Máquina de Aprendizagem Mínima: aspectos teóricos e práticos

Abstract

Minimal learning machine (MLM) is a supervised learning algorithm based on a linear mapping between input and output distance matrices. After the learning step, the corresponding output prediction for new incoming inputs is achieved by solving a multilateration problem. Given its simple formulation, quick training and good accuracy, MLM has been used successfully in several applications. However, it should be noted that such models still need a detailed analysis of some of their steps. This paper presents three contributions regarding the MLM analysis. Initially will be presented the proof of the interpolation capacity of an MLM model. Afterwards, experimental analyzes of the steps of selection of reference points and estimation of the output will be presented.