Abstract:Granger causality, commonly used for inferring causal structures from time series data, has been adopted in widespread applications across various fields due to its intuitive explainability and high compatibility with emerging deep neural network prediction models. To alleviate challenges in better deciphering causal structures unambiguously from time series, the use of interventional data has become a practical approach. However, existing methods have yet to be explored in the context of imperfect interventions with unknown targets, which are more common and often more beneficial in a wide range of real-world applications. Additionally, the identifiability issues of Granger causality with unknown interventional targets in complex network models remain unsolved. Our work presents a theoretically-grounded method that infers Granger causal structure and identifies unknown targets by leveraging heterogeneous interventional time series data. We further illustrate that learning Granger causal structure and recovering interventional targets can mutually promote each other. Comparative experiments demonstrate that our method outperforms several robust baseline methods in learning Granger causal structure from interventional time series data.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: **How to learn the Granger causal structure from heterogeneous interventional time - series data and identify unknown intervention targets**. Specifically, the paper focuses on how to more accurately infer causal relationships in time - series data in the face of imperfection and unknown intervention targets. ### Background and Challenges 1. **Applications of Granger Causality** - Granger causality is widely used in multiple fields, such as economics, bioinformatics, and geoinformatics. It discovers causal relationships by analyzing time - series data. - Existing methods work well when dealing with linear causal relationships but perform poorly when facing nonlinear causal relationships. 2. **Importance of Intervention Data** - Observational data can only identify the causal structure within the Markov equivalence class (MEC), while interventional data can further narrow this range to the interventional Markov equivalence class (I - MEC), thereby improving the identifiability of the causal structure. - However, most existing methods require known intervention targets, which are often infeasible in practical applications because intervention targets are usually unknown. 3. **Limitations of Existing Methods** - Existing methods face challenges when dealing with time - series data with unknown intervention targets, especially in complex network models, and the identifiability problem of Granger causality has not been solved yet. ### Contributions of the Paper 1. **Task Formalization** - The author formalizes the task of learning the Granger causal structure from heterogeneous interventional time - series data and takes into account the situation where the intervention targets are unknown. 2. **Proposing the IGC Method** - A theoretically guaranteed method - Interventional Granger Causal structure learning (IGC) is proposed. This method can simultaneously infer the Granger causal structure and identify unknown intervention targets. - This method utilizes interventional time - series data across multiple fields and can distinguish between non - intervened and intervened variables in different environments. 3. **Theoretical Guarantee** - It is proved that in the unknown - target setting, exactly minimizing the proposed optimization objective can identify the (I, D)-Markov equivalence class, thus solving the identifiability problem of Granger causality. 4. **Experimental Verification** - Through a large number of experiments, including synthetic data and real - world time - series data, it is verified that the proposed IGC method is superior to several existing robust baseline methods. ### Summary The core of this paper is to propose an IGC method that can handle time - series data with unknown intervention targets, thereby improving the accuracy and reliability of inferring causal relationships in complex situations. This method not only has theoretical breakthroughs but also shows significant advantages in practical applications.

Learning Flexible Time-windowed Granger Causality Integrating Heterogeneous Interventional Time Series Data

Inductive Granger Causal Modeling for Multivariate Time Series

Spatial-Temporal Causality Modeling for Industrial Processes with a Knowledge-Data Guided Reinforcement Learning

CUTS: Neural Causal Discovery from Irregular Time-Series Data

CUTS: Neural Causal Discovery from Unstructured Time-Series Data

Towards the Disappearing Truth: Fine-Grained Joint Causal Influences Learning with Hidden Variable-Driven Causal Hypergraphs in Time Series

Deep Recurrent Modelling of Granger Causality with Latent Confounding

Variable-lag Granger Causality and Transfer Entropy for Time Series Analysis

Detecting High-Dimensional Causal Networks Using Randomly Conditioned Granger Causality

Visual Causality Analysis of Event Sequence Data.

EffCause: Discover Dynamic Causal Relationships Efficiently from Time-Series

Granger causality and time series forecasting

Learning Mixtures of Unknown Causal Interventions

CUTS+: High-dimensional Causal Discovery from Irregular Time-series

Estimating Granger Causality with Unobserved Confounders via Deep Latent-Variable Recurrent Neural Network

Permutation-Based Causal Structure Learning with Unknown Intervention Targets

Dynamic Window-level Granger Causality of Multi-channel Time Series

Partial Granger Causality—eliminating Exogenous Inputs and Latent Variables

A Novel Granger Causality Method Based on HSIC-Lasso for Revealing Nonlinear Relationship Between Multivariate Time Series

Structure Learning from Time Series with False Discovery Control

GCFormer: Granger Causality Based Attention Mechanism for Multivariate Time Series Anomaly Detection