Research map / Label-efficient and robust learning / Recognition and long-tail
Mitigating Embedding and Class Assignment Mismatch in Unsupervised Image Classification
Sungwon Han, Sungwon Park, Sungkyu Park, Sundong Kim, Meeyoung Cha
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 |
|---|---|---|---|
| add consistency preserving loss Lcp (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 58.2 -> 81.0 | Improved |
| add weight normalization in the classifier (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 56.0 -> 81.0 | Improved |
| add temperature tau_c<1 in softmax (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 19.5 -> 81.0 | Improved |
| use Super-AND instead of AND for stage-1 embedding (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 73.9 -> 81.0 | Improved |
| use Super-AND instead of Lemniscate for stage-1 embedding (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 63.8 -> 81.0 | Improved |
| add pretrained Super-AND initialization vs random (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 58.6 -> 81.0 | Improved |
| replace k-means with second-stage refinement (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 38.3 -> 81.0 | Improved |
| replace hierarchical clustering with second-stage refinement (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 49.9 -> 81.0 | Improved |
| separate embedding pretraining and class assignment into two stages instead of joint (ablation) | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 68.4 -> 81.0 | Improved |
| use two-stage model instead of IIC | Two-stage model · CIFAR-10 (unsupervised) | accuracy: 61.7 -> 81.0 | Improved |
| use two-stage model instead of IIC | Two-stage model · CIFAR-100 | accuracy: 20.3 -> 28.1 | Improved |
| add our pretraining to Mean Teacher | Mean Teacher · CIFAR-10 (4000 labels) | accuracy: 84.2 -> 88.0 | Improved |
Closest papers by meaning
- Augmenting Supervised Neural Networks with Unsupervised Objectives for Large-scale Image Classification (ICML 2016 · similarity 0.87)
- Unsupervised Deep Learning by Neighbourhood Discovery (ICML 2019 · similarity 0.87)
- Unsupervised Deep Learning via Affinity Diffusion (AAAI 2020 · similarity 0.87)
- Learning Robust Visual-Semantic Embeddings (ICCV 2017 · similarity 0.86)
- Unsupervised Learning for Class Distribution Mismatch (ICML 2025 · similarity 0.86)
- On the Provable Advantage of Unsupervised Pretraining (ICLR 2024 · similarity 0.84)
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.
