Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Cross-Domain Few-Shot Learning With Task-Specific Adapters
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 residual matrix adapters and pre-classifier alignment to URL | URL on Meta-Dataset · Meta-Dataset, Average All | accuracy: 76.6 -> 79.5 | Helped |
| Add residual matrix adapters and pre-classifier alignment to URL | URL on Meta-Dataset · Meta-Dataset, Average Unseen | accuracy: 70.6 -> 78.1 | Helped |
| Add residual matrix adapters and pre-classifier alignment to an ImageNet-trained ResNet-18 | SDL-ResNet-18 · Meta-Dataset, Average All | accuracy: 61.4 -> 70.7 | Helped |
| Add residual matrix adapters and pre-classifier alignment to an ImageNet-trained ResNet-34 | SDL-ResNet-34 · Meta-Dataset, Average All | accuracy: 74.3 -> 74.9 | Helped |
| Learn adapter weights directly on the support set instead of predicting them with an auxiliary network | Task-specific adapters · Meta-Dataset, unseen domains | accuracy | Helped |
| Use residual rather than serial adapter connections (ablation) | Task-specific adapters · Meta-Dataset, especially unseen domains | accuracy | Helped |
| Use matrix rather than channelwise adapter parameterization (ablation) | Directly learned residual adapters · Meta-Dataset | accuracy | Mixed |
| Add pre-classifier alignment to residual adapters (ablation) | Ad-R-M · Meta-Dataset | accuracy | Helped |
| Add adapters to more backbone blocks rather than only the last block (ablation) | Residual-adapter model · Meta-Dataset, unseen domains | accuracy | Helped |
| Decompose large residual adapter matrices with N=32 (ablation) | Residual-adapter model · Meta-Dataset | accuracy: around 4% vs 13% | No clear effect |
| Optimize task-specific parameters for 20 rather than 40 iterations (ablation) | MDL-based adapter model · Meta-Dataset, multiple domains | accuracy | Mixed |
| Use residual adapters in five-way one-shot tasks | URL · Meta-Dataset, five-way one-shot, Average Seen | accuracy: 73.4 -> 73.5 | No clear effect |
Closest papers by meaning
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.90)
- Looking Wider for Better Adaptive Representation in Few-Shot Learning (AAAI 2021 · similarity 0.89)
- Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Data (NeurIPS 2021 · similarity 0.89)
- Leveraging Normalization Layer in Adapters with Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.89)
- Task-Adaptive Prompted Transformer for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.88)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.87)
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.