Research map / Vision-language and multimodal / Visual question answering
Attention in Reasoning: Dataset, Analysis, and Modeling
Shi Chen, Ming Xin Jiang, Jinhui Yang, Qi Zhao
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 AiR-M progressive attention and reasoning supervision | UpDown · GQA test-dev | accuracy: 51.31 -> 53.46 | Improved |
| add AiR-M supervision | BAN · GQA test-dev | accuracy: 50.14 -> 53.36 | Improved |
| supervise attention with PAAN adversarial learning | UpDown · GQA test-dev | accuracy: 51.31 -> 48.03 | Hurt |
| supervise attention with HAN human-like attention | UpDown · GQA test-dev | accuracy: 51.31 -> 49.96 | Hurt |
| add attention supervision with ASM | MUTAN · GQA test-dev | accuracy: 50.78 -> 51.46 | Improved |
| add attention loss (L_alpha) to AiR-M reasoning supervision (ablation) | AiR-M · GQA test-dev | accuracy: 50.01 -> 53.46 | Improved |
| add reasoning-operation loss (L_r) to AiR-M attention supervision (ablation) | AiR-M · GQA test-dev | accuracy: 50.33 -> 53.46 | Improved |
| use multi-step progressive supervision instead of single aggregated attention (ablation) | AiR-M · GQA test-dev | accuracy: 52.84 -> 53.46 | Improved |
| add incorrect-attention supervision on top of correct attention (AiR-C) (ablation) | UpDown · GQA test-dev | accuracy: 52.96 -> 53.74 | Improved |
| add incorrect-attention supervision on top of correct attention (AiR-C) (ablation) | UpDown-360 · IQVA | accuracy: 40.55 -> 41.10 | Improved |
| add incorrect-attention supervision on top of correct attention (AiR-C) (ablation) | UpDown · GQA | AiR-E: 1.74 -> 2.02 | Improved |
| replace model attention with random attention (ablation) | pre-trained UpDown · GQA | accuracy change: -7.39% | Hurt |
Closest papers by meaning
- Cognitive Mirrors: Exploring the Diverse Functional Roles of Attention Heads in LLM Reasoning (NeurIPS 2025 · similarity 0.87)
- Stop When Further Reasoning Won’t Help: Attention-State Adaptive Generation in Reasoning Models (ICML 2026 · similarity 0.87)
- Seeing but Not Believing: Probing the Disconnect Between Visual Attention and Answer Correctness in VLMs (ICLR 2026 · similarity 0.86)
- Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas (ICML 2025 · similarity 0.86)
- Caption This, Reason That: VLMs Caught in the Middle (NeurIPS 2025 · similarity 0.86)
- FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning (ICLR 2026 · similarity 0.86)
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.
