Abstract
Discovering the underlying Directed Acyclic Graph (DAG) from time seriesobservational data is highly challenging due to the dynamic nature and complexnonlinear interactions between variables. Existing methods typically search forthe optimal DAG by optimizing an objective function but face scalabilitychallenges, as their computational demands grow exponentially with thedimensional expansion of variables. To this end, we propose LOCAL, a highlyefficient, easy-to-implement, and constraint-free method for recovering dynamiccausal structures. LOCAL is the first attempt to formulate a quasi-maximumlikelihood-based score function for learning the dynamic DAG equivalent to theground truth. Building on this, we introduce two adaptive modules that enhancethe algebraic characterization of acyclicity: Asymptotic Causal Mask Learning(ACML) and Dynamic Graph Parameter Learning (DGPL). ACML constructs causalmasks using learnable priority vectors and the Gumbel-Sigmoid function,ensuring DAG formation while optimizing computational efficiency. DGPLtransforms causal learning into decomposed matrix products, capturing dynamiccausal structure in high-dimensional data and improving interpretability.Extensive experiments on synthetic and real-world datasets demonstrate thatLOCAL significantly outperforms existing methods and highlight LOCAL'spotential as a robust and efficient method for dynamic causal discovery.