This work attempts to give an overview over the modern questions in statistical learning sparked by the success of deep learning. In particular, benign overfitting and overparameterization are discussed by means of kernel ridge(less) regression. Due to recent interest in kernel methods in this field, the underlying theory is developed in-depth. Finally, a simple model is proposed that seems to reveal many of the interesting aspects of deep learning, and this is demonstrated experimentally by a brief discussion of the "double descent" phenomenon.

High-capacity hypothesis spaces in modern statistical learning

WELLMEIER, LUCA
2022/2023

Abstract

This work attempts to give an overview over the modern questions in statistical learning sparked by the success of deep learning. In particular, benign overfitting and overparameterization are discussed by means of kernel ridge(less) regression. Due to recent interest in kernel methods in this field, the underlying theory is developed in-depth. Finally, a simple model is proposed that seems to reveal many of the interesting aspects of deep learning, and this is demonstrated experimentally by a brief discussion of the "double descent" phenomenon.
2022
High-capacity hypothesis spaces in modern statistical learning
benign overfitting
reproducing kernel
regularized least
Sobolev spaces
bias-variance trade-
File in questo prodotto:
File Dimensione Formato  
thesis.pdf

accesso aperto

Dimensione 1.01 MB
Formato Adobe PDF
1.01 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/46193