Research map / Label-efficient and robust learning / Recognition and long-tail
FANoise: Singular Value-Adaptive Noise Modulation for Robust Multimodal Representation Learning
Jiaoyang Li, Jun Fang, Tianhao Gao, Xiaohui Zhang, Zhiyuan Liu, Chao Liu, Pengzhang Liu, Qixia Jiang
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 FANoise sublinear noise | VLM2Vec Phi3.5-V · MMEB (36 datasets) | overall Precision@1: 60.1 -> 60.8 | Improved |
| add FANoise sublinear noise | VLM2Vec Qwen2-VL-2B · MMEB (36 datasets) | overall Precision@1: 60.1 -> 61.1 | Improved |
| add FANoise sublinear noise | VLM2Vec LLaVA-1.6-LR · MMEB (36 datasets) | overall Precision@1: 55.0 -> 59.2 | Improved |
| add FANoise sublinear noise | VLM2Vec LLaVA-1.6-HR · MMEB (36 datasets) | overall Precision@1: 62.9 -> 66.4 | Improved |
| add FANoise sublinear noise | VLM2Vec Qwen2-VL-7B · MMEB (36 datasets) | overall Precision@1: 65.8 -> 66.6 | Improved |
| add FANoise sublinear noise | VLM2Vec LLaVA-1.6-LR · MMEB VQA | VQA score: 50.3 -> 43.1 | Hurt |
| add FANoise sublinear noise | VLM2Vec Qwen2-VL-2B · MMEB grounding | Grounding score: 73.4 -> 72.3 | Hurt |
| add uniform scaling noise (ablation) | VLM2Vec Qwen2-VL-2B · MMEB (small-budget study) | overall Precision@1: 60.06 -> 60.93 | Improved |
| scale noise linearly with singular values (ablation) | VLM2Vec Qwen2-VL-2B · MMEB (small-budget study) | overall Precision@1: 60.06 -> 60.69 | Improved |
| scale noise sublinearly (sqrt) rather than uniformly (ablation) | VLM2Vec Qwen2-VL-2B · MMEB (small-budget study) | overall Precision@1: 60.93 -> 61.08 | Improved |
| inject noise at output layer instead of input layer (ablation) | VLM2Vec Qwen2-VL-2B · MMEB (noise position study) | overall Precision@1: 48.46 -> 48.73 | Improved |
| increase noise strength alpha too high (ablation) | FANoise_ss · MMEB (noise strength study) | overall Precision@1 | Hurt |
Closest papers by meaning
- Embracing Unimodal Aleatoric Uncertainty for Robust Multimodal Fusion (CVPR 2024 · similarity 0.85)
- Asymmetric Reinforcing Against Multi-Modal Representation Bias (AAAI 2025 · similarity 0.84)
- Aligning Multimodal Representations through an Information Bottleneck (ICML 2025 · similarity 0.84)
- Unity by Diversity: Improved Representation Learning for Multimodal VAEs (NeurIPS 2024 · similarity 0.84)
- Explore How to Inject Beneficial Noise in MLLMs (AAAI 2026 · similarity 0.84)
- Understanding the Robustness of Multi-modal Contrastive Learning to Distribution Shift (ICLR 2024 · similarity 0.84)
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.
