On the properties of random feature methods

Thesis

In this thesis, we study the theoretical properties of the random feature methods, which are initially proposed to address the scalability issues of the kernel methods. Although empirically efficient, existing theoretical analysis of random feature methods often provide pessimistic bounds. Hence, this thesis starts with theoretical investigation of the generalization properties of random feature methods. Our first goal is to understand the statistical consistency of random feature estimato...

Expand abstract

Authors: Li, Z

• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: benign overfitting; kernel methods; supervised learning; random features
• Subjects: Statistics; Machine learning