Research map / Probabilistic and scientific modeling / Causal discovery
TopoDistill: Distilling Global System Topology for Causal Discovery in Multivariate Time Series
Zehao Liu, Pengfei Jiao, Yuhan Wu, Jianqi Yang, Yuyu Yin
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-guided distillation from the global embedder (ablation) | TopoDistill · fMRI | AUC-PR: 0.596 -> 0.744 | Improved |
| add topology-guided distillation from the global embedder (ablation) | TopoDistill · Causal Rivers | AUC-PR: 0.415 -> 0.508 | Improved |
| add temporal contrastive learning (ablation) | TopoDistill · fMRI | AUC-PR: 0.642 -> 0.744 | Improved |
| add temporal contrastive learning (ablation) | TopoDistill · Causal Rivers | AUC-PR: 0.452 -> 0.508 | Improved |
| use topology and contrastive losses instead of window reconstruction objective (ablation) | TopoDistill · Causal Rivers | AUC-ROC: 0.672 -> 0.813 | Improved |
| use topology and contrastive losses instead of window reconstruction objective (ablation) | TopoDistill · fMRI | SHD: 0.100 -> 0.067 | Improved |
| add topology distillation (ablation) | TopoDistill · Causal Rivers | SHD: 0.200 -> 0.150 | Improved |
| replace TCN delay embedder with MLP | TopoDistill · Double Pendulum | AUC-ROC: 0.888 -> 0.912 | No clear effect |
| replace TCN delay embedder with MLP | TopoDistill · Causal Rivers | AUC-PR: 0.508 -> 0.525 | No clear effect |
| replace time-delay embedding with learned TopoDistill embedder | CCM pipeline · Lorenz-96 with noise | AUC-ROC/AUC-PR | Improved |
| replace CCM with TopoDistill-TCN | CCM · Double Pendulum | SHD: 0.167 -> 0.067 | Improved |
| replace CCM with TopoDistill-TCN | CCM · Causal Rivers | AUC-ROC: 0.594 -> 0.794 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CUTS+ — CUTS: Neural Causal Discovery from Irregular Time-Series Data (ICLR 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Tangent Space Causal Inference: Leveraging Vector Fields for Causal Discovery in Dynamical Systems (NeurIPS 2024 · similarity 0.89)
- Hybrid Top-Down Global Causal Discovery with Local Search for Linear and Nonlinear Additive Noise Models (NeurIPS 2024 · similarity 0.88)
- Learning the Causal Structure of Networked Dynamical Systems under Latent Nodes and Structured Noise (AAAI 2024 · similarity 0.87)
- Learning Causal Relations from Subsampled Time Series with Two Time-Slices (ICML 2024 · similarity 0.87)
- CausalDynamics: A large‐scale benchmark for structural discovery of dynamical causal models (NeurIPS 2025 · similarity 0.87)
- Structured Temporal Causality for Interpretable Multivariate Time Series Anomaly Detection (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.
