Research map / Vision-language and multimodal / Cross-modal learning
Towards Understanding Modality Interaction in Multimodal Language Models via Partial Information Decomposition
Wanlong Fang, Tianle Zhang, Wen Tao, Alvin Chan
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 PID-guided sample reweighting instead of uniform LoRA | LoRA on Qwen2.5-VL-7B · MMBench | accuracy: 89.1 -> 90.2 | Improved |
| use PID-guided sample reweighting instead of uniform LoRA | LoRA on Qwen2.5-VL-7B · MMStar | accuracy: 62.0 -> 64.3 | Improved |
| use PID-guided sample reweighting instead of uniform LoRA | LoRA on Qwen2.5-VL-7B · POPE | accuracy: 87.2 -> 88.5 | Improved |
| use PID-guided sample reweighting instead of uniform LoRA | LoRA on Qwen2.5-VL-7B · MMMU | accuracy: 54.0 -> 53.5 | Hurt |
| use PID-guided sample reweighting instead of uniform LoRA | LoRA on Qwen2.5-VL-7B · PMC-VQA | accuracy: 53.0 -> 52.7 | Hurt |
| use PID-guided reweighting instead of random reweighting (ablation) | LoRA on Qwen2.5-VL-7B · MMStar | accuracy: 61.5 -> 64.3 | Improved |
| use PID-guided reweighting instead of accuracy-based mining (ablation) | LoRA on Qwen2.5-VL-7B · MMStar | accuracy: 62.5 -> 64.3 | Improved |
| use PID-guided reweighting instead of modality-ablation selection (ablation) | LoRA on Qwen2.5-VL-7B · MMStar | accuracy: 63.0 -> 64.3 | Improved |
| PID-guided reweighting on LLaVA-OV-7B | LoRA · MMStar | accuracy: 60.0 -> 61.5 | Improved |
| PID-guided reweighting on Gemma3-12B | LoRA · MMStar | accuracy: 56.2 -> 54.3 | Hurt |
| PID-guided reweighting on Gemma3-12B | LoRA · MMMU | accuracy: 59.3 -> 54.1 | Hurt |
| PID-guided reweighting on Gemma3-12B | LoRA · MMBench | accuracy: 83.1 -> 83.3 | No clear effect |
Closest papers by meaning
- XModBench: Benchmarking Cross-Modal Capabilities and Consistency in Omni-Language Models (ICLR 2026 · similarity 0.90)
- A Comprehensive Information-Decomposition Analysis of Large Vision-Language Models (ICLR 2026 · similarity 0.90)
- Compose and Fuse: Revisiting the Foundational Bottlenecks in Multimodal Reasoning (ICLR 2026 · similarity 0.90)
- Diagnosing and Mitigating Modality Interference in Multimodal Large Language Models (NeurIPS 2026 · similarity 0.89)
- Seeing is Understanding: Unlocking Causal Attention into Modality-Mutual Attention for Multimodal LLMs (ICML 2026 · similarity 0.89)
- Measuring Cross-Modal Synergy: A Benchmark for VLM Explainability (NeurIPS 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.
