XIE, Antoine (2023) Advanced kernel methods for machine learning: Deep kernel/Indefinite Kernel/Out-of sample extension PRE - Research Project, ENSTA.

[img]
Preview
PDF
1000Kb

Abstract

Kernel methods are a class of methods widely used in machine learning to solve non-linear problems using a linear algorithm. These kernels are generally used via the kernel trick. This technique allows to map the data into a higher dimension space which would be more suited. However, kernels need to satisfy conditions before being used, those conditions are often highly restrictive. One of these conditions is symmetry. But, asymmetry is naturally present in lot of problems and often provides information as in directed graphs. So, the development of algorithms that can work with a wider class of kernels will improve existing algorithms. This report shows how it is possible to learn directly from asymmetric kernels in machine learning algorithms, and seeks to develop new techniques involving asymmetry.

Item Type:Thesis (PRE - Research Project)
Uncontrolled Keywords:Machine learning, Asymmetric kernels, Directed graphs, Least-Square Support Vector Machine, Classification
Subjects:Information and Communication Sciences and Technologies
ID Code:9638
Deposited By:Antoine XIE
Deposited On:28 août 2023 10:05
Dernière modification:28 août 2023 10:05

Repository Staff Only: item control page