Scalable state-space models for spatio-temporal geostatistics

Jacopo Rodeschini (University of Bergamo)

Wednesday 25th March 12:00-12:30
Maths 311B

Abstract

This presentation introduces a novel low-rank approximation of the State-Space Model (SSM) with spatially correlated innovations for multivariate spatio-temporal data. Dimensionality reduction is achieved by embedding the Stochastic Partial Differential Equation (SPDE) approach, providing a finite-dimensional representation of the innovation processes. Parameter estimation and likelihood derivatives are implemented in the Python/JAX package geossm, enabling automatic differentiation and scalable execution across CPU cores with native GPU and TPU support.

Add to your calendar

Download event information as iCalendar file (only this event)