arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.25473cs.LG
收藏

Causal-INSIGHT: 探索时间模型以提取因果结构

Causal-INSIGHT: Probing Temporal Models to Extract Causal Structure

Queen's University Belfast(贝尔法斯特女王大学)

Benjamin Redden, Hui Wang, Shuyan Li

AI总结:

本文提出Causal-INSIGHT框架,用于从训练好的时间预测模型中提取因果结构,通过干预启发的输入钳制分析,结合Qbic准则实现稀疏性与结构复杂度的平衡。

摘要:

理解多变量时间序列中的有向时间交互对于解释复杂动态系统及其训练模型至关重要。我们提出了Causal-INSIGHT,一种模型无关的后验解释框架,用于从训练好的时间预测器中提取模型隐含的、预测器依赖的、有向的时间滞后影响结构。与其推断数据生成过程的因果结构不同,Causal-INSIGHT分析固定预训练预测器在推理时对系统性干预启发的输入钳制的响应。从这些响应中,我们构建了有向时间影响信号,反映了预测器用于预测所依赖的依赖关系,并引入Qbic,一种稀疏性感知的图选择标准,可在不需地面真实图标签的情况下平衡预测保真度和结构复杂度。在合成、模拟和现实基准上的实验表明,Causal-INSIGHT在不同骨干架构上具有良好的泛化能力,保持了竞争性的结构准确性,并在应用于现有预测器时显著提高了时间延迟定位性能。

英文摘要:

Understanding directed temporal interactions in multivariate time series is essential for interpreting complex dynamical systems and the predictive models trained on them. We present Causal-INSIGHT, a model-agnostic, post-hoc interpretation framework for extracting model-implied (predictor-dependent), directed, time-lagged influence structure from trained temporal predictors. Rather than inferring causal structure at the level of the data-generating process, Causal-INSIGHT analyzes how a fixed, pre-trained predictor responds to systematic, intervention-inspired input clamping applied at inference time. From these responses, we construct directed temporal influence signals that reflect the dependencies the predictor relies on for prediction, and introduce Qbic, a sparsity-aware graph selection criterion that balances predictive fidelity and structural complexity without requiring ground-truth graph labels. Experiments across synthetic, simulated, and realistic benchmarks show that Causal-INSIGHT generalizes across diverse backbone architectures, maintains competitive structural accuracy, and yields significant improvements in temporal delay localization when applied to existing predictors.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑