Filtered Conformal Ellipsoids for Graph-Native Time Series
Researchers have introduced filtered conformal ellipsoids, a novel method for joint prediction sets in multivariate time series. This approach utilizes a state-space filter to emit predictive means and covariances, which are then calibrated using split-conformal methods. The framework aims to control single events while adapting to cross-coordinate dependencies, benefiting from learned predictive covariances without relying on Gaussian tail probabilities for coverage. AI
IMPACT Introduces a new framework for multivariate time series prediction, potentially improving accuracy in complex sequential data analysis.