Informatika Tanszék

Felde

Imre

Szénási Sándor

Log Categorization Strategies for Scarce Data Scenarios

IEEE 25th International Symposium on Computational Intelligence and Informatics (CINTI 2025)

Institute of Electrical and Electronics Engineers (IEEE) (

BUDAPEST

Magyarország

2025.11.18.

2025.11.20.

eural networks have become increasingly widespread in recent years, with growing applications across various industries and scientific fields; however, their effective training typically requires large datasets, which are often difficult to obtain due to practical or physical constraints. In the case of log files, few articles address the problem of limited data, as large software-based systems often facilitate the collection of substantial quantities of high-quality data. However, generating an adequate amount of log data can still be non-trivial and may require a significant amount of time. In this article, we review a range of techniques aimed at reducing training data requirements while preserving accuracy including Data Augmentation, Generative Adversarial Networks, Label Smoothing, Ensemble Learning, Transfer Learning, Born Again Neural Networks, Knowledge Evolution, Dense-SparseDense Training, and Dynamic Neural Regeneration.These techniques offer promising solutions for training neural networks with limited data, and these approaches could be successfully applied to the categorization of a limited number of log files, which could be explored in future work.