Research map / Probabilistic and scientific modeling / Causal discovery
Structure Learning from Time-Series Data with Lag-Agnostic Structural Prior
Taiyu Ban, Changxin Rong, Xiangyu Wang, Lyuzhou Chen, Yanze Gao, Xin Wang, Huanhuan Chen
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 lag-agnostic prior (logic-dual, Init Data, 10% priors) | DYNOTEARS · 20-node ER4 Gaussian | SHD: 13.00 -> 8.00 | Improved |
| add lag-agnostic prior (logic-dual, Init Data, 50% priors) | DYNOTEARS · 20-node ER4 Gaussian | SHD: 13.00 -> 5.50 | Improved |
| use process-equivalent binary-masked loss instead of max-based loss (Init Data) (ablation) | DYNOTEARS · 20-node ER4 Gaussian, 30% priors | SHD: 7.83 -> 5.83 | Improved |
| use logic-dual loss instead of max-based loss (Init 0) (ablation) | DYNOTEARS · 20-node ER4 Gaussian, 50% priors | SHD: 7.83 -> 5.93 | Improved |
| add data-driven initialization to logic-dual loss (ablation) | DYNOTEARS* · 20-node ER4 Gaussian, 5% priors | SHD: 10.87 -> 10.50 | Improved |
| add data-driven initialization to max-based loss (ablation) | DYNOTEARSˆ · 20-node ER4 Gaussian, 5% priors | SHD: 12.00 -> 11.50 | Improved |
| add data-driven initialization to logic-dual loss (ablation) | DYNOTEARS* · DREAM4-189, 75% priors | test-set regression loss: 4.27 -> 3.40 | Improved |
| add data-driven initialization to logic-dual loss (ablation) | DYNOTEARS* · DREAM4-63, 75% priors | test-set regression loss: 5.28 -> 6.22 | Hurt |
| replace perfect lag knowledge with 50% wrong lags (lag-specific prior) versus lag-agnostic prior | LS baseline · 20-node ER4 Gaussian, 50% priors | SHD: 4.83 -> 11.83 | Hurt |
| add lag-agnostic prior with correct summary edges versus random lag priors | DYNOTEARS · DREAM4-189, 75% priors | test-set regression loss: 5.43 -> 3.25 | Improved |
| add lag-agnostic prior (binary-masked, Init Data) | DYNOTEARS · DREAM4-189, 75% priors | AUROC: 0.65 -> 0.92 | Improved |
| use max-based loss with Init Data instead of Init 0 (ablation) | DYNOTEARSˆ · DREAM4-63, 25% priors | AUROC: 0.70 -> 0.58 | Hurt |
Closest papers by meaning
- Stable Partial Order Constraints for Temporal Causal Structure Learning (NeurIPS 2026 · similarity 0.91)
- From Temporal to Contemporaneous Iterative Causal Discovery in the Presence of Latent Confounders (ICML 2023 · similarity 0.89)
- Learning Time-Aware Causal Representation for Model Generalization in Evolving Domains (ICML 2025 · similarity 0.88)
- Learning Causal Relations from Subsampled Time Series with Two Time-Slices (ICML 2024 · similarity 0.87)
- On the Identification of Temporal Causal Representation with Instantaneous Dependence (ICLR 2025 · similarity 0.87)
- Discovering Mixtures of Structural Causal Models from Time Series Data (ICML 2024 · 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.
