Estimating Mutual Information between Time Series and Temporal Event Sequences Across Diverse Analysis Tasks

The article presents a nonparametric mutual information estimator designed to quantify dependence between continuous time series and discrete temporal event sequences without requiring data transformation or discretization. This method addresses issues related to quantization and event redundancy by employing a latent event clustering strategy, resulting in improved accuracy and robustness across various tasks, including causality analysis and feature selection. The proposed framework enhances the analysis of heterogeneous temporal data, positioning it as a versatile tool akin to Pearson correlation for homogeneous datasets, which is significant for practitioners working with diverse temporal data types.