Magnetic Declination Statistics (Dusheti 1935-1989) and Deep Self-Learning Model

Authors

  • Tengiz V. Kiria
  • Abesalom A. Esakia
  • Manana M. Nikolaishvili
  • Elene D. Lomadze

Keywords:

Earth’s magnetic field, Adam Deep Learning network.

Abstract

Nowadays, there are many new instruments available for studying the parameters of the Earth’s magnetic field with higher precision and more discretization. Moreover, data processing techniques have been developing on strong mathematical basis. The paper presents a rather long-term (1935 - 1950, total 19332 records) data of Dusheti Observatory on the statistics of magnetic declination (1) and considers the possibilities of so called machine learning (ML), a widespread method nowadays. It gives hypotheses to prove certain hidden regularities and periodicity of some geomagnetic parameters and determines so called “storages” of high statistical reliability, which are the etalon samples we use to build attribution function by use of so called Adam Deep Learning network (2

References

Glatzmaiers G.A., Roberts P.H. Three-dimensional self-consistent computer simulation of a geomagnetic field reversal. Nature. 377(6546), Х, 1995, pp. 203–209.

Loshchilov I., Hutter F. Fixing Weight Decay Regularization in Adam. Archives: 1711.05101 v. 2, 2017.

Kingma D.P., Ba J. A Method for Stochastic Optimization. 22 Dec 2014 (v.1), last revised 30 Jan 2017 (this version, v. 9).

Anastasis K. Deep Arbitrage-Free Learning in a Generalized HJM Framework via Arbitrage-Regularization Data". 2020.

Downloads

Published

2021-10-11

How to Cite

Kiria, T. V. ., Esakia, A. A. ., Nikolaishvili, M. M., & Lomadze, E. D. . (2021). Magnetic Declination Statistics (Dusheti 1935-1989) and Deep Self-Learning Model. JOURNAL OF THE GEORGIAN GEOPHYSICAL SOCIETY, 24(1). Retrieved from https://openjournals.ge/index.php/GGS/article/view/2877