Flexible Regression in Neural Networks
Quentin Edward Seifert, University of Göttingen, Germany
Co-authors: Tobias Hepp, Friedrich-Alexander-Universität Erlangen-Nuremburg; Benjamin Säfken, TU Clausthal; Elisabeth Bergherr, University of Göttingen
Abstract: We propose a series of nonparametric regression models based on gradient descent based training methods to tackle challenges that can arise in classical flexible regression settings. Our applications include the estimation of pediatric reference intervals using finite mixture regression models, probabilistic load forecasting on very large smart meter data using simultaneous quantile regression and the analysis of parking data using functional regression.