Research map / Graph learning / GNN architectures
Deep Constraint-Based Propagation in Graph Neural Networks
Matteo Tiezzi, Giuseppe Marra, Stefano Melacci, Marco Maggini
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 |
|---|---|---|---|
| train with Lagrangian constraint-based propagation instead of iterative fixed-point GNN* training | GNN* · Clique localization | accuracy (%): 91.86 -> 93.07 | Improved |
| train with Lagrangian constraint-based propagation instead of iterative fixed-point GNN* training | GNN* · Subgraph matching | accuracy (%): 95.86 -> 96.30 | No clear effect |
| use squared G instead of lin G (ablation) | LP-GNN · Clique localization | accuracy (%): 84.61 -> 93.07 | Improved |
| use unilateral abs G instead of bilateral lin G (ablation) | LP-GNN · Clique localization | accuracy (%): 84.61 -> 88.80 | Improved |
| increase epsilon from 0.00 to 0.10 for abs G (ablation) | LP-GNN · Clique localization | accuracy (%): 88.80 -> 85.88 | Hurt |
| use multiple state layers instead of a single layer (ablation) | LP-GNN · NCI1 | accuracy (%): 68.4 -> 74.9 | Improved |
| use multiple state layers instead of a single layer (ablation) | LP-GNN · IMDB-B | accuracy (%): 65.3 -> 76.2 | Improved |
| use multiple state layers instead of a single layer (ablation) | LP-GNN · PTC | accuracy (%): 64.4 -> 67.9 | Improved |
| LP training vs original GNN* training | GNN* · NCI1 | accuracy (%): 51.5 -> 68.4 | Improved |
| LP training vs original GNN* training | GNN* · IMDB-B | accuracy (%): 60.9 -> 65.3 | Improved |
| use LP-GNN with one state layer instead of GIN | GIN · IMDB-B (1 state layer) | accuracy (%): 52 -> 65.3 | Improved |
| use 2 state layers instead of 5 (ablation) | LP-GNN · IMDB-B | accuracy (%): 76.2 -> 73.7 | Mixed |
Closest papers by meaning
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.90)
- A Generalized Neural Diffusion Framework on Graphs (AAAI 2024 · similarity 0.90)
- Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.89)
- AutoSGNN: Automatic Propagation Mechanism Discovery for Spectral Graph Neural Networks (AAAI 2025 · similarity 0.88)
- On the Interplay between Graph Structure and Learning Algorithms in Graph Neural Networks (ICML 2025 · similarity 0.88)
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · 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.
