Research map / Graph learning / Graph self-supervised learning
Robust Graph Structure Learning via Multiple Statistical Tests
Yaohua Wang, Fangyi Zhang, Ming Lin, Senzhang Wang, Xiuyu Sun, Rong Jin
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 multiple-test AUC similarity instead of single-test similarity as k grows from 5 to 40 | Sim-M vs Sim-S on MS-Celeb Part1 · MS-Celeb Part1, k=5 to k=40 | AUC: 96.80 -> 97.59; 96.21 -> 97.72; 95.33 -> 97.21; 94.00 -> 95.66 | Improved |
| Use multiple-test similarity when the neighbourhood contains too much noise | Sim-M on MS-Celeb Part1 · MS-Celeb Part1, k=80 to k=160 | AUC: 92.05 -> 91.32; 90.76 -> 86.58; 90.65 -> 83.90 | Hurt |
| Use multiple-test similarity rather than single-test similarity | Sim-M on MSMT17 · MSMT17, k=5 to k=20 | AUC: 89.23 -> 89.74; 89.33 -> 90.02; 87.78 -> 88.63 | Improved |
| Use multiple-test similarity rather than single-test similarity | Sim-M on VeRi-776 · VeRi-776, k=5 to k=40 | AUC: 90.96 -> 90.40; 89.34 -> 90.09; 85.86 -> 86.54; 83.05 -> 82.89 | Mixed |
| Add A_band attention to the GCN | GCN · MS-Celeb Part1, MSMT17, VeRi-776 | mAP: 90.15 -> 95.97; 83.51 -> 86.04; 85.27 -> 87.59 | Improved |
| Use the proposed learned graph features with Infomap rather than original features | Infomap clustering · MS-Celeb Part1 | Pairwise F-score: 93.91 -> 94.94 | Improved |
| Add Q-attention to self-attention in the combined A_band design (ablation) | GCN w/ A_band on MS-Celeb Part1 · MS-Celeb Part1 | mAP: 88.10 -> 95.97 | Improved |
| Use A_X rather than A_self as the input to Q-attention (ablation) | GCN attention architecture on MS-Celeb Part1 · MS-Celeb Part1 | mAP: 90.10 -> 92.26 | Improved |
| Add learnable W_qart query and key weights (ablation) | A_band on MS-Celeb Part1 · MS-Celeb Part1 | mAP: 93.67 -> 95.97 | Improved |
| Use learned weighted addition rather than unweighted addition to fuse attention maps (ablation) | A_band · MS-Celeb, MSMT17, VeRi-776 | mAP: 92.83 -> 95.97; 84.95 -> 86.04; 85.97 -> 87.59 | Improved |
| Use learned weighted addition rather than element-wise multiplication to fuse attention maps (ablation) | A_band · MS-Celeb, MSMT17, VeRi-776 | mAP: 95.71 -> 95.97; 85.59 -> 86.04; 85.53 -> 87.59 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Ada-NETS — Ada-NETS: Face Clustering via Adaptive Neighbour Discovery in the Structure Space (ICLR 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning MLPs on Graphs: A Unified View of Effectiveness, Robustness, and Efficiency (ICLR 2023 · similarity 0.87)
- No Metric to Rule Them All: Toward Principled Evaluations of Graph-Learning Datasets (ICML 2025 · similarity 0.87)
- Beyond Redundancy: Information-aware Unsupervised Multiplex Graph Structure Learning (NeurIPS 2024 · similarity 0.85)
- Distribution Learning Meets Graph Structure Sampling (NeurIPS 2025 · similarity 0.85)
- Neural Gaussian Similarity Modeling for Differential Graph Structure Learning (AAAI 2024 · similarity 0.85)
- Refining Latent Homophilic Structures over Heterophilic Graphs for Robust Graph Convolution Networks (AAAI 2024 · similarity 0.85)
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.
