Vitenskapelig artikkel   2021

Hubin, Aliaksandr; Storvik, Geir Olve; Frommlet, Florian

Publikasjonsdetaljer

Tidsskrift:

The journal of artificial intelligence research, vol. 72, p. 901–942–41, 2021

Internasjonale standardnumre:

Trykt: 1076-9757
Elektronisk: 1943-5037

Lenker:

SAMMENDRAG: jair.org/index.php/jair/article/view/13047
FULLTEKST: jair.org/index.php/jair/article/view/13047/26740
DOI: doi.org/10.1613/JAIR.1.13047

Regression models are used in a wide range of applications providing a powerful scientific tool for researchers from different fields. Linear, or simple parametric, models are often not sufficient to describe complex relationships between input variables and a response. Such relationships can be better described through flexible approaches such as neural networks, but this results in less interpretable models and potential overfitting. Alternatively, specific parametric nonlinear functions can be used, but the specification of such functions is in general complicated. In this paper, we introduce a flexible approach for the construction and selection of highly flexible nonlinear parametric regression models. Nonlinear features are generated hierarchically, similarly to deep learning, but have additional flexibility on the possible types of features to be considered. This flexibility, combined with variable selection, allows us to find a small set of important features and thereby more interpretable models. Within the space of possible functions, a Bayesian approach, introducing priors for functions based on their complexity, is considered. A genetically modified mode jumping Markov chain Monte Carlo algorithm is adopted to perform Bayesian inference and estimate posterior probabilities for model averaging. In various applications, we illustrate how our approach is used to obtain meaningful nonlinear models. Additionally, we compare its predictive performance with several machine learning algorithms.