Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Feature Learning via Non-Parametric Instance Discrimination
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 |
|---|---|---|---|
| Replace parametric class weights with stored instance features | Instance discrimination on CIFAR-10 · CIFAR-10 | top-1 accuracy (nearest neighbor, %): 63.0 -> 80.8 | Helped |
| Approximate non-parametric softmax using NCE with m=1 negatives (ablation) | Non-Param Softmax · CIFAR-10 | top-1 accuracy (nearest neighbor, %): 80.8 -> 42.5 | Hurt |
| Increase the NCE negative count from m=1 to m=4096 (ablation) | Non-Param Softmax · CIFAR-10 | top-1 accuracy (nearest neighbor, %): 42.5 -> 80.4 | Helped |
| Use ResNet-50 instead of AlexNet as the backbone | Instance discrimination · ImageNet | top-1 accuracy (conv5 linear, %): 35.6 -> 54.0 | Helped |
| Increase the embedding size from 32 to 128 (ablation) | Instance discrimination with ResNet-18 · ImageNet | top-1 accuracy (%): 34.0 -> 41.0 | Helped |
| Increase the embedding size from 128 to 256 (ablation) | Instance discrimination with ResNet-18 · ImageNet | top-1 accuracy (%): 41.0 -> 40.1 | Hurt |
| Increase the training data fraction from 0.1% to 100% (ablation) | Instance discrimination with ResNet-18 · ImageNet | accuracy (%): 3.9 -> 41.0 | Helped |
| Add proximal regularization to stabilize training (ablation) | NCE instance discrimination · ImageNet | learned representation performance | Helped |
Closest papers by meaning
- Train a One-Million-Way Instance Classifier for Unsupervised Visual Representation Learning (AAAI 2021 · similarity 0.91)
- Parametric Instance Classification for Unsupervised Visual Feature learning (NeurIPS 2020 · similarity 0.89)
- Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning (CVPR 2022 · similarity 0.88)
- Unsupervised Feature Learning by Cross-Level Instance-Group Discrimination (CVPR 2021 · similarity 0.88)
- Unsupervised Learning of Discriminative Attributes and Visual Representations (CVPR 2016 · similarity 0.88)
- Self-Organizing Visual Prototypes for Non-Parametric Representation Learning (ICML 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.