[2402.03726] Studying Granger Causality from Occasion-wise Self-attentive Hawkes Processes

Obtain a PDF of the paper titled Studying Granger Causality from Occasion-wise Self-attentive Hawkes Processes, by Dongxia Wu and seven different authors

Obtain PDF
HTML (experimental)

Summary:We deal with the issue of studying Granger causality from asynchronous, interdependent, multi-type occasion sequences. Particularly, we’re fascinated with discovering instance-level causal buildings in an unsupervised method. Occasion-level causality identifies causal relationships amongst particular person occasions, offering extra fine-grained data for decision-making. Current work within the literature both requires sturdy assumptions, equivalent to linearity within the depth perform, or heuristically outlined mannequin parameters that don’t essentially meet the necessities of Granger causality. We suggest Occasion-wise Self-Attentive Hawkes Processes (ISAHP), a novel deep studying framework that may straight infer the Granger causality on the occasion occasion stage. ISAHP is the primary neural level course of mannequin that meets the necessities of Granger causality. It leverages the self-attention mechanism of the transformer to align with the rules of Granger causality. We empirically reveal that ISAHP is able to discovering complicated instance-level causal buildings that can’t be dealt with by classical fashions. We additionally present that ISAHP achieves state-of-the-art efficiency in proxy duties involving type-level causal discovery and instance-level occasion sort prediction.