Research map / Graph learning / GNN architectures
Learn to Propagate Reliably on Noisy Affinity Graphs
Lei Yang, Qingqiu Huang, Huaiyi Huang, Linning Xu, Dahua Lin
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 |
|---|---|---|---|
| increase maximum graph patch size from 100 to 3000 (ablation) | reliable propagation · ImageNet 10K vertices, 50% noise | accuracy: 64.63 -> 65.83 | Improved |
| increase maximum patch size from 3000 to 10000 (ablation) | reliable propagation · ImageNet 10K vertices, 50% noise | accuracy: 65.83 -> 65.92 | No clear effect |
| increase maximum patch size from 3000 to 10000 (cost) (ablation) | reliable propagation · ImageNet 10K vertices, 50% noise | runtime: 316s -> 1077s | Hurt |
| increase maximum patch size from 100 to 3000 (cost) (ablation) | reliable propagation · ImageNet 10K vertices, 50% noise | runtime: 34s -> 316s | Hurt |
| use the proposed confidence inclusion instead of design (1) (ablation) | reliable propagation · ImageNet noisy graph | accuracy: 61.5 -> 65.8 | Improved |
| concatenate confidence to vertex features (design 2) instead of the proposed design (ablation) | reliable propagation · ImageNet noisy graph | accuracy: 65.8 -> 65.9 | No clear effect |
| concatenate confidence to vertex features (design 2) instead of the proposed design (cost) (ablation) | reliable propagation · ImageNet noisy graph | memory: 168M -> 205M | Hurt |
| use estimated confidence to choose samples for annotation (active learning) versus random or GCN-based selection (ablation) | reliable propagation · ImageNet, 1% labels, 50% noise | accuracy | Improved |
| train face model on pseudo labels from the proposed method versus GCN or CDP pseudo labels (ablation) | reliable propagation · MegaFace, Ms-Celeb-1M 1K identities | accuracy | Improved |
| compare proposed method against SGC under 50% noise | reliable propagation · ImageNet noisy graph, noise 50% | accuracy: 65.63 -> 68.25 | Improved |
| compare proposed method against SGC under 0% noise | reliable propagation · ImageNet noisy graph, noise 0% | accuracy: 84.78 -> 85.16 | No clear effect |
| compare proposed method against GCN on Ms-Celeb-1M at 10% noise | reliable propagation · Ms-Celeb-1M noisy graph, noise 10% | accuracy: 99.6 -> 97.59 | Mixed |
Closest papers by meaning
- Affinity-Aware Graph Networks (NeurIPS 2023 · similarity 0.84)
- Learn Beneficial Noise as Graph Augmentation (ICML 2025 · similarity 0.83)
- Learning with Noisy Correspondence (IJCV 2024 · similarity 0.83)
- Training Robust Graph Neural Networks by Modeling Noise Dependencies (NeurIPS 2025 · similarity 0.82)
- Is Noise Conditioning Necessary? A Unified Theory of Unconditional Graph Diffusion Models (NeurIPS 2025 · similarity 0.82)
- Geometric Generative Modeling with Noise-Conditioned Graph Networks (ICML 2025 · similarity 0.81)
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.
