Research map / Graph learning / GNN architectures
Shared Growth of Graph Neural Networks via Prompted Free-Direction Knowledge Distillation
Kaituo Feng, Yikun Miao, Changsheng Li, Ye Yuan, Guoren Wang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add the RL knowledge judge (vs distilling every node both ways) (ablation) | FreeKD with GCN-GCN · Cora | Micro-F1: 85.83 -> 86.53 | Improved |
| use RL agent instead of cross-entropy-loss heuristic for direction (ablation) | FreeKD with GCN-GCN · Chameleon | Micro-F1: 35.77 -> 37.48 | Improved |
| use RL agent instead of cross-entropy-loss heuristic for direction (ablation) | FreeKD with GCN-GCN · Cora | Micro-F1: 85.89 -> 86.53 | Improved |
| add structure-level actions (ablation) | FreeKD with GCN-GCN · Cora | Micro-F1: 86.13 -> 86.53 | Improved |
| select agent-selected neighborhood instead of all neighbors (ablation) | FreeKD with GCN-GCN · Chameleon | Micro-F1: 36.54 -> 37.48 | Improved |
| add structure-level distillation to node-level distillation (ablation) | FreeKD with GCN-GCN · Chameleon | Micro-F1: 36.18 -> 37.48 | Improved |
| add information preservation loss to FreeKD-Prompt (ablation) | FreeKD-Prompt with GCN-GCN · Chameleon | Micro-F1: 37.91 -> 40.04 | Improved |
| add diversity loss to FreeKD-Prompt (ablation) | FreeKD-Prompt with GCN-GCN · Chameleon | Micro-F1: 39.30 -> 40.04 | Improved |
| add information preservation loss to FreeKD-Prompt (ablation) | FreeKD-Prompt with GCN-GCN · Cora | Micro-F1: 86.74 -> 87.33 | Improved |
| add FreeKD mutual distillation between two GCNs | GCN · Chameleon | Micro-F1: 33.09 -> 37.48 | Improved |
| add prompt-based augmentation on top of FreeKD | FreeKD with GCN-GCN · Texas | Micro-F1: 60.00 -> 65.41 | Improved |
| add FreeKD-Prompt to GraphSAGE | GraphSAGE · PPI | Micro-F1: 69.28 -> 72.95 | Improved |
Closest papers by meaning
- Boosting Graph Neural Networks via Adaptive Knowledge Distillation (AAAI 2023 · similarity 0.91)
- Multi-Scale Distillation from Multiple Graph Neural Networks (AAAI 2022 · similarity 0.91)
- Demystifying GNN-to-MLP Knowledge Transfer: Theoretical Grounding and Dual-Stream Distillation Method (AAAI 2026 · similarity 0.90)
- Large Language Model Meets Graph Neural Network in Knowledge Distillation (AAAI 2025 · similarity 0.88)
- VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs (ICLR 2024 · similarity 0.87)
- All You Need in Knowledge Distillation Is a Tailored Coordinate System (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.
