Research map / Graph learning / GNN architectures
Do We Need Anisotropic 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 |
|---|---|---|---|
| Use isotropic EGC-M with multiple aggregators instead of anisotropic PNA | EGC-M · ZINC | MAE: 0.320 ± 0.032 -> 0.281 ± 0.007 | Helped |
| Use isotropic EGC-M with multiple aggregators instead of anisotropic PNA | EGC-M · CIFAR-10 Superpixels | accuracy: 70.21 ± 0.15 -> 71.03 ± 0.42 | Helped |
| Use isotropic EGC-M with multiple aggregators instead of anisotropic PNA | EGC-M · MolHIV | ROC-AUC: 79.05 ± 1.32 -> 78.18 ± 1.53 | Hurt |
| Use isotropic EGC-M with multiple aggregators instead of anisotropic PNA | EGC-M · Code-V2 | F1: 0.1570 ± 0.0032 -> 0.1595 ± 0.0019 | Helped |
| Use EGC-S instead of the strongest listed baseline | EGC-S · Arxiv | accuracy: 71.92 ± 0.21 -> 72.21 ± 0.17 | Helped |
| Add multiple aggregator functions to EGC-S to form EGC-M (ablation) | EGC · ZINC | MAE: 0.364 ± 0.020 -> 0.281 ± 0.007 | Helped |
| Add multiple aggregator functions to EGC-S to form EGC-M (ablation) | EGC · CIFAR-10 Superpixels | accuracy: 66.92 ± 0.37 -> 71.03 ± 0.42 | Helped |
| Apply an activation function to combination weights (ablation) | EGC-S · ZINC | MAE: 0.364 ± 0.020 -> 0.435 ± 0.010 (Hardtanh); 0.366 ± 0.008 (Sigmoid); 0.404 ± 0.010 (Softmax) | Hurt |
| Use DropEdge with p = 0.1 (ablation) | EGC-S · ZINC | MAE: 0.364 ± 0.020 -> 0.468 ± 0.007 | Hurt |
| Use DropEdge with p = 0.5 (ablation) | EGC-S · ZINC | MAE: 0.364 ± 0.020 -> 0.629 ± 0.023 | Hurt |
| Apply GraphSAINT node sampling with p = 0.1 (ablation) | EGC-S · ZINC | MAE: 0.364 ± 0.020 -> 0.631 ± 0.012 | Hurt |
| Replace the chosen EGC-M aggregator set with symmetric normalization + max + min (ablation) | EGC-M · CIFAR-10 Superpixels | accuracy: 71.03 ± 0.42 -> 70.05 ± 1.14 | Hurt |
Closest papers by meaning
- Directional Graph Networks (ICML 2021 · similarity 0.90)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.89)
- Analyzing the Expressive Power of Graph Neural Networks in a Spectral Perspective (ICLR 2021 · similarity 0.89)
- Graph Neural Networks Inspired by Classical Iterative Algorithms (ICML 2021 · similarity 0.88)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.88)
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · similarity 0.87)
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.