HEPA: A Self-Supervised Horizon-Conditioned Event Predictive Architecture for Time Series

TL;DR

HEPA is a self-supervised architecture that predicts critical events in multivariate time series by learning temporal dynamics from unlabeled data and fine-tuning for specific event prediction tasks.

Contribution

Introduces HEPA, a novel horizon-conditioned, self-supervised framework that outperforms existing methods across diverse time series event prediction benchmarks.

Findings

HEPA surpasses leading architectures on at least 10 of 14 benchmarks.

Uses an order of magnitude less labeled data and fewer parameters.

Effectively predicts diverse event types across multiple domains.

Abstract

Critical events in multivariate time series, from turbine failures to cardiac arrhythmias, demand accurate prediction, yet labeled data is scarce because such events are rare and costly to annotate. We introduce HEPA (Horizon-conditioned Event Predictive Architecture), built on two key principles. First, a causal Transformer encoder is pretrained via a Joint-Embedding Predictive Architecture (JEPA): a horizon-conditioned predictor learns to forecast future representations rather than future values, forcing the encoder to capture predictable temporal dynamics from unlabeled data alone. Second, we freeze the encoder and finetune only the predictor toward the target event, producing a monotonic survival cumulative distribution function (CDF) over horizons. With fixed architecture and optimiser hyperparameters across all benchmarks, HEPA handles water contamination, cyberattack detection, volatility regimes, and eight further event types across 11 domains, exceeding leading time-series architectures including PatchTST, iTransformer, MAE, and Chronos-2 on at least 10 of 14 benchmarks, with an order of magnitude fewer tuned parameters and, on lifecycle datasets, an order of magnitude less labeled data.