Research map / Label-efficient and robust learning / Self-supervised and contrastive
The Pursuit of Human Labeling: A New Perspective on Unsupervised Learning
Artyom Gadetsky, Maria Brbic
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 transductive instead of inductive setting (DINO) | HUME · STL-10 | ACC: 90.8 -> 93.2 | Improved |
| use DINOv2 instead of BiT as the regularizing representation (inductive) (ablation) | HUME · CIFAR-100-20 | ACC: 47.8 -> 55.5 | Improved |
| use CLIP instead of BiT as the second representation (inductive) (ablation) | HUME · STL-10 | ACC: 87.5 -> 90.2 | Improved |
| compare unsupervised HUME with supervised linear probe | HUME · CIFAR-100-20 | ACC: 72.5 -> 55.5 | Hurt |
| compare unsupervised HUME with supervised linear probe | HUME · STL-10 | ACC: 88.9 -> 93.2 | Improved |
| replace SPICE with HUME | deep clustering · STL-10 | ACC: 86.2 -> 90.8 | Improved |
| replace SPICE with HUME | deep clustering · CIFAR-100-20 | ACC: 46.8 -> 55.5 | Improved |
| replace best baseline with HUME | deep clustering · ImageNet-1000 | ACC: 41.1 -> 51.1 | Improved |
| use HUME reliable samples instead of random ground truth labels | FreeMatch · CIFAR-10, 1 label/class | accuracy: 83.3 -> 91.2 | Improved |
| use HUME reliable samples instead of random ground truth labels | FreeMatch · STL-10, 100 labels/class | accuracy: 94.0 -> 91.3 | Hurt |
| use HUME reliable samples instead of ground truth labels | FreeMatch · CIFAR-10, 400 labels/class | accuracy: 95.7 -> 94.3 | Hurt |
| use MOCO instead of SimCLR representation (ablation) | HUME · STL-10 | ACC: 86.9 -> 90.8 | Improved |
Closest papers by meaning
- Towards Good Practices for Efficiently Annotating Large-Scale Image Classification Datasets (CVPR 2021 · similarity 0.87)
- Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning (CVPR 2022 · similarity 0.87)
- Labelling unlabelled videos from scratch with multi-modal self-supervision (NeurIPS 2020 · similarity 0.87)
- Learning Rare Category Classifiers on a Tight Labeling Budget (ICCV 2021 · similarity 0.87)
- Masked Unsupervised Self-training for Label-free Image Classification (ICLR 2023 · similarity 0.87)
- Navigating Real-World Partial Label Learning: Unveiling Fine-Grained Images with Attributes (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.
