Research map / Graph learning / GNN architectures
SLAPS: Self-Supervision Improves Structure Learning for 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 self-supervised denoising to latent graph learning (ablation) | SLAPS on Cora and Citeseer · Cora and Citeseer | accuracy | Helped |
| add self-supervision to SLAPS (ablation) | Cora nodes not connected to labeled nodes · Cora, unconnected test-node group | accuracy: 38.0% | Helped |
| add self-supervision to SLAPS (ablation) | Cora nodes connected to at least one labeled node · Cora, connected test-node group | accuracy: 8.9% | Helped |
| add self-supervision (ablation) | SLAPS with FP generator and 5 labeled nodes per class · Cora | accuracy: 16.7% | Helped |
| add self-supervision (ablation) | SLAPS with FP generator and 5 labeled nodes per class · Citeseer | accuracy: 22.0% | Helped |
| add self-supervision (ablation) | SLAPS with FP generator and 20 labeled nodes per class · Cora and Citeseer | accuracy: 10.0%; 7.0% | Helped |
| add self-training (ablation) | SLAPS · Cora, Citeseer, Cora390, Citeseer370, Pubmed | accuracy: 72.8 ± 0.8 -> 74.2 ± 0.5; 70.5 ± 1.1 -> 73.1 ± 1.0; 75.3 ± 1.0 -> 75.5 ± 0.7; 73.0 ± 0.9 -> 73.3 ± 0.6; 74.4 ± 0.6 -> 74 | Mixed |
| add AdaEdge (ablation) | SLAPS · Cora, Citeseer, Cora390, Citeseer370 | accuracy: 72.8 ± 0.8 -> 72.8 ± 0.7; 70.5 ± 1.1 -> 70.6 ± 1.5; 75.3 ± 1.0 -> 75.2 ± 0.6; 73.0 ± 0.9 -> 72.6 ± 1.4 | Mixed |
| use self-supervision-only two-stage graph learning instead of joint label and feature objectives (ablation) | SLAPS on Cora with FP generator · Cora | accuracy: +8.4% | Helped |
| increase GCN depth from 2 to 4 layers without self-supervision (ablation) | SLAPS without self-supervision · Cora | accuracy: 66.2% -> 67.1% | Helped |
| increase GCN depth from 4 to 6 layers without self-supervision (ablation) | SLAPS without self-supervision · Cora | accuracy: 67.1% -> 55.8% | Hurt |
| use mean adjacency symmetrization rather than max or no symmetrization (ablation) | SLAPS with MLP generator · Cora and Citeseer | accuracy | Helped |
Closest papers by meaning
- Graph Alignment for Benchmarking Graph Neural Networks and Learning Positional Encodings (ICML 2026 · similarity 0.88)
- Uncovering the Redundancy in Graph Self-supervised Learning Models (NeurIPS 2024 · similarity 0.88)
- Edge Prompt Tuning for Graph Neural Networks (ICLR 2025 · similarity 0.87)
- On the Interplay between Graph Structure and Learning Algorithms in Graph Neural Networks (ICML 2025 · similarity 0.87)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.87)
- Two Heads Are Better Than One: Boosting Graph Sparse Training via Semantic and Topological Awareness (ICML 2024 · 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.