Probabilistic Forecasting of Irregular Time Series via Conditional Flows
This addresses forecasting challenges in fields like health care and astronomy, offering a novel method for handling irregular data with missing values, though it is incremental in improving existing probabilistic approaches.
The paper tackles probabilistic forecasting of irregularly sampled multivariate time series with missing values by proposing ProFITi, a model using conditional normalizing flows that learns joint distributions without fixed-shape assumptions, achieving 4 times higher likelihood than the previous best model.
Probabilistic forecasting of irregularly sampled multivariate time series with missing values is an important problem in many fields, including health care, astronomy, and climate. State-of-the-art methods for the task estimate only marginal distributions of observations in single channels and at single timepoints, assuming a fixed-shape parametric distribution. In this work, we propose a novel model, ProFITi, for probabilistic forecasting of irregularly sampled time series with missing values using conditional normalizing flows. The model learns joint distributions over the future values of the time series conditioned on past observations and queried channels and times, without assuming any fixed shape of the underlying distribution. As model components, we introduce a novel invertible triangular attention layer and an invertible non-linear activation function on and onto the whole real line. We conduct extensive experiments on four datasets and demonstrate that the proposed model provides $4$ times higher likelihood over the previously best model.