Sparse kernel learning with LASSO and Bayesian inference algorithm.

Area of Science:

• Machine Learning
• Statistical Modeling
• Kernel Methods

Background:

• Kernelized LASSO has been explored in recent literature.
• Existing methods for sparse regression include Relevance Vector Machines (RVM) and Local Regularization Assisted Orthogonal Least Squares Regression (LROLS).

Purpose of the Study:

• To develop a generative Bayesian learning and inference approach for learning kernels under the LASSO formulation.
• To propose a new robust learning algorithm for sparse kernel models.

Main Methods:

• Adopting a generative Bayesian learning and inference approach.
• Developing a novel algorithm for learning regularized parameters and kernel hyperparameters.
• Comparing the proposed method with RVM and LROLS.

Main Results:

• The proposed algorithm produces a sparse kernel model.
• The algorithm demonstrates capability in learning regularized parameters and kernel hyperparameters.
• The new algorithm exhibits considerable computational advantages over existing methods.

Conclusions:

• The proposed Bayesian learning approach offers an effective method for kernelized LASSO.
• The algorithm provides a robust and computationally efficient solution for sparse kernel model construction.
• This work advances sparse regression techniques through a novel kernel learning framework.