Research map / Label-efficient and robust learning / Domain adaptation and generalization
Boosting the Generalization Capability in Cross-Domain Few-Shot Learning via Noise-Enhanced Supervised Autoencoder
Hanwen Liang, Qiong Zhang, Peng Dai, Juwei Lu
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 classification loss on reconstructed images (SAE -> NSAE), CE+CE (ablation) | NSAE ResNet10 · ISIC 5-shot | accuracy: 52.28 -> 54.05 | Improved |
| add classification loss on reconstructed images (SAE -> NSAE), CE+CE (ablation) | NSAE Conv4 · CUB 5-shot | accuracy: 52.74 -> 55.35 | Improved |
| add classification loss on reconstructed images instead of doubling original-image loss (SAE(*) -> NSAE) (ablation) | NSAE ResNet10 · ISIC 5-shot | accuracy: 52.11 -> 54.05 | Improved |
| use two-step fine-tuning instead of one-step (NSAE(-) -> NSAE), CE+CE (ablation) | NSAE ResNet10 · ISIC 5-shot | accuracy: 53.52 -> 54.05 | Improved |
| pre-train with NSAE instead of a single encoder, CE+CE | Fine-tune baseline · ISIC 5-shot | accuracy: 48.11 -> 54.05 | Improved |
| pre-train with NSAE instead of a single encoder, BSR+CE | BSR baseline · EuroSAT 5-shot | accuracy: 80.89 -> 84.33 | Improved |
| pre-train with NSAE instead of single encoder, CE+CE | Fine-tune baseline · CUB 5-shot | accuracy: 64.14 -> 68.51 | Improved |
| pre-train with NSAE instead of single encoder, CE+CE | Fine-tune baseline · CUB 50-shot | accuracy: 89.61 -> 89.42 | No clear effect |
| add data augmentation and label propagation to NSAE (BSR+CE) | NSAE (BSR+CE) · ISIC 5-shot | accuracy: 55.27 -> 56.85 | Improved |
| add data augmentation and label propagation to NSAE (BSR+CE) | NSAE (BSR+CE) · ChestX 50-shot | accuracy: 38.52 -> 41.80 | Mixed |
| use distance-based fine-tuning instead of CE fine-tuning when pre-training with CE (ablation) | CE pre-training ResNet10 · ISIC 5-shot | accuracy: 51.28 -> 51.62 | Mixed |
| use distance-based fine-tuning instead of CE fine-tuning when pre-training with BSR (ablation) | BSR pre-training ResNet10 · EuroSAT 5-shot | accuracy: 80.89 -> 80.13 | Hurt |
Closest papers by meaning
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.88)
- Switch to Generalize: Domain-Switch Learning for Cross-Domain Few-Shot Classification (ICLR 2022 · similarity 0.88)
- Understanding Cross-Domain Few-Shot Learning Based on Domain Similarity and Few-Shot Difficulty (NeurIPS 2022 · similarity 0.88)
- Reconstruction Target Matters in Masked Image Modeling for Cross-Domain Few-Shot Learning (AAAI 2025 · similarity 0.87)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.87)
- Leveraging Normalization Layer in Adapters with Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning (AAAI 2024 · 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.
