Probabilistic Forecasting via Autoregressive Flow Matching
This work addresses forecasting challenges in domains like dynamical systems and real-world applications, offering a method that combines autoregressive benefits with generative modeling, though it appears incremental as it builds on existing flow matching and autoregressive techniques.
The authors tackled probabilistic forecasting of multivariate time series by proposing FlowTime, a generative model that samples future trajectories from a learned conditional distribution, achieving strong extrapolation performance and well-calibrated uncertainty estimates.
In this work, we propose FlowTime, a generative model for probabilistic forecasting of multivariate timeseries data. Given historical measurements and optional future covariates, we formulate forecasting as sampling from a learned conditional distribution over future trajectories. Specifically, we decompose the joint distribution of future observations into a sequence of conditional densities, each modeled via a shared flow that transforms a simple base distribution into the next observation distribution, conditioned on observed covariates. To achieve this, we leverage the flow matching (FM) framework, enabling scalable and simulation-free learning of these transformations. By combining this factorization with the FM objective, FlowTime retains the benefits of autoregressive models -- including strong extrapolation performance, compact model size, and well-calibrated uncertainty estimates -- while also capturing complex multi-modal conditional distributions, as seen in modern transport-based generative models. We demonstrate the effectiveness of FlowTime on multiple dynamical systems and real-world forecasting tasks.