A0484
Title: Function-on-scalar regression via first-order gradient-based optimization
Authors: Quentin Edward Seifert - Georg-August-Universitaet Goettingen (Germany) [presenting]
Elisabeth Bergherr - Georg-August-Univerität Göttingen (Germany)
Tobias Hepp - University of Erlangen-Nuremberg (Germany)
Abstract: Functional regression models allow for the inclusion of functional covariates and responses. Due to the nature of the data, the models can quickly become computationally expensive, even with a comparably low number of observations. To address this increased complexity, gradient descent-based functional regression is introduced. The idea is to fit functional regression models using gradient descent-based optimization algorithms and estimate the model parameters as one would estimate the parameters of neural networks. The proposed model provides an easily scalable and customizable alternative to established approaches. Preliminary simulation results show that the approach performs reliably. The approach is applied to supermarket parking data recorded during the first months of the Covid-19 pandemic in Germany to analyze the effect of the contact restrictions introduced during this period on consumer behavior.
