Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Diversity Transfer Network for Few-Shot Learning
Read the paperCheck 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 |
|---|---|---|---|
| transfer within-class feature offsets from known categories to support features (ablation) | DTN · miniImageNet 5-way 1-shot | accuracy: 60.14% ± 1.24% -> 63.45% ± 0.86% | Helped |
| transfer within-class feature offsets from known categories (ablation) | DTN · miniImageNet 5-way 5-shot | accuracy: 75.57% ± 0.96% -> 77.91% ± 0.62% | Helped |
| replace Delta-encoder generation with diversity transfer (ablation) | DTN · miniImageNet 5-way 1-shot | accuracy: 60.38% ± 1.12% -> 61.95% ± 0.85% | Helped |
| replace Delta-encoder generation with diversity transfer (ablation) | DTN · miniImageNet 5-way 5-shot | accuracy: 73.44% ± 0.92% -> 76.52% ± 0.65% | Helped |
| use OAT rather than two-stage training (ablation) | DTN · miniImageNet 5-way 1-shot | accuracy: 61.95% ± 0.85% -> 63.45% ± 0.86% | Helped |
| use OAT rather than two-stage training (ablation) | DTN · miniImageNet 5-way 5-shot | accuracy: 76.52% ± 0.65% -> 77.91% ± 0.62% | Helped |
| organize auxiliary and meta-training epochs with OAT rather than stochastic AT (ablation) | DTN · miniImageNet 5-way 1-shot | accuracy: 62.65% ± 0.86% -> 63.45% ± 0.86% | Helped |
| organize auxiliary and meta-training epochs with OAT rather than stochastic AT (ablation) | DTN · miniImageNet 5-way 5-shot | accuracy: 77.12% ± 0.65% -> 77.91% ± 0.62% | Helped |
| add auxiliary task co-training with OAT instead of naive meta-training (ablation) | DTN · miniImageNet 5-way 1-shot | accuracy: 59.81% ± 1.36% -> 63.45% ± 0.86% | Helped |
| add auxiliary task co-training with OAT instead of naive meta-training (ablation) | DTN · miniImageNet 5-way 5-shot | accuracy: 74.97% ± 1.01% -> 77.91% ± 0.62% | Helped |
| increase generated feature count from 2 to 64 (ablation) | DTN · miniImageNet 5-way 1-shot | accuracy: 60.71% (-2.74%) -> 63.45% (-0.00%) | Helped |
| increase generated feature count from 2 to 64 (ablation) | DTN · miniImageNet 5-way 5-shot | accuracy: 75.56% (-2.35%) -> 77.91% (-0.00%) | Helped |
Closest papers by meaning
- Meta-Transfer Learning for Few-Shot Learning (CVPR 2019 · similarity 0.88)
- Diversity-enhancing Generative Network for Few-shot Hypothesis Adaptation (ICML 2023 · similarity 0.87)
- Partial Is Better Than All: Revisiting Fine-tuning Strategy for Few-shot Learning (AAAI 2021 · similarity 0.87)
- A Unified Few-Shot Classification Benchmark to Compare Transfer and Meta Learning Approaches (NeurIPS 2021 · similarity 0.86)
- Meta-Learning for Semi-Supervised Few-Shot Classification (ICLR 2018 · similarity 0.86)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (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.