Research map / Vision-language and multimodal / Cross-modal learning
Parameter-Efficient Adaptation for MLLMs via Implicit Modality Decomposition
Mingfang Zhang, Yunhong Wang, Lu Wang, Jiaxin Chen
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 both MSDC and MAAC constraints to LoRA (without layer-wise allocation) (ablation) | LoRA · MUSIC-AVQA (LLaMA2) | accuracy: 73.41 -> 76.91 | Improved |
| add both constraints to LoRA (without layer-wise allocation) (ablation) | LoRA · AVE (LLaMA2) | accuracy: 69.84 -> 75.22 | Improved |
| add layer-wise growth of shared parameter ratio (ablation) | IMoD · MUSIC-AVQA (LLaMA2) | accuracy: 76.91 -> 77.31 | Improved |
| add layer-wise growth of shared parameter ratio (ablation) | IMoD · AVE (LLaMA2) | accuracy: 75.22 -> 75.77 | Improved |
| set layer-wise growth rate alpha to 0.4 instead of 0 (ablation) | IMoD · MUSIC-AVQA (LLaMA2) | accuracy: 76.91 -> 77.31 | Improved |
| vary gradient weights lambda1 and lambda2 around optimum (ablation) | IMoD · MUSIC-AVQA | accuracy | No clear effect |
| replace LoRA with IMoD | LoRA on LLaMA2 · MUSIC-AVQA | accuracy: 73.41 -> 77.31 | Improved |
| replace MokA with IMoD | MokA on LLaMA2 · AIR-Bench speech-en foundation | accuracy: 39.64 -> 37.99 | Hurt |
| explicit separate LoRA per modality (Multiple LoRA) instead of one LoRA | LoRA on LLaMA2 · MUSIC-AVQA | accuracy: 73.41 -> 72.66 | Hurt |
| use IMoD instead of MokA for inference efficiency | MokA · MUSIC-AVQA (LLaMA2) | inference latency: 116% -> 100% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MokA — MokA: Multimodal Low-Rank Adaptation for MLLMs (NeurIPS 2025) · better on 14 of 14 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MokA: Multimodal Low-Rank Adaptation for MLLMs (NeurIPS 2025 · similarity 0.91)
- Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs (NeurIPS 2026 · similarity 0.90)
- Explore How to Inject Beneficial Noise in MLLMs (AAAI 2026 · similarity 0.89)
- ModaVerse: Efficiently Transforming Modalities with LLMs (CVPR 2024 · similarity 0.89)
- Learn from Downstream and Be Yourself in Multimodal Large Language Models Fine-Tuning (ICML 2025 · similarity 0.88)
- Diagnosing and Mitigating Modality Interference in Multimodal Large Language Models (NeurIPS 2026 · similarity 0.88)
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.
