Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Deep Learning via Affinity Diffusion
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 strongly connected subgraphs via affinity diffusion instead of plain k-NN groups (ablation) | PAD · CIFAR100 (kNN/FC) | accuracy (%): 34.5 -> 48.7 | Helped |
| use strongly connected subgraphs instead of k-NN groups (ablation) | PAD · CIFAR10 (kNN/FC) | accuracy (%): 77.5 -> 81.5 | Helped |
| add cyclic constraint in subgraph search (ablation) | PAD · CIFAR100 (kNN/FC) | accuracy (%): 30.6 -> 48.7 | Helped |
| add cyclic constraint in subgraph search (ablation) | PAD · SVHN (kNN/FC) | accuracy (%): 91.1 -> 91.2 | No clear effect |
| add scale (size) threshold on subgraphs (ablation) | PAD · CIFAR10 (kNN/FC) | accuracy (%): 20.2 -> 81.5 | Helped |
| add hard positive enhancement loss (ablation) | PAD · CIFAR10 (kNN/FC) | accuracy (%): 69.8 -> 81.5 | Helped |
| use KL divergence rather than feature cosine similarity for HPE (ablation) | PAD · CIFAR100 (kNN/FC) | accuracy (%): 39.1 -> 48.7 | Helped |
| increase SCS max size s from 10 to 100 (ablation) | PAD · CIFAR10 (kNN/FC) | accuracy (%): 81.5 -> 73.5 | Hurt |
| vary graph density k (1,3,5,10) (ablation) | PAD · CIFAR10 (kNN/FC) | accuracy (%): 78.3, 79.3, 81.5, 79.7 | No clear effect |
| compare PAD against IIC on STL10 clustering | PAD · STL10 clustering | clustering accuracy (%): 59.8 -> 46.5 | Hurt |
Closest papers by meaning
- Unsupervised Deep Learning by Neighbourhood Discovery (ICML 2019 · similarity 0.91)
- Unsupervised Learning of Discriminative Attributes and Visual Representations (CVPR 2016 · similarity 0.89)
- Joint Unsupervised Learning of Deep Representations and Image Clusters (CVPR 2016 · similarity 0.88)
- Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning (CVPR 2022 · similarity 0.88)
- Beyond Words: Augmenting Discriminative Richness via Diffusions in Unsupervised Prompt Learning (CVPR 2025 · similarity 0.87)
- An Efficient Framework for Enhancing Discriminative Models via Diffusion Techniques (AAAI 2025 · 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.