Título: On the use of kernel functions in Minimal Learning Machines
Autores: Freire, Ananda;Souza, Amauri
The Minimal Learning Machine is a recently proposed supervised method in which learning consists of fitting a multiresponse linear regression model between distances computed from the input and output spaces. This paper approaches the use of Minimal Learning Machines in combination with positive definite kernel functions. Particularly, we investigate if computing distances in feature spaces rather than in the original data space is beneficial for the MLM in the context of regression and classification tasks. This can be accomplished since the kernel trick allows us to calculate inner products (and consequently Euclidean distances) in feature spaces. We compare the standard MLM to its kernel variants on real-world problems.
Código DOI: 10.21528/CBIC2017-98
Artigo em pdf: cbic-paper-98.pdf
Arquivo BibTeX: cbic-paper-98.bib