Research map / Graph learning / GNN architectures
Leap of FAITH from GNN-to-MLP: Fairness Aware Inference via DisTillation of GrapH Knowledge
Vipul Kumar Singh, Jyotismita Barman, Sandeep Kumar, Tapan K. Gandhi, , Jayadeva
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 |
|---|---|---|---|
| replace GLNN-style soft-label distillation with FAITH (energy alignment + l2,1 fairness) | GNN-to-MLP distillation · Cora (transductive) | individual unfairness IF x10^3: 71.13 -> 12.06 | Improved |
| use FAITH instead of TINED global energy alignment | MLP student, SAGE teacher · Pubmed (transductive) | accuracy (%): 75.51 -> 81.53 | Improved |
| use FAITH instead of KRD | MLP student, Co-author CS · CS (inductive) | accuracy (%): KRD 92.50 vs FAITH 92.74 | Mixed |
| use FAITH instead of KRD in transductive CS | MLP student · CS (transductive) | accuracy (%): 93.99 -> 93.37 | Hurt |
| add neighborhood-guided Dirichlet energy distribution (DED) block (ablation) | FAITH · transductive node classification | accuracy/fairness | Improved |
| add oracle similarity alignment (OSA, l2,1) block (ablation) | FAITH · transductive node classification | accuracy/fairness | Improved |
| add counterfactual invariance group-fairness term (FAITH*) (ablation) | FAITH · Credit and Income | BPC (group+individual fairness balance) | Improved |
| FAITH without group fairness term | FAITH · Credit and Income | group fairness | Hurt |
| use l2,1 norm instead of l2 norm for similarity regularizer (ablation) | FAITH · node classification (Appendix F) | sparsity and utility-fairness trade-off | Improved |
| change teacher GNN to GCN, GAT or APPNP (ablation) | FAITH · Cora, Pubmed, A-Photo | accuracy and IF | No clear effect |
| vary hyperparameters alpha, beta, gamma (ablation) | FAITH · Cora | accuracy and IF | No clear effect |
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.
- TINED — TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation (ICML 2025) · better on 6 of 6 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
- VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs (ICLR 2024 · similarity 0.91)
- Chasing Fairness in Graphs: A GNN Architecture Perspective (AAAI 2024 · similarity 0.91)
- Distilling Graph Geometry: Knowledge Gap from GNNs to MLPs (NeurIPS 2026 · similarity 0.91)
- TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation (ICML 2025 · similarity 0.89)
- Mirage: Model-agnostic Graph Distillation for Graph Classification (ICLR 2024 · similarity 0.88)
- Graph Distillation with Eigenbasis Matching (ICML 2024 · similarity 0.88)
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.
