Research map / Vision-language and multimodal / CLIP and zero-shot
Parameter-efficient Tuning of Large-scale Multimodal Foundation Model
Haixin Wang, Xinlong Yang, Jianlong Chang, Dian Jin, Jinan Sun, Shikun Zhang, Xiao Luo, Qi Tian
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 |
|---|---|---|---|
| remove informative context enhancement (Aurora w/o C) (ablation) | Aurora · Flickr30K / DiDemo / MSCOCO | recall | Hurt |
| replace context with random vectors (Aurora w. R) (ablation) | Aurora · cross-modal retrieval | recall | Hurt |
| use uniform average of queries instead of adaptive weights (Aurora w. M) (ablation) | Aurora · cross-modal retrieval | recall | Hurt |
| remove gated query transformation (Aurora w/o Q) (ablation) | Aurora · MSCOCO / DiDemo | recall | Hurt |
| initialize gated transformation with ones instead of zeros (Aurora w. O) (ablation) | Aurora · MSCOCO / DiDemo | recall | Hurt |
| share parameters across multimodal encoder layers (ablation) | Aurora · FLICKR30K | recall: average improvement of 0.1 | Improved |
| increase CP decomposition rank R (ablation) | Aurora · Flickr30K / DiDemo | recall | Improved |
| increase rank from r=64 to r=128 (ablation) | Aurora · MSCOCO I->T | R@1: 80.2 -> 80.7 | Improved |
| increase rank from r=64 to r=128 (ablation) | Aurora · FLICKR30K I->T | R@1: 96.8 -> 97.2 | Improved |
| Aurora r=64 instead of UniAdapter r=512 on video retrieval | UniAdapter (r=512) · MSR-VTT | R@1: 50.6 -> 52.4 | Improved |
| Aurora r=64 instead of UniAdapter r=512 on VQAv2 test-dev | UniAdapter (r=512) · VQAv2 test-dev | accuracy: 75.44 -> 77.69 | Improved |
| Aurora r=64 instead of LoRA in zero-shot retrieval | BLIP + LoRA · DiDemo zero-shot | R@1: 43.4 -> 44.6 | Improved |
Closest papers by meaning
- Re-Imagining Multimodal Instruction Tuning: A Representation View (ICLR 2025 · similarity 0.87)
- A Wander Through the Multimodal Landscape: Efficient Transfer Learning via Low-rank Sequence Multimodal Adapter (AAAI 2025 · similarity 0.86)
- ReconBoost: Boosting Can Achieve Modality Reconcilement (ICML 2024 · similarity 0.86)
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning (NeurIPS 2025 · similarity 0.86)
- When to Align, When to Predict: A Phase Diagram for Multimodal Learning (NeurIPS 2026 · similarity 0.86)
- Improving Multimodal Learning Balance and Sufficiency through Data Remixing (ICML 2025 · 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.
