A1203
Title: On space-time local interaction machines
Authors: Dionissios Hristopulos - Technical University of Crete (Greece) [presenting]
Abstract: The focus is on the development of local interaction machines (SLIMs) for scalar and vector (multi-output) spatiotemporal processes. SLIMs are inspired by statistical physics models and employ sparse precision matrices. Hence, they can significantly improve the computational efficiency of regression tasks compared to the Gaussian process framework. The construction of positive-definite precision matrices that can incorporate general distance measures is discussed. Parameter estimation and regression equations based on these precision matrices are presented, and example datasets are used to illustrate the application of SLIMs.
