search for




 

The learning of artficial neural network using density power divergence - Focusing on regression problem
Journal of the Korean Data & Information Science Society 2024;35:411-20
Published online May 31, 2024;  https://doi.org/10.7465/jkdi.2024.35.3.411
© 2024 Korean Data and Information Science Society.

Moosup Kim1

Department of Statistics, Keimyung University
Correspondence to: This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT): Grant No. RS-2023-00243752
1 Assistant professor, Department of Statistics, Keimyung University, Daegu 42601, Korea. E-mail: moosupkim@kmu.ac.kr
Received March 2, 2024; Revised April 5, 2024; Accepted April 9, 2024.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
The learning of artificial neural network implicitly assumes the integrity of data. However, real data may be contaminated by outliers, which leads to the performance degradation of the resulting model. Therefore, a robust learning method is required for training the model using contaminated data. This paper considers the learning method based on density power divergence with respect to regression problem. A robust loss function is derived by applying minimum density power divergence principle to the problem and the learning algorithm is adapted to it. A simulation study shows that the proposed method is indeed robust. Moreover, an empirical rule of the tuning parameter selection is dealt with.
Keywords : Artificial neural network, density power divergence, regression problem, robustness, tuning parameter