Research map / Architectures and efficiency / RNNs and normalization
Memory Efficient Meta-Learning with Large Images
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 |
|---|---|---|---|
| Train Simple CNAPs with LITE on 224 images using EfficientNet-B0 rather than 84 images using the same backbone | Simple CNAPs on ORBIT · ORBIT clean videos | frame accuracy: 70.3 (2.1) -> 82.7 (1.7) | Helped |
| Train Simple CNAPs with LITE on large images (ablation) | Simple CNAPs on VTAB+MD · VTAB+MD MD-v2 | accuracy: 68.4 -> 73.9 | Helped |
| Use LITE with 224 images rather than Simple CNAPs without LITE on 84 images (ablation) | Simple CNAPs on VTAB+MD · VTAB+MD VTAB (all) | accuracy: 44.7 -> 51.4 | Helped |
| Increase the LITE back-propagated support subset from |H|=1 to |H|=40 (ablation) | Simple CNAPs on VTAB+MD · VTAB+MD MD-v2 | accuracy: 72.8 -> 73.9 | Helped |
| Increase the LITE back-propagated support subset from |H|=1 to |H|=40 (ablation) | Simple CNAPs on VTAB+MD · VTAB+MD VTAB (all) | accuracy: 51.2 -> 51.4 | No clear effect |
| Use full support gradients instead of LITE with |H|=40 (ablation) | Simple CNAPs on VTAB+MD · VTAB+MD VTAB (all), 84 images | accuracy: 42.7 -> 44.7 | Helped |
| Use LITE with EfficientNet-B0 instead of Simple CNAPs without LITE (ablation) | Simple CNAPs on ORBIT · ORBIT clutter videos | video accuracy: 69.9 (2.3) -> 71.9 (2.3) | Helped |
| Use LITE with EfficientNet-B0 instead of Simple CNAPs without LITE (ablation) | Simple CNAPs on ORBIT · ORBIT clutter videos | frame accuracy: 65.6 (1.9) -> 65.6 (1.9) | No clear effect |
| Use LITE with EfficientNet-B0 instead of Simple CNAPs without LITE (ablation) | Simple CNAPs on ORBIT · ORBIT clean videos | video accuracy: 88.3 (2.1) -> 91.8 (1.8) | Helped |
| Use LITE with EfficientNet-B0 instead of Simple CNAPs without LITE (ablation) | Simple CNAPs on ORBIT · ORBIT clean videos | frame accuracy: 81.4 (1.8) -> 82.7 (1.7) | Helped |
| Use 224 images instead of 84 images | Meta-learners on ORBIT · ORBIT | classification accuracy | Helped |
| Replace the ResNet-18 feature extractor with EfficientNet-B0 (ablation) | Simple CNAPs · ORBIT | classification accuracy | Helped |
Closest papers by meaning
- A Unified Few-Shot Classification Benchmark to Compare Transfer and Meta Learning Approaches (NeurIPS 2021 · similarity 0.89)
- Meta-Learning without Memorization (ICLR 2020 · similarity 0.89)
- A Dual Attention Network with Semantic Embedding for Few-Shot Learning (AAAI 2019 · similarity 0.88)
- Meta-Transfer Learning for Few-Shot Learning (CVPR 2019 · similarity 0.88)
- One Meta-tuned Transformer is What You Need for Few-shot Learning (ICML 2024 · similarity 0.87)
- Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · 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.