Shylock: Causal Discovery in Multivariate Time Series based on Hybrid Constraints

TL;DR

The paper tackles causal discovery in multivariate time series under few-shot data conditions and with time-delayed effects. It introduces Shylock, a neural-network framework that combines attention-based local causal discovery with parameter sharing and group dilated convolutions, and a differentiable global DAG constraint to enforce acyclicity. Through a lightweight synthetic data generator and evaluation on both synthetic and FMRI datasets, Shylock outperforms state-of-the-art methods NOTEARS and TCDF, particularly in low-data regimes, while maintaining interpretability via the attention matrix. The approach offers practical impact by enabling robust causal inference in settings with limited data and time delays, and is complemented by the Tcausal library for easy deployment on EarthDataMiner.

Abstract

Causal relationship discovery has been drawing increasing attention due to its prevalent application. Existing methods rely on human experience, statistical methods, or graphical criteria methods which are error-prone, stuck at the idealized assumption, and rely on a huge amount of data. And there is also a serious data gap in accessing Multivariate time series(MTS) in many areas, adding difficulty in finding their causal relationship. Existing methods are easy to be over-fitting on them. To fill the gap we mentioned above, in this paper, we propose Shylock, a novel method that can work well in both few-shot and normal MTS to find the causal relationship. Shylock can reduce the number of parameters exponentially by using group dilated convolution and a sharing kernel, but still learn a better representation of variables with time delay. By combing the global constraint and the local constraint, Shylock achieves information sharing among networks to help improve the accuracy. To evaluate the performance of Shylock, we also design a data generation method to generate MTS with time delay. We evaluate it on commonly used benchmarks and generated datasets. Extensive experiments show that Shylock outperforms two existing state-of-art methods on both few-shot and normal MTS. We also developed Tcausal, a library for easy use and deployed it on the EarthDataMiner platform

Figures (6)

• Figure 1: Temporal causal graph representation.
• Figure 2: Approach Overview of Shylock.
• Figure 3: Group dilated convolution.
• Figure 4: F1 scores on the synthetic datasets. The chart name consists of two qualifiers. The first qualifier describes the method, the second describes node_num $n$.
• Figure 5: Comparison of fitting ability between Shylock and TCDF on datasets of different sizes.