Research map / Graph learning / GNN architectures
Wasserstein Graph Neural Networks for Graphs With Missing Attributes
Zhixian Chen, Tengfei Ma, Yangqiu Song, Yang Wang
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 low-dimensional SVD distributional embedding instead of high-dimensional attribute distributions (ablation) | WGNN_prob · synthetic graph, high missing rates | accuracy | Improved |
| use SVD latent factors instead of random orthogonal latent factors (ablation) | WGNN_rand · synthetic graph, all missing settings | accuracy | Improved |
| add low-dimensional embedding instead of direct high-dimensional distributions (ablation) | WGNN_hd · Cora, r_p=0.1 | accuracy: 0.439 -> below 0.6029 GCN_nofeat | Improved |
| use Wasserstein aggregation instead of GCN aggregation on U_k (ablation) | GCN_U · Cora, partially missing | accuracy | Improved |
| use Wasserstein aggregation instead of Mean aggregation on U_k (ablation) | MEAN_U · Cora, partially missing | accuracy | Mixed |
| add Euclidean embedding with V_k (Lambda_k V_k^T) before prediction (ablation) | WGNN_svd,U · Cora, partially missing | accuracy | Improved |
| add Wasserstein aggregation with SVD embedding at highest entirely missing level | GCN and other imputation baselines · Cora/Citeseer/PubMed, r_e=0.9 | accuracy: at least 6%, 10% and 9% higher accuracy | Improved |
| add Wasserstein aggregation with SVD embedding at highest partially missing level | GCN and other imputation baselines · Cora/Citeseer/PubMed, r_p=0.9 | accuracy: 5%, 8% and 5% higher accuracy | Improved |
| give MIDA 30% more observed values | MIDA baseline · Cora/Citeseer/PubMed | accuracy | No clear effect |
| increase layers L and aggregation steps h (up to L=5, h=9) | WGNN_svd · Cora, r_e=0.9 | accuracy | Improved |
| use Multi-WGNN instead of IGMC for matrix completion | IGMC · Flixster | RMSE: 0.872 -> 0.883 | No clear effect |
| use Multi-WGNN instead of IGMC for matrix completion | IGMC · MovieLens-100K | RMSE: 0.905 -> 0.910 | No clear effect |
Closest papers by meaning
- Learning on Attribute-Missing Graphs (TPAMI 2020 · similarity 0.89)
- Representing Graphs via Gromov-Wasserstein Factorization (TPAMI 2022 · similarity 0.88)
- Wasserstein Embedding for Graph Learning (ICLR 2021 · similarity 0.88)
- Attribute-Missing Graph Clustering Network (AAAI 2024 · similarity 0.86)
- Subgraph Invariant Learning Towards Large-Scale Graph Node Classification (AAAI 2025 · similarity 0.85)
- Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (ICLR 2025 · 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.
