Research map / Architectures and efficiency / RNNs and normalization
RSC: Accelerate Graph Neural Networks Training via Randomized Sparse Computations
Zirui Liu, CHEN SHENGYUAN, Kaixiong Zhou, Daochen Zha, Xiao Huang, Xia Hu
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 |
|---|---|---|---|
| Approximate backward sparse operations using layer-wise resource allocation | GCN · ogbn-proteins | AUC: 71.99±0.66 -> 71.60±0.45 | No clear effect |
| Approximate backward sparse operations using layer-wise resource allocation | GraphSAGE (full batch) · ogbn-proteins | speedup: 1.60× | Improved |
| Approximate backward sparse operations | Sparse-dense matrix multiplication on Reddit · Reddit | backward operation time: 44.23 -> 3.81 (11.6×) | Improved |
| Approximate the forward sparse operation only (ablation) | Two-layer GCN · Reddit | accuracy: 95.39±0.04 -> 16.45±0.39 | Hurt |
| Approximate sparse operations in both forward and backward passes (ablation) | Two-layer GCN · Reddit | accuracy: 95.39±0.04 -> 80.74±1.00 | Hurt |
| Add the switching mechanism, using exact operations in the final training stage (ablation) | GCN + RSC · ogbn-proteins | AUC: 71.60± 0.66 -> 72.19 ± 0.79 | Improved |
| Add caching of sampled sparse matrices (ablation) | GCN + RSC · ogbn-proteins | speedup: 1.19× -> 1.60× | Improved |
| Add caching of sampled sparse matrices (ablation) | GCN + RSC · ogbn-proteins | AUC: 71.60± 0.66 -> 69.80 ± 0.60 | Hurt |
| Combine caching with switching rather than caching alone (ablation) | GCN + RSC · ogbn-proteins | AUC: 69.80 ± 0.60 -> 71.60± 0.45 | Improved |
| Allocate samples by layer importance instead of uniform allocation (ablation) | RSC · Reddit | accuracy at matched speedup | Improved |
| Increase budget C (ablation) | RSC · Reddit | accuracy | Improved |
Closest papers by meaning
- Towards Scalable and Deep Graph Neural Networks via Noise Masking (AAAI 2025 · similarity 0.88)
- Two Heads Are Better Than One: Boosting Graph Sparse Training via Semantic and Topological Awareness (ICML 2024 · similarity 0.88)
- Random Search Neural Networks for Efficient and Expressive Graph Learning (NeurIPS 2025 · similarity 0.88)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.88)
- The Graphon Limit Hypothesis: Understanding Neural Network Pruning via Infinite Width Analysis (NeurIPS 2025 · similarity 0.87)
- Graph Coarsening via Supervised Granular-Ball for Scalable Graph Neural Network Training (AAAI 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.
