Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Exploring Complementary Strengths of Invariant and Equivariant Representations for Few-Shot Learning
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 |
|---|---|---|---|
| jointly enforce transformation invariance and equivariance rather than invariance alone (ablation) | few-shot representation learning · miniImageNet 5-way 1-shot | accuracy: 64.64±0.80 -> 66.82± 0.80 | Helped |
| jointly enforce transformation invariance and equivariance rather than equivariance alone (ablation) | few-shot representation learning · miniImageNet 5-way 1-shot | accuracy: 66.09±0.80 -> 66.82± 0.80 | Helped |
| add multi-head distillation instead of supervised logit-level distillation (ablation) | joint invariant-equivariant model · miniImageNet 5-way 1-shot | accuracy: 66.95±0.78 -> 67.28±0.80 | Helped |
| increase the transformation set from M=16 to M=20 (ablation) | few-shot representation learning · CIFAR-FS validation 5-way 1-shot | accuracy: 68.20±0.92 -> 68.07±0.90 | Hurt |
| add jointly sampled affine transformations (ablation) | invariant-equivariant representation learning · miniImageNet 5-way 5-shot | accuracy: 82.96±0.53 vs 84.35±0.51 | Hurt |
| add the proposed invariant and equivariant objectives to supervised training (ablation) | few-shot learning · CIFAR-FS 5-way 1-shot | accuracy: 71.50±0.80 -> 76.83±0.82 | Helped |
| add invariant and equivariant objectives (ablation) | few-shot learning · FC100 5-way 1-shot | accuracy: 42.60±0.70 -> 47.38±0.79 | Helped |
| add invariant and equivariant objectives (ablation) | few-shot learning · miniImageNet 5-way 1-shot | accuracy: 62.02±0.63 -> 66.82± 0.80 | Helped |
| add multi-head self-distillation to the proposed model (ablation) | Ours · CIFAR-FS 5-way 1-shot | accuracy: 76.83±0.82 -> 77.87±0.85 | Helped |
| add multi-head self-distillation to the proposed model (ablation) | Ours · FC100 5-way 5-shot | accuracy: 64.43±0.77 -> 65.30±0.76 | Helped |
| add jointly learned invariant and equivariant representations (ablation) | few-shot representation learning · miniImageNet, CIFAR-FS, and FC100 5-way 1-shot | accuracy | Helped |
| add standard auxiliary SSL tasks instead of the proposed joint invariant-equivariant objectives (ablation) | miniImageNet few-shot learning · miniImageNet 5-way 1-shot | accuracy: 66.82± 0.80 vs 65.25±0.80 | Hurt |
Closest papers by meaning
- DiffKendall: A Novel Approach for Few-Shot Learning with Differentiable Kendall's Rank Correlation (NeurIPS 2023 · similarity 0.90)
- Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation (ICLR 2020 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
- Self-supervised learning of Split Invariant Equivariant representations (ICML 2023 · similarity 0.88)
- Meta-Learning With Differentiable Convex Optimization (CVPR 2019 · similarity 0.88)
- Few-shot Implicit Function Generation via Equivariance (CVPR 2025 · 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.