Zhenyu-LIAO / RMT4LSSVM

A Random Matrix Approach for Least Squares SVM Analysis

RMT4LS-SVM

A Random Matrix Approach to Least Squares SVM

This page contains a simple demo using Python 3 of the theoretical results in the following paper:

A Large Dimensional Analysis of Least Squares Support Vector Machines

where recent advances in matrix matrix theory are used to analyze the performance of LS-SVM, a variant of classical SVM.

About the code

Comparison between theory and practice is available for data from

• MNIST database
• Gaussian mixture model

for Gaussian and ploynomial kernels.

Dependencies

To be able to test this code requires the following:

• Python: tested with version 3.6
• Numpy and Scipy
• Matplotlib for visulazation
• Scikit-learn for MNIST dataset

We strongly recommend you to use Jupyter nootbook to have a direct illustration within your web browsers: here.

Contact information

• Zhenyu LIAO
  • Ph.D. student at CentraleSupelec, Paris, France
  • Website: https://zhenyu-liao.github.io/
  • E-mail: [e-mail]
• Prof. Romain COUILLET
  • Professor at CentraleSupelec, Paris, France
  • Website: http://romaincouillet.hebfree.org/
  • E-mail: [e-mail]