Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Latent Representations in Neural Networks for Clustering through Pseudo Supervision and Graph-based Activity Regularization
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 |
|---|---|---|---|
| Use the proposed pseudo-supervision and ACOL/GAR approach instead of IMSAT | Unsupervised clustering · SVHN-test | ACC: 57.30%(± 3.90) -> 76.80%(± 1.30) | Helped |
| Use the proposed pseudo-supervision and ACOL/GAR approach instead of JULE | Unsupervised clustering · USPS-full | ACC: 95.00% -> 96.51%(± 0.26) | Helped |
| Use the proposed pseudo-supervision and ACOL/GAR approach instead of IMSAT | Unsupervised clustering · MNIST-test | ACC: 98.40%(± 0.40) -> 98.32%(± 0.08) | No clear effect |
| Use four rotation transformations instead of all eight (ablation) | Our approach · MNIST-test | ACC: 98.32%(± 0.08) -> 97.80%(± 0.18) | Hurt |
| Use only T1 and T2 rotations instead of all eight transformations (ablation) | Our approach · MNIST-test | ACC: 98.32%(± 0.08) -> 72.52%(± 6.20) | Hurt |
| Use only T1 and T3 rotations instead of all eight transformations (ablation) | Our approach · MNIST-test | ACC: 98.32%(± 0.08) -> 96.84%(± 0.29) | Hurt |
| Use T1 and T3 rather than T1 and T2 (ablation) | Our approach · MNIST-test | ACC: 72.52%(± 6.20) -> 96.84%(± 0.29) | Helped |
| Cluster the Z representation instead of F (ablation) | Our approach · MNIST-test | ACC: 98.32%(± 0.08) -> 98.16%±(0.14) | No clear effect |
| Use scaling or shearing transformations to make pseudo parent-classes (ablation) | Our approach · MNIST | clustering quality | Hurt |
| Use translation or random pixel permutation transformations to make pseudo parent-classes (ablation) | Our approach · MNIST | clustering quality | Hurt |
| Increase the number of rotation-based transformations (ablation) | Our approach · MNIST | ACC: 72.52%(± 6.20) -> 96.84%(± 0.29) -> 97.80%(± 0.18) -> 98.32%(± 0.08) | Helped |
| Reduce F dimensionality up to approximately 1024 (ablation) | Our approach · MNIST | clustering accuracy | No clear effect |
Closest papers by meaning
- Semi-Supervised Learning via Compact Latent Space Clustering (ICML 2018 · similarity 0.87)
- Unsupervised Clustering using Pseudo-semi-supervised Learning (ICLR 2020 · similarity 0.87)
- Unsupervised Action Segmentation by Joint Representation Learning and Online Clustering (CVPR 2022 · similarity 0.85)
- Deep Subspace Clustering Networks (NeurIPS 2017 · similarity 0.85)
- Unsupervised Deep Embedding for Clustering Analysis (ICML 2016 · similarity 0.85)
- Clustering Properties of Self-Supervised Learning (ICML 2025 · 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.