Scalable Sample-Level Causal Discovery in Event Sequences via Autoregressive Density Estimation

TL;DR

This paper introduces TRACE, a scalable autoregressive-based framework for causal discovery in single sequences of events, capable of handling high-dimensional data and long-range dependencies efficiently.

Contribution

The paper presents TRACE, a novel method that uses autoregressive density estimation for causal discovery from a single event sequence, supporting delayed effects and large vocabularies.

Findings

Robust performance across various baselines and vocab sizes

Supports delayed causal effects in event sequences

Effective in large-scale applications like vehicle diagnostics

Abstract

We study causal discovery from a single observed sequence of discrete events generated by a stochastic process, as encountered in vehicle logs, manufacturing systems, or patient trajectories. This regime is particularly challenging due to the absence of repeated samples, high dimensionality, and long-range temporal dependencies of the single observation during inference. We introduce TRACE, a scalable framework that repurposes autoregressive models as pretrained density estimators for conditional mutual information estimation. TRACE infers the summary causal graph between event types in a sequence, scaling linearly with the event vocabulary and supporting delayed causal effects, while being fully parallel on GPUs. We establish its theoretical identifiability under imperfect autoregressive models. Experiments demonstrate robust performance across different baselines and varying vocabulary sizes including an application to root-cause analysis in vehicle diagnostics with over 29,100 event types.