Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Few-Shot Learning With Embedded Class Models and Shot-Free Meta Training
Avinash Ravichandran, Rahul Bhotika, Stefano Soatto
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 |
|---|---|---|---|
| use embedded class models with learned prototypes instead of mean prototypes | Prototypical Networks · miniImageNet 5-way 1-shot (C64) | accuracy (%): 43.88 -> 49.07 | Improved |
| use embedded class models, train with 1-shot and test 5-shot | Prototypical Networks · miniImageNet 5-way (C64) | accuracy (%): 58.33 -> 64.98 | Improved |
| use embedded class models instead of mean prototypes | Prototypical Networks · tieredImageNet 5-way 1-shot (C64) | accuracy (%): 41.36 -> 48.19 | Improved |
| use embedded class models instead of mean prototypes | Prototypical Networks · CIFAR-FS 5-way 1-shot (C64) | accuracy (%): 50.74 -> 55.14 | Improved |
| use embedded class models when train and test shots match at 5 | Prototypical Networks · miniImageNet 5-way 5-shot (C64) | accuracy (%): 65.49 -> 65.73 | No clear effect |
| use embedded class models when train and test shots match at 5 | Prototypical Networks · tieredImageNet 5-way 5-shot (C64) | accuracy (%): 65.51 -> 65.50 | No clear effect |
| lift class identity dimension to 2x the embedding dimension (ablation) | Embedded class models · miniImageNet 5-way 1-shot (C64) | accuracy (%): 49.07 -> 51.46 | Improved |
| increase class identity dimension from 2x to 10x (ablation) | Embedded class models · miniImageNet 5-way 1-shot (C64) | accuracy (%): 51.46 -> 51.32 | No clear effect |
| use 2x class identity dimension (ablation) | Embedded class models (ResNet-12) · tieredImageNet 5-way 1-shot | accuracy (%): 63.99 -> 66.87 | Improved |
| use 2x class identity dimension (ablation) | Embedded class models (ResNet-12) · miniImageNet 5-way | accuracy (%): 1-shot 59.00 -> 60.64; 5-shot 77.46 -> 77.02 | Mixed |
| widen ResNet-12 filters to (64,160,320,640) (ablation) | Embedded class models · miniImageNet 5-way 1-shot | accuracy (%): 59.00 -> 59.04 | No clear effect |
| replace ADAM with SGD (ablation) | Embedded class models (ResNet-12) · miniImageNet 5-way | accuracy (%): 1-shot 59.41 -> 59.00; 5-shot 76.75 -> 77.46; 10-shot 81.33 -> 82.33 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TADAM — TADAM: Task dependent adaptive metric for improved few-shot learning (NeurIPS 2018) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking Few-shot Image Classification: A Good Embedding is All You Need? (ECCV 2020 · similarity 0.90)
- Few-Shot Learning via Embedding Adaptation With Set-to-Set Functions (CVPR 2020 · similarity 0.90)
- Few-Shot Learning With Global Class Representations (ICCV 2019 · similarity 0.90)
- A Closer Look at Few-shot Classification (ICLR 2019 · similarity 0.90)
- Scaling Few-Shot Learning for the Open World (AAAI 2024 · 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)
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.
