📄 中文摘要
提出了一种名为CausalTimePrior的框架，用于生成配对的观察性和干预性时间序列的合成时间结构因果模型（TSCMs）。该框架解决了现有时间序列基准缺乏干预数据的问题，现有基准虽然能够生成具有真实因果图的观察数据，但无法提供训练因果基础模型所需的干预数据。CausalTimePrior支持可配置的因果图结构、非线性自回归机制、状态切换动态以及多个干预目标的生成，旨在推动因果推断在时间序列数据中的应用。
因果基础模型的干预时间序列先验
出处: Interventional Time Series Priors for Causal Foundation Models
发布: 2026年3月13日
📄 English Summary
Interventional Time Series Priors for Causal Foundation Models
CausalTimePrior is proposed as a principled framework for generating synthetic temporal structural causal models (TSCMs) that provide paired observational and interventional time series. This framework addresses the limitation of existing time series benchmarks, which generate observational data with ground-truth causal graphs but lack the necessary interventional data for training causal foundation models. CausalTimePrior supports configurable causal graph structures, nonlinear autoregressive mechanisms, regime-switching dynamics, and multiple interventional targets, aiming to enhance the application of causal inference in time series data.