Probabilistic Learning of Multivariate Time Series with Temporal Irregularity
This addresses a practical problem for applications relying on multivariate time series forecasting, such as finance or healthcare, by handling real-world irregularities, though it is incremental as it builds on existing probabilistic methods.
The paper tackles probabilistic forecasting of multivariate time series with temporal irregularities like nonuniform intervals and misaligned variables, proposing an end-to-end framework that models these irregularities and captures joint distributions at arbitrary continuous-time points, achieving superior performance in experiments on real-world datasets.
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.