Research map / Graph learning / GNN architectures
DropGNN: Random Dropouts Increase the Expressiveness of Graph Neural Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 independent random node-dropout runs and aggregate their embeddings | GIN · Limits 1, Limits 2, 4-cycles, LCC, Triangles, Skip-circles | test accuracy: Limits 1: 0.50 ± 0.00 -> 1.00 ± 0.00; Limits 2: 0.50 ± 0.00 -> 1.00 ± 0.00; 4-cycles: 0.50 ± 0.00 -> 1.00 ± 0.01; LCC: 0 | Helped |
| Add random node-dropout runs to GIN | GIN · MUTAG, PTC, PROTEINS, IMDB-BINARY | accuracy: MUTAG: 89.4 ± 5.6 -> 90.4 ± 7.0; PTC: 64.6 ± 7.0 -> 66.3 ± 8.6; PROTEINS: 76.2 ± 2.8 -> 76.3 ± 6.1; IMDB-BINARY: 75.1 ± | Helped |
| Add random node-dropout runs to GIN | GIN · IMDB-MULTI | accuracy: 52.3 ± 2.8 -> 51.4 ± 2.8 | Hurt |
| Add random node-dropout runs | MPNN · QM9 μ, α, ε_LUMO, Δε, ⟨R²⟩, ZPVE | MAE: μ: 0.358 -> 0.059*; α: 0.89 -> 0.173*; ε_LUMO: 0.00623 -> 0.00177*; Δε: 0.0066 -> 0.00282*; ⟨R²⟩: 28.5 -> 0.392*; ZPVE: | Helped |
| Add random node-dropout runs | MPNN · QM9 ε_HOMO, U₀, U, H, G, Cᵥ | MAE: ε_HOMO: 0.00541 -> 0.00193*; U₀: 2.05 -> 0.0409*; U: 2.0 -> 0.0536*; H: 2.02 -> 0.0481*; G: 2.02 -> 0.0508*; Cᵥ: 0.42 -> | Hurt |
| Add random node-dropout runs | 1-GNN · QM9, 12 properties | MAE | Mixed |
| Increase the number of independent runs (ablation) | DropGNN · Limits 1, 4-cycles, Triangles | accuracy | Helped |
| Vary dropout probability over the tested range (ablation) | DropGNN · Limits 1, 4-cycles, Triangles | accuracy | No clear effect |
Closest papers by meaning
- Effects of Dropout on Performance in Long-range Graph Learning Tasks (NeurIPS 2025 · similarity 0.89)
- Almost Surely Asymptotically Constant Graph Neural Networks (NeurIPS 2024 · similarity 0.88)
- An Empirical Study of Realized GNN Expressiveness (ICML 2024 · similarity 0.87)
- Aggregation Buffer: Revisiting DropEdge with a New Parameter Block (ICML 2025 · similarity 0.87)
- Pruning for GNNs: Lower Complexity with Comparable Expressiveness (ICML 2025 · similarity 0.87)
- Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness (ICLR 2024 · similarity 0.86)
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.