Research map / Vision-language and multimodal / Multimodal LLMs
Information-Regularized Attention for Visual-Centric Reasoning
Guohao Sun, Xiaofang Wang, Yash Patel, Mengchen Liu, Zhiqiang Tao, Praveen Krishnan
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 IRA during SFT | InternVL2-8B · MME | score: 1792 -> 1959 | Improved |
| add IRA during SFT | InternVL2-8B · MMStar | accuracy: 60.2 -> 58.8 | Hurt |
| add IRA during SFT | InternVL2.5-8B · MMMU | accuracy: 46.4 -> 47.6 | Improved |
| add IRA during SFT | LLaVA-OneVision-8B · OK-VQA | accuracy: 41.7 -> 47.4 | Improved |
| add IRA during SFT | InternVL2-8B · MuirBench | accuracy: 38.3 -> 36.8 | Hurt |
| add IRA during SFT | InternVL2.5-8B · VLM-are-blind | accuracy: 33.9 -> 37.3 | Improved |
| add uncertainty-weighted KL (ablation) | IRA · Ablation average (InternVL2-8B) | average score: 60.7 -> 61.3 | Improved |
| add data-dependent prior centered on visual embedding (ablation) | IRA · Ablation average (InternVL2-8B) | average score: 60.6 -> 61.3 | Improved |
| add IRA vs vanilla SFT (ablation) | InternVL2-8B · Ablation average | average score: 59.5 -> 61.3 | Improved |
| apply IRA to layers 60%-80% rather than all layers 0%-100% (ablation) | InternVL2.5-8B · Layer-depth study | Improve (%): -0.93% -> +0.58% | Improved |
| apply IRA to layers 20%-80% instead of 60%-80% (ablation) | InternVL2.5-8B · Layer-depth study | Improve (%): +0.58% -> -2.08% | Hurt |
| apply IRA to layers 50%-70% instead of 60%-80% (ablation) | InternVL2.5-8B · Layer-depth study | Improve (%): +0.51% -> +0.58% | No clear effect |
Closest papers by meaning
- Mitigating Mask Prior Drift and Positional Attention Collapse in Large Diffusion Vision-Language Models (ICML 2026 · similarity 0.90)
- Large Vision-Language Models Get Lost in Attention (ICML 2026 · similarity 0.90)
- Predictive Regularization Against Visual Representation Degradation in Multimodal Large Language Models (CVPR 2026 · similarity 0.90)
- Lifting the Veil on Visual Information Flow in MLLMs: Unlocking Pathways to Faster Inference (CVPR 2025 · similarity 0.89)
- Visual Enhanced Depth Scaling for Multimodal Latent Reasoning (NeurIPS 2026 · similarity 0.89)
- Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-training (ICLR 2026 · 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.
