Research map / Graph learning / GNN architectures
Graph Ladling: Shockingly Simple Parallel GNN Training without Intermediate Communication
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 |
|---|---|---|---|
| Greedily interpolate independently trained model ingredients rather than use one vanilla model | GCN · Cora | accuracy: 82.39 -> 83.52 | Helped |
| Greedily interpolate independently trained model ingredients rather than use one vanilla model | GCNII · PubMed | accuracy: 78.06 -> 79.44 | Helped |
| Use node-wise sampling for soup ingredients instead of edge-wise sampling (ablation) | Data-centric GNN soup · Reddit | accuracy: 92.66±0.34 % -> 97.28 ± 0.08 % | Helped |
| Use node-wise sampling for soup ingredients instead of layer-wise sampling (ablation) | Data-centric GNN soup · OGBN-ArXiv | accuracy: 68.30 ± 0.54 % -> 72.83 ± 0.21 % | Helped |
| Use graph partition soup instead of ClusterGCN | Graph partition training · Flickr | accuracy: 51.20 ± 0.13% -> 52.23 ± 0.12 % | Helped |
| Increase the ingredient count from 10 to 100 (ablation) | Our Data Partition Soup · OGBN-ArXiv | performance: 71.426 -> 72.814 | Helped |
| Add intermediate souping every 100 epochs during training (ablation) | Data partition GCN soup · OGBN-ArXiv | performance: -0.745% | Hurt |
| Reduce GraphSAGE neighborhood sample size to one-fourth and use node-sampled soup | GraphSAGE on OGBN-products · OGBN-products | accuracy: 79.5 ± 0.36 -> 80.42 ± 0.41 | Helped |
| Reduce GraphSAGE neighborhood sample size to one-fourth and use node-sampled soup | GraphSAGE on OGBN-products · OGBN-products | Act. Memory(MB): 415.94 -> 369.51 | Helped |
| Reduce GraphSAGE neighborhood sample size to one-fourth and use node-sampled soup | GraphSAGE on OGBN-products · OGBN-products | Throughput(iter/sec): 37.69 -> 44.30 | Helped |
| Use layer-wise sampled soup rather than node-wise sampled soup (ablation) | Data-centric GNN soup · Flickr | accuracy: 54.47 ± 0.13 % -> 51.08 ± 0.22 % | Hurt |
Closest papers by meaning
- MuseGNN: Forming Scalable, Convergent GNN Layers that Minimize a Sampling-Based Energy (ICLR 2025 · similarity 0.89)
- Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.89)
- Towards Scalable and Deep Graph Neural Networks via Noise Masking (AAAI 2025 · similarity 0.88)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.88)
- Dynamic Rescaling for Training GNNs (NeurIPS 2024 · similarity 0.88)
- GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning (NeurIPS 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.