Research map / Architectures and efficiency / Attention and transformers
Controlled Dynamics Attractor Transformer
Cheng Zhang, Minnan Luo, Zesheng Yang, MING LI, Yong-Jin Liu, Qinghua Zheng
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 topology-conditioned mutual excitation W (ablation) | CDAT · DD | accuracy (%): 83.1 -> 98.5 | Improved |
| add topology-conditioned mutual excitation W (ablation) | CDAT · NCI1 | accuracy (%): 82.8 -> 93.5 | Improved |
| add self-inhibition (ablation) | CDAT · graph classification (DD, NCI1, NCI109, MUTAG) | accuracy (%): DD 98.09 -> 98.50; NCI1 92.74 -> 93.50; NCI109 92.07 -> 94.30; MUTAG 98.46 -> 98.80 | Improved |
| use rank r = 4 instead of r = 8 for W (ablation) | CDAT · NCI1 | accuracy (%): 77.0 -> 93.5 | Improved |
| use low-rank symmetric W instead of full-rank W (ablation) | CDAT · NCI1 | accuracy (%): 82.9 -> 93.5 | Improved |
| use low-rank symmetric W instead of full-rank W (ablation) | CDAT · DD | accuracy (%): 98.0 -> 98.5 | No clear effect |
| use r = 4 instead of r = 1 (ablation) | CDAT · NCI1 | accuracy (%): 87.6 -> 93.5 | Improved |
| add CDAT dynamics on top of ET | ET · Amazon 1% anomaly detection | AUC (%): 91.9 -> 94.5 | Improved |
| add CDAT dynamics on top of ET | ET · T-Finance 1% anomaly detection | AUC (%): 92.2 -> 95.91 | Improved |
| add CDAT dynamics on top of ET | ET · NCI1 / NCI109 / MUTAG | accuracy (%): NCI1 +3.4%, NCI109 +3.8%, MUTAG +2.2% | Improved |
| add CDAT dynamics on top of ET | ET · PROTEINS | accuracy (%): 90.3 -> 87.1 | Hurt |
| add CDAT dynamics on top of ET | ET · Amazon 70% anomaly detection | AUC (%): 97.1 -> 93.0 | Hurt |
Closest papers by meaning
- A Dynamical Systems-Inspired Pruning Strategy for Addressing Oversmoothing in Graph Attention Networks (ICML 2025 · similarity 0.87)
- Recurrent Self-Attention Dynamics: An Energy-Agnostic Perspective from Jacobians (NeurIPS 2025 · similarity 0.86)
- Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning (ICML 2026 · similarity 0.84)
- Let's do the time-warp-attend: Learning topological invariants of dynamical systems (ICLR 2024 · similarity 0.84)
- Dissecting the Interplay of Attention Paths in a Statistical Mechanics Theory of Transformers (NeurIPS 2024 · similarity 0.84)
- Associative Transformer (CVPR 2025 · similarity 0.84)
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.
