Arrow: Accelerator for Time Series Causal Discovery with Time Weaving

Yuanyuan Yao, Yuan Dong, Lu Chen, Kun Kuang, Ziquan Fang, Cheng Long, Yunjun Gao, TIANYI LI

Abstract: Current causal discovery methods for time series data can effectively address a variety of scenarios; however, they remain constrained by inefficiencies. The significant inefficiencies arise primarily from the high computational costs associated with binning, the uncertainty in selecting appropriate time lags, and the extensive sets of candidate variables. To achieve both high efficiency and effectiveness of causal discovery, we introduce an accelerator termed ARROW. It incorporates an innovative concept termed “Time Weaving” that efficiently encodes time series data to well capture the dynamic trends, thereby mitigating computational complexity. We also propose a novel time lag discovery strategy utilizing XOR operations, which derives a theorem to obtain the optimal time lag and significantly enhances the efficiency using XOR operations. To optimize the search space for causal relationships, we design an efficient pruning strategy that intelligently identifies the most relevant candidate variables, enhancing the efficiency and accuracy of causal discovery. We applied ARROW to four different types of time series causal discovery algorithms and evaluated it on 25 synthetic and real-world datasets. The results demonstrate that, compared to the original algorithms, ARROW achieves up to 153x speedup while achieving higher accuracy in most cases.

Lay Summary:

We aim to enable efficient causal discovery in time series data. However, existing methods often suffer from low efficiency when dealing with complex time lags, high-dimensional variable combinations, and intensive computation.

Our paper introduces a simple yet highly effective idea: to discover causal relationships in time series, we can start from the overall trends and use a lightweight approach to quickly capture temporal dependencies between variables. This might seem surprising, as traditional methods often rely on complex modeling and heavy computation to handle challenges like time lags and non-linearity. Our method builds on two key ideas: Time Weaving, which captures dynamic temporal trend changes with low overhead, and a XOR-based strategy for fast time-lag discovery, while intelligent variable selection minimizes redundant computation.

Our research results provide new insights into accelerating causal discovery algorithms and demonstrate that, through ARROW, we can not only significantly speed up the causal discovery process but also achieve better results in most scenarios.

Link To Code: https://github.com/XiangguanMu/arrow

Primary Area: General Machine Learning->Causality

Keywords: time series, causal discovery, accelerator

Submission Number: 5454