Research map / Large language models / Reasoning and chain of thought
AdaThinking-E: One-Token Entropy Regulation for Adaptive Thinking
Zining Wang, Tongkun Guan, Boming Chen, Zhentao Guo, Jianqiang Liu, chao jin, Chen Duan, Kai zhou, Pengfei Yan, Wei Shen, Xiaokang Yang
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 |
|---|---|---|---|
| add RL stage with entropy regulation after cold start (ablation) | AdaThinking-E Qwen2.5-VL-7B · ChartQA | accuracy: 87.6 -> 89.1 | Improved |
| add RL stage with entropy regulation after cold start (ablation) | AdaThinking-E Qwen2.5-VL-7B · OCR-Reasoning | accuracy: 21.6 -> 26.5 | Improved |
| use two-stage entropy curriculum instead of constant high entropy (ablation) | AdaThinking-E · ChartQA | accuracy: 88.1 -> 89.1 | Improved |
| use two-stage entropy curriculum instead of constant high entropy (ablation) | AdaThinking-E · OCR-Reasoning | accuracy: 23.7 -> 26.5 | Improved |
| use two-stage entropy curriculum instead of constant low entropy (ablation) | AdaThinking-E · OCR-Reasoning | accuracy: 18.7 -> 26.5 | Improved |
| use two-stage entropy curriculum instead of constant low entropy (ablation) | AdaThinking-E · ChartQA | accuracy: 87.8 -> 89.1 | Improved |
| add accuracy and propensity guidance terms to decision feedback (ablation) | AdaThinking-E · ChartQA | thinking ratio: 39.9 -> 24.3 | Improved |
| add accuracy and propensity guidance terms to decision feedback (ablation) | AdaThinking-E · OCR-Reasoning | accuracy: 25.7 -> 26.5 | Improved |
| lower inflection point gamma from 0.6 to 0.4 (beta 10) (ablation) | AdaThinking-E · ChartQA | thinking ratio: 39.3 -> 24.3 | Improved |
| raise smoothing coefficient beta from 5 to 10 (gamma 0.4) (ablation) | AdaThinking-E · OCR-Reasoning | accuracy: 26.1 -> 26.5 | Improved |
| lower propensity threshold Delta_r from 0.625 to 0.375 (ablation) | AdaThinking-E · ChartQA | accuracy: 88.4 -> 89.1 | Improved |
| replace GRPO/DAPO uniform token reward with decoupled entropy reward (ablation) | AdaThinking-E · training dynamics | thinking ratio | Improved |
Closest papers by meaning
- Reasoning with Exploration: An Entropy Perspective (AAAI 2026 · similarity 0.90)
- ARES: Multimodal Adaptive Reasoning via Difficulty-Aware Token-Level Entropy Shaping (ICLR 2026 · similarity 0.90)
- Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning (NeurIPS 2025 · similarity 0.90)
- Learning When to Think: Shaping Adaptive Reasoning in R1-Style Models via Multi-Stage RL (NeurIPS 2025 · similarity 0.89)
- PEAR: Phase Entropy Aware Reward for Efficient Reasoning (ICLR 2026 · similarity 0.89)
- Thinker: Learning to Think Fast and Slow (NeurIPS 2025 · similarity 0.89)
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.
