Multitask Learning for Estimation of Magnetic Parameters Using Pattern Recognition

Multitask Learning for Estimation of Magnetic Parameters Using Pattern Recognition

Machine learning (ML) approaches present an effective technique for accurately and efficiently predicting device parameters. Using these techniques, we introduce a multi-task convolutional neural network (CNN) model and support vector regression (SVR) model that is intended to precisely estimate two...

Full description

Saved in:
Bibliographic Details
Main Authors: Anubha Sehgal, Shipra Saini, Hemkant Nehete, Kunal Kranti Das, Sourajeet Roy, Brajesh Kumar Kaushik
Format: Article
Language:English
Published: IEEE 2024-01-01
Series:IEEE Open Journal of Nanotechnology
Subjects:
Machine learning
pattern recognition
micromagnetic
Dzyaloshinskii Moriya interaction
parameter estimation
exchange stiffness
Online Access:https://ieeexplore.ieee.org/document/10748362/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Machine learning (ML) approaches present an effective technique for accurately and efficiently predicting device parameters. Using these techniques, we introduce a multi-task convolutional neural network (CNN) model and support vector regression (SVR) model that is intended to precisely estimate two important parameters of magnetic systems such as the Dzyaloshinskii-Moriya interaction (DMI) constant and the exchange constant (A<sub>ex</sub>). The magnetic Hamiltonian encapsulates various energy components, including exchange energy, DMI, Zeeman energy, and anisotropy energy, wherein factors such as saturation magnetization, DMI strength, exchange stiffness, and anisotropy constants influence their magnitudes. Conventionally, the estimation of these parameters has been computationally intensive and time-consuming. The CNN and SVR models can simultaneously estimate both the DMI constant and the exchange constant, making it a versatile tool for magnetic system characterization. The custom CNN model performs best for the DMI constant and A<sub>ex</sub> with R<sup>2</sup> scores of 0.991 and 0.998 respectively. The SVR model achieves R<sup>2</sup> scores of 0.927 and 0.989 for DMI constant and A<sub>ex</sub> respectively. The estimated values are in good agreement with true values, thus emphasizing the potential of ML methods for pattern recognition.
ISSN:2644-1292

Similar Items