Machine learning and deep learning algorithms for prediction of different parameters of a 2D permanent magnetic lattice

Abstract

In this paper, we consider artificial intelligence algorithms in Machine Learning (ML) and Deep Learning (DL) models as an innovative method for improving computational efficiency to predict a large amount of data related to the microtrap parameters of a 2D permanent magnetic lattice. This periodic array is created by magnetic slabs in the x and y directions, which are influenced by three components of an external bias magnetic field (B1x, B1y, B1z). By considering these components as input data and changing B1z, the central minimum coordinate of magnetic microtraps (xmin,ymin,zmin) and magnetic field minima Bmin are obtained as output data. We assume 60% of the data for training and the remaining 40% for validation. By using the ML algorithms such as linear model, Decision Tree (DT), random forest, Gradient Boosting Tree (GBT), and multi-layer perceptron from DL algorithm, we show that the GBT and DT algorithms represent the best prediction results by evaluation metrics R2 between 0.998 and 1 for all quantities and mean squared error value 4.1266×10−7 for zmin. Other algorithms demonstrate different results for the considered parameters that will be discussed.