Research map / Graph learning / GNN architectures
GNN-FiLM: Graph Neural Networks with Feature-wise Linear Modulation
Marc Brockschmidt
This paper tried a common change: Add dropout
Add dropout 0.2 · PPI · Avg. Micro-F1: 0.924 ± 0.004 -> 0.928 ± 0.005 Improved
What 795 accepted papers report for this change
- 549 improved
- 87 no clear effect
- 12 mixed
- 147 hurt
- Don't Drop Dropout: Optimizing Layer Sparsity for Efficient LLM Training and Inference ICML 2026 · apply layer dropout to whole transformer blocks instead of attention and FFN sub-blocks separately · 20 TPP val loss, 271M · val loss: 2.377 -> 2.367Improved
- Quantifying the Uncertainty of Foundation Models with Singular Value Ensembles ICML 2026 · use MC Dropout · vision and SST-2 · accuracy: 86.3 -> 76.6 on Flowers102Hurt
- Motion Planning in Compressed Representation Spaces ICML 2026 · add nested dropout during training · WOMD guided generation · goal FDE: 4.145 -> 1.104Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 target-node-conditioned feature-wise modulation for incoming messages | GNN-FiLM · QM9 | MAE: alpha: 4.22 ± 0.45 -> 3.75 ± 0.11; U0: 5.55 ± 0.38 -> 5.43 ± 0.96; R2: 14.86 ± 1.62 -> 15.59 ± 1.38 | Mixed |
| Use target-node-conditioned feature-wise modulation for incoming messages | GNN-FiLM · PPI | Avg. Micro-F1: 0.992 ± 0.000 vs 0.992 ± 0.000 | No clear effect |
| Use target-node-conditioned feature-wise modulation for incoming messages | GNN-FiLM · VarMisuse SeenProjTest | accuracy: 87.0 ± 0.2% vs 87.2 ± 1.5% | No clear effect |
| Move the nonlinearity before message aggregation (ablation) | GNN-FiLM · PPI, QM9, and VarMisuse | performance | Improved |
| Use different layer weights at each propagation step instead of sharing them (ablation) | All models · PPI, QM9, and VarMisuse | performance | Improved |
| Generalize GAT to distinguish edge types and add a self-loop edge type (ablation) | GAT · PPI | Avg. Micro-F1: 0.953 ± 0.002 -> 0.989 ± 0.001 | Improved |
| Add dropout 0.2 (ablation) | GAT, dimension 256 · PPI | Avg. Micro-F1: 0.924 ± 0.004 -> 0.928 ± 0.005 | Improved |
| Increase representation dimension from 256 to 320 (ablation) | GAT without dropout · PPI | Avg. Micro-F1: 0.924 ± 0.004 -> 0.942 ± 0.004 | Improved |
| Add dropout 0.2 and increase representation dimension to 320 (ablation) | GAT · PPI | Avg. Micro-F1: 0.924 ± 0.004 -> 0.953 ± 0.002 | Improved |
| Replace standard GAT edge handling with relation-specific edges and self-loops (ablation) | GAT · PPI | Avg. Micro-F1: 0.953 ± 0.002 -> 0.986 ± 0.001 | Improved |
| Use a full dynamic matrix hypernetwork (RGDCN) rather than feature-wise modulation (ablation) | RGDCN · Not reported | training stability | Hurt |
| Move the nonlinearity before message aggregation in other GNN types (ablation) | GNN types other than GNN-FiLM · Not reported | performance | No clear effect |
Closest papers by meaning
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.88)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.86)
- Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.85)
- TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation (ICML 2025 · similarity 0.85)
- Feature Distribution on Graph Topology Mediates the Effect of Graph Convolution: Homophily Perspective (ICML 2024 · similarity 0.85)
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 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.
