An application of kernel random matrices to the optimization of the parameter in the RBF kernel
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
In recent years, significant progress has been made in the development and application of kernel methods for many practical machine learning and data analysis. The kernel methods are used for a range of problems such as classification, regression, and more general unsupervised learning. The most significant feature of kernel methods is their ability to solve nonlinear problems using linear techniques. This approach avoids complex optimization problems and enables the analysis of large datasets. In this study, we will provide a method to determine the optimal Gaussian kernel parameter and to estimate the number of clusters by applying the theory of random matrices to the kernel matrices. Namely if we can regard a kernel random matrix as Wishart matrix, then the eigenvalue distribution can be approximated by the Marchenko-Pastur distribution of parameter 1, which means that the noise component in a kernel matrix can be extracted by the Marchenko-Pastur distribution of parameter 1. It also leverages the power-law relationship between the kernel parameter and the inter-cluster distance.
Fields of Study & Subjects
Random Matrices and Applications
Advanced Clustering Algorithms Research
Statistical Methods and Applications
Publication details
- Journal / source
- Institutional Repositories DataBase (IRDB)
- Volume
- 76
- Issue
- 1-2
- Pages
- 1–15
- Publication date
- 1 February 2026
- Language
- EN