Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Granger Causality from Instance-wise Self-attentive Hawkes Processes

Dongxia Wu, Tsuyoshi Idé, Aurélie Lozano, Georgios Kollias, Jiří Navrátil, Naoki Abe, Yi-An Ma, Rose Yu

TL;DR

This paper tackles learning Granger causality in asynchronous, multi-type event sequences by proposing ISAHP, a neural point process that preserves additive structure while employing instance-aware embeddings through self-attention. By parameterizing the kernel and decay terms with per-event embeddings and attention-guided interactions, ISAHP enables direct instance-level causal inference and coherent aggregation to type-level causality. Empirically, ISAHP achieves state-of-the-art results on type-level causal discovery and instance-level event-type prediction, while providing qualitative evidence of correct instance-level causal reasoning and synergistic effects. The approach offers practical value for fine-grained causal analysis in domains like medical diagnosis and real-world event data, and it reduces computational overhead by avoiding post-hoc attribution steps common in competing methods.

Abstract

We address the problem of learning Granger causality from asynchronous, interdependent, multi-type event sequences. In particular, we are interested in discovering instance-level causal structures in an unsupervised manner. Instance-level causality identifies causal relationships among individual events, providing more fine-grained information for decision-making. Existing work in the literature either requires strong assumptions, such as linearity in the intensity function, or heuristically defined model parameters that do not necessarily meet the requirements of Granger causality. We propose Instance-wise Self-Attentive Hawkes Processes (ISAHP), a novel deep learning framework that can directly infer the Granger causality at the event instance level. ISAHP is the first neural point process model that meets the requirements of Granger causality. It leverages the self-attention mechanism of the transformer to align with the principles of Granger causality. We empirically demonstrate that ISAHP is capable of discovering complex instance-level causal structures that cannot be handled by classical models. We also show that ISAHP achieves state-of-the-art performance in proxy tasks involving type-level causal discovery and instance-level event type prediction.

Learning Granger Causality from Instance-wise Self-attentive Hawkes Processes

TL;DR

This paper tackles learning Granger causality in asynchronous, multi-type event sequences by proposing ISAHP, a neural point process that preserves additive structure while employing instance-aware embeddings through self-attention. By parameterizing the kernel and decay terms with per-event embeddings and attention-guided interactions, ISAHP enables direct instance-level causal inference and coherent aggregation to type-level causality. Empirically, ISAHP achieves state-of-the-art results on type-level causal discovery and instance-level event-type prediction, while providing qualitative evidence of correct instance-level causal reasoning and synergistic effects. The approach offers practical value for fine-grained causal analysis in domains like medical diagnosis and real-world event data, and it reduces computational overhead by avoiding post-hoc attribution steps common in competing methods.

Abstract

We address the problem of learning Granger causality from asynchronous, interdependent, multi-type event sequences. In particular, we are interested in discovering instance-level causal structures in an unsupervised manner. Instance-level causality identifies causal relationships among individual events, providing more fine-grained information for decision-making. Existing work in the literature either requires strong assumptions, such as linearity in the intensity function, or heuristically defined model parameters that do not necessarily meet the requirements of Granger causality. We propose Instance-wise Self-Attentive Hawkes Processes (ISAHP), a novel deep learning framework that can directly infer the Granger causality at the event instance level. ISAHP is the first neural point process model that meets the requirements of Granger causality. It leverages the self-attention mechanism of the transformer to align with the principles of Granger causality. We empirically demonstrate that ISAHP is capable of discovering complex instance-level causal structures that cannot be handled by classical models. We also show that ISAHP achieves state-of-the-art performance in proxy tasks involving type-level causal discovery and instance-level event type prediction.
Paper Structure (29 sections, 13 equations, 2 figures, 6 tables)

This paper contains 29 sections, 13 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Type-Level Granger Causality
  4. Instance-Level Granger Causality
  5. Related Work
  6. Multivariate Hawkes process.
  7. Neural Point Processes.
  8. Instance-wise Self-attentive Hawkes Processes
  9. Intensity Function
  10. Self-attentive Architecture
  11. Self-Attention.
  12. Kernel matrix, background intensity, and decay rate functions
  13. Maximum Likelihood Estimation
  14. Experiments
  15. Experimental Set-up
  16. ...and 14 more sections

Figures (2)

  • Figure 1: Instance-wise Self-attentive Hawkes Processes (ISAHP) architecture. The input is an event sequence identified by timestamp and event type. the output is an intensity function parameterized by background intensity, kernel matrix, and decay rate.
  • Figure 2: Instance-level causality analysis. The weight of the edge from the first event to the third is what we compare for the synergistic (left) and non-synergistic (right) sequences. Red numbers represent successful cases and blue numbers represent failure cases. ISAHP is the only one that successfully captures the synergistic effect at the instance level.