Research map / Label-efficient and robust learning / Noisy labels
Reliable Few-Shot Learning Under Dual Noises
Ji Zhang, Jingkuan Song, Lianli Gao, Nicu Sebe, Heng Tao Shen
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 DETA++ to six task-adaptation baselines | TSA, eTT, full-FT (MoCo, CLIP, DeiT, SwinT) · vanilla Meta-Dataset (ID-denoising) | avg accuracy: +2.1 to +5.2 | Improved |
| add DETA++ to six baselines under OOD noise | six baselines · OOD-polluted Meta-Dataset | avg accuracy: +2.7 to +6.6 | Improved |
| add DETA++ instead of DETA | full-FT SwinT · vanilla Meta-Dataset | avg accuracy: 85.4 -> 86.5 | Improved |
| add DETA++ to TSA in standard protocol | TSA (ImageNet-only) · Meta-Dataset standard | avg accuracy: 70.7 -> 73.6 | Improved |
| add CoRA and clean prototype loss (ablation) | baselines with NCC · Meta-Dataset 30% OOD | avg accuracy: 78.0 -> 79.8 (ID), 67.6 -> 69.2 (OOD) | Improved |
| add LocalNCC (ablation) | baselines with NCC · Meta-Dataset 30% OOD | avg accuracy: 79.8 -> 80.4 (ID), 69.2 -> 70.2 (OOD) | Improved |
| add noise entropy maximization loss (ablation) | baselines with NCC · Meta-Dataset 30% OOD | avg accuracy: 80.4 -> 81.2 (ID), 70.2 -> 71.6 (OOD) | Improved |
| add IntraSwap (ablation) | DETA++ · Meta-Dataset 30% OOD | avg accuracy: 81.2 -> 81.7 (ID), 71.6 -> 72.1 (OOD) | Improved |
| vary noise-loss coefficient beta from 0.01 to 0.3 (ablation) | DETA++ on full-FT MoCo · Meta-Dataset 30% OOD | avg accuracy: 75.5 -> 77.8 (ID), 63.5 -> 68.0 (OOD) | Improved |
| increase beta from 0.3 to 1.0 (ablation) | DETA++ on full-FT MoCo · Meta-Dataset 30% OOD | avg accuracy: 77.8 -> 75.8 (ID) | Hurt |
| set region-filter threshold to 0.3 instead of 1.0 (ablation) | DETA++ on full-FT MoCo · Meta-Dataset 30% OOD | avg accuracy: 75.4 -> 77.8 (ID), 65.5 -> 68.0 (OOD) | Improved |
| use region size 112 instead of 48 (224x224 images) (ablation) | DETA++ on full-FT MoCo · Meta-Dataset 30% OOD | avg accuracy: 74.0 -> 77.8 (ID), 63.2 -> 68.0 (OOD) | Improved |
Closest papers by meaning
- Leveraging Normalization Layer in Adapters with Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.85)
- Task-Adaptive Prompted Transformer for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.84)
- Noisy Test-Time Adaptation in Vision-Language Models (ICLR 2025 · similarity 0.84)
- Dual-Level Curriculum Meta-Learning for Noisy Few-Shot Learning Tasks (AAAI 2024 · similarity 0.84)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · similarity 0.83)
- Proxy Denoising for Source-Free Domain Adaptation (ICLR 2025 · similarity 0.83)
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.
