Implementing a batch size hyperparameter with logic tensor networks to predict the visibility of the aurora

Authors

Erik Mayer, Pittsburg State UniversityFollow
Jhonatan Granadeno, Pittsburg State University
Jacob Luton, Pittsburg State University
Ty Woolven, Pittsburg State University
Estevan Hernandez, Black & Veatch
Tao Wu, Bethel University
Hongsheng He, University of Alabama
Elizabeth MacDonald, Goddard Space Flight Center

Files

Description

In this research, Logic Tensor Networks (LTNs) are used to predict the visibility of the aurora. LTNs implement first-order logic using neural networks. LTNs were used in the classification of conditions for which the aurora could be visible. However, due to the large amount of training data, the LTNs experienced convergence issues during training. To resolve these issues, a batch size hyperparameter was implemented for which only a subset of the training vectors was used each epoch. This allowed the training of the LTNs to successfully converge.

Department of Primary Author

Automotive and Engineering Technology

Affiliation of Primary Author

Faculty

Publication Date

2026

Recommended Citation

Mayer, Erik; Granadeno, Jhonatan; Luton, Jacob; Woolven, Ty; Hernandez, Estevan; Wu, Tao; He, Hongsheng; and MacDonald, Elizabeth, "Implementing a batch size hyperparameter with logic tensor networks to predict the visibility of the aurora" (2026). Posters. 9.
https://digitalcommons.pittstate.edu/ai-posters-2026/9