Flow-based Conformal Prediction for Multi-dimensional Time Series

Time series prediction is a crucial task in sequential decision-making. With the increasing use of black-box models for time series prediction, the need for uncertainty quantification has become more critical. Conformal prediction has gained attention as a reliable uncertainty quantification framework. However, conformal prediction for time series faces two key challenges: (1) effectively leveraging sequential correlations in features and non-conformity scores, and (2) handling multi-dimensional outcomes. To address these challenges, we propose a novel conformal prediction method for time series using flow with classifier-free guidance. We provide theoretical guarantees by establishing an exact non-asymptotic marginal coverage and a finite-sample bound on conditional coverage for our method. Evaluations on real-world multi-dimensional time series datasets demonstrate that our method constructs significantly smaller prediction sets while maintaining target coverage, outperforming existing baselines.