Title: Scalable and adaptive smooth modelling
Authors: Alkeos Tsokos - UCL (United Kingdom) [presenting]
Ioannis Kosmidis - University of Warwick and The Alan Turing Institute (United Kingdom)
Abstract: An approach is developed to modelling with smooth components that produces locally adaptive fits while being tuning parameter free. The approach relies on representing smooth functions as linear combinations of b-splines whose coefficients are equipped with a particular sparsity inducing prior. Because the approach is tuning parameter free, it scales well with the number of smooth components estimated. We provide an efficient algorithm to compute the estimates and apply the method to additive modelling and scalar on function regression. We demonstrate its performance on simulated data and show some examples of real world use.