 66.2 $\pm$ 0.60 | Helped |
| Select examples with the same answer type instead of random examples | Few-shot-CoT GPT-3 · TabMWP development (1,000 examples) | accuracy: 65.2 $\pm$ 4.01 -> 67.9 $\pm$ 0.38 | Helped |
| Select examples from the same grade level instead of random examples | Few-shot-CoT GPT-3 · TabMWP development (1,000 examples) | accuracy: 65.2 $\pm$ 4.01 -> 67.9 $\pm$ 1.87 | Helped |
| Select the most complex examples by table-cell count | Few-shot-CoT GPT-3 · TabMWP development (1,000 examples) | accuracy: 65.2 $\pm$ 4.01 -> 64.0 $\pm$ 0.42 | Hurt |
| Increase randomly selected in-context examples from 2 to 4 | Few-shot-CoT GPT-3 · TabMWP development (1,000 examples) | accuracy: 65.2 $\pm$ 4.01 -> 67.7 $\pm$ 0.78 | Helped |
| Increase randomly selected in-context examples from 4 to 5 | Few-shot-CoT GPT-3 · TabMWP development (1,000 examples) | accuracy: 67.7 $\pm$ 0.78 -> 67.5 $\pm$ 0.98 | Hurt |
| Remove the table context from the input (ablation) | Zero-shot GPT-3 on TabMWP · TabMWP development | accuracy: 6.10% | Hurt |
| Remove the question text from the input (ablation) | Zero-shot GPT-3 on TabMWP · TabMWP development | accuracy: 7.00% | Hurt |
| Add more training examples beyond around 160 (ablation) | PromptPG · TabMWP development | accuracy | Hurt |
| Increase the candidate pool beyond its effective threshold (ablation) | PromptPG · TabMWP development | accuracy | Hurt |
Closest papers by meaning
- Template-Driven LLM-Paraphrased Framework for Tabular Math Word Problem Generation (AAAI 2025 · similarity 0.89)
- We-Math 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning (ICLR 2026 · similarity 0.88)
- StrategyLLM: Large Language Models as Strategy Generators, Executors, Optimizers, and Evaluators for Problem Solving (NeurIPS 2024 · similarity 0.87)
- CurES: From Gradient Analysis to Efficient Curriculum Learning for Reasoning LLMs (ICLR 2026 · similarity 0.87)
- MathSE: Improving Multimodal Mathematical Reasoning via Self-Evolving Iterative Reflection and Reward-Guided Fine-Tuning (AAAI 2026 · similarity 0.87)
- MathScale: Scaling Instruction Tuning for Mathematical Reasoning (ICML 2024 · 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.