Probabilistic Learning of Multivariate Time Series with Temporal Irregularity

TL;DR

This paper introduces a novel probabilistic framework for multivariate time series forecasting that effectively handles temporal irregularities and variable misalignments using continuous normalizing flows and a factorized likelihood.

Contribution

It presents an end-to-end model that captures complex joint distributions at arbitrary times, addressing limitations of uniform sampling assumptions in prior methods.

Findings

Outperforms existing methods on real-world datasets

Effectively models non-Gaussian data distributions

Handles irregular sampling and misaligned variables

Abstract

Probabilistic forecasting of multivariate time series is essential for various downstream tasks. Most existing approaches rely on the sequences being uniformly spaced and aligned across all variables. However, real-world multivariate time series often suffer from temporal irregularities, including nonuniform intervals and misaligned variables, which pose significant challenges for accurate forecasting. To address these challenges, we propose an end-to-end framework that models temporal irregularities while capturing the joint distribution of variables at arbitrary continuous-time points. Specifically, we introduce a dynamic conditional continuous normalizing flow to model data distributions in a non-parametric manner, accommodating the complex, non-Gaussian characteristics commonly found in real-world datasets. Then, by leveraging a carefully factorized log-likelihood objective, our approach captures both temporal and cross-sectional dependencies efficiently. Extensive experiments on a range of real-world datasets demonstrate the superiority and adaptability of our method compared to existing approaches.