T-Rep: Representation Learning for Time Series using Time-Embeddings

TL;DR

T-Rep is a self-supervised learning method that creates detailed time-aware representations of multivariate time series, improving performance and robustness in classification, forecasting, and anomaly detection tasks.

Contribution

It introduces T-Rep, a novel approach that learns time-specific embeddings to capture temporal features and handle missing data effectively.

Findings

Outperforms existing self-supervised methods in multiple tasks

Demonstrates robustness in missing data scenarios

Provides interpretable latent space visualizations

Abstract

Multivariate time series present challenges to standard machine learning techniques, as they are often unlabeled, high dimensional, noisy, and contain missing data. To address this, we propose T-Rep, a self-supervised method to learn time series representations at a timestep granularity. T-Rep learns vector embeddings of time alongside its feature extractor, to extract temporal features such as trend, periodicity, or distribution shifts from the signal. These time-embeddings are leveraged in pretext tasks, to incorporate smooth and fine-grained temporal dependencies in the representations, as well as reinforce robustness to missing data. We evaluate T-Rep on downstream classification, forecasting, and anomaly detection tasks. It is compared to existing self-supervised algorithms for time series, which it outperforms in all three tasks. We test T-Rep in missing data regimes, where it proves more resilient than its counterparts. Finally, we provide latent space visualisation experiments, highlighting the interpretability of the learned representations.