Research map / Graph learning / Knowledge and temporal graphs
Global-Lens Transformers: Adaptive Token Mixing for Dynamic Link Prediction
Tao Zou, Chengfeng Wu, Tianxi Liao, Junchen Ye, Bowen Du
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 |
|---|---|---|---|
| replace self-attention with the adaptive token mixer (GLFormer) | DyGFormer · MOOC, transductive AP | AP: 87.52 -> 87.87 | Improved |
| replace self-attention with the adaptive token mixer (GLFormer) | TGN · LastFM, transductive AP | AP: 77.07 -> 84.96 | Improved |
| replace self-attention with the adaptive token mixer (GLFormer) | TGAT · Enron, transductive AP | AP: 71.12 -> 75.48 | Improved |
| replace self-attention with the adaptive token mixer (GLFormer) | CAWN · Wikipedia, transductive AP | AP: 98.76 -> 98.67 | No clear effect |
| replace self-attention with MLP token mixer | CAWN · MOOC, transductive AP | AP: 80.15 -> 82.86 | Improved |
| replace self-attention with average pooling over recent neighbors | DyGFormer · MOOC, transductive AP | AP: 87.52 -> 84.91 | Hurt |
| replace self-attention with pooling or MLP (attention-free) | five Transformer baselines · Wikipedia, MOOC, LastFM, Enron | AP | No clear effect |
| remove learnable order parameters (w/o LP) (ablation) | GLFormer · Wikipedia, MOOC, LastFM, Enron | AP | Improved |
| add relative-time weights (versus w/o RT) (ablation) | GLFormer · Wikipedia, MOOC, LastFM, Enron | AP | Improved |
| use GELU instead of ReLU (ablation) | GLFormer · Wikipedia, MOOC, LastFM, Enron | AP | Improved |
| add residual connections (versus w/o ResNet) (ablation) | GLFormer · Wikipedia, MOOC, LastFM, Enron | AP | Improved |
| add channel mixers (versus w/o CM) (ablation) | GLFormer · Wikipedia, MOOC, LastFM, Enron | AP | Improved |
Closest papers by meaning
- Future Link Prediction Without Memory or Aggregation (NeurIPS 2025 · similarity 0.89)
- Supra-Laplacian Encoding for Transformer on Dynamic Graphs (NeurIPS 2024 · similarity 0.88)
- A Deep Probabilistic Framework for Continuous Time Dynamic Graph Generation (AAAI 2025 · similarity 0.87)
- Unifying and Enhancing Graph Transformers via a Hierarchical Mask Framework (NeurIPS 2025 · similarity 0.87)
- Even Sparser Graph Transformers (NeurIPS 2024 · similarity 0.87)
- A Dynamical Systems-Inspired Pruning Strategy for Addressing Oversmoothing in Graph Attention Networks (ICML 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.
