Speaker
Anders Kvellestad
(University of Oslo)
Description
Gaussian processes regression provides a principled Bayesian approach to regression problems. But many applications of GP regression are hampered by the fact that the training time for a naive GP implementation scales as the number of data points cubed. In this talk I will present a new software package, GPTreeO, that enables continual learning with GPs, i.e. a GP regression model that continually adapts to a never-ending data stream, and I will discuss how GPTreeO can be used in computationally expensive parameter estimation studies.
Primary author
Anders Kvellestad
(University of Oslo)
