Research map / Continual learning / Online and replay-based
Hierarchical Prototype Networks for Continual Graph Representation Learning
Xikun Zhang, Dongjin Song, Dacheng Tao
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 N-prototypes to A-prototypes (ablation) | HPNs · Cora (3 tasks) | accuracy mean (AM): 89.2 -> 91.7 | Improved |
| add C-prototypes on top of A- and N-prototypes (ablation) | HPNs · Cora (3 tasks) | accuracy mean (AM): 91.7 -> 93.7 | Improved |
| add L_div to L_cls (ablation) | HPNs · Cora (3 tasks) | accuracy mean (AM): 92.4 -> 92.9 | Improved |
| add L_dis to L_cls (ablation) | HPNs · Cora (3 tasks) | accuracy mean (AM): 92.4 -> 92.8 | Improved |
| add both L_div and L_dis to L_cls (ablation) | HPNs · Cora (3 tasks) | accuracy mean (AM): 92.4 -> 93.7 | Improved |
| add L_div and L_dis to L_cls (ablation) | HPNs · Cora (3 tasks) | forgetting mean (FM): +0.8 -> +0.6 | No clear effect |
| replace GAT+TWP/MAS-style continual learning baselines with HPNs | GAT backbone · OGB-Arxiv (20 tasks) | accuracy mean (AM): 81.8 -> 85.8 | Improved |
| use HPNs instead of joint training (GAT) | OGB-Products · OGB-Products (23 tasks) | accuracy mean (AM): 73.7 -> 80.1 | Improved |
| use HPNs instead of joint training (GAT) | Cora · Cora (3 tasks) | accuracy mean (AM): 93.9 -> 93.7 | No clear effect |
| use HPNs instead of GEM (replay) in memory use | GAT on OGB-Products · OGB-Products (23 tasks) | final parameter count: 2,220,032 -> 4,908 | Improved |
| increase number of AFEs and prototype dimension (ablation) | HPNs · OGB-Arxiv (20 tasks) | AM and FM | Improved |
| vary prototype threshold t_A from 0.01 to 0.4 (ablation) | HPNs · Cora (3 tasks) | AM and FM | No clear effect |
Closest papers by meaning
- A Topology-aware Graph Coarsening Framework for Continual Graph Learning (NeurIPS 2024 · similarity 0.87)
- Towards Continuous Reuse of Graph Models via Holistic Memory Diversification (ICLR 2025 · similarity 0.87)
- Towards Continual Knowledge Graph Embedding via Incremental Distillation (AAAI 2024 · similarity 0.86)
- Exploring Rationale Learning for Continual Graph Learning (AAAI 2025 · similarity 0.86)
- A Selective Learning Method for Temporal Graph Continual Learning (ICML 2025 · similarity 0.86)
- GraphKeeper: Graph Domain-Incremental Learning via Knowledge Disentanglement and Preservation (NeurIPS 2025 · similarity 0.86)
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.
