Nonparametric regression lies in the space between simplistic linear models and opaque machine learning tools. Nonparametric models are flexible, allowing the data to dictate the shape of relationships between variables, but are also easier to interpret, and hence allow for greater insight into what may be learned from them, when compared to, e.g., deep learning. This lecture provides an overview of nonparametric regression, focusing on some methods, and a bit of the underlying statistical theory. There will be a discussion of the implications of the curse of dimensionality, and how the additive nonparametric model can be a particularly appealing option to circumvent this "curse." The relationship between additive models and simple neural network architectures is instructive.
The seminar will be done remote only.
M. Girone, M. Elsing, L. Moneta, M. Pierini
Event co-organised with the PHYSTAT Committee