Shallow vs. Deep Learning: Prioritizing Efficiency in Next Generation Networks
Rafael Bastos Teixeira, Leonardo Almeida, Pedro Rodrigues, Mário Antunes, Diogo Gomes, Rui L. Aguiar · 2024
With the exponential growth of mobile network traffic and diverse application demands, traditional network management methods struggle to meet the stringent requirements of 5th Generation (5G) and Beyond-5G (B5G) networks. Artificial Intelligence (AI), particularly Deep Learning (DL), has emerged as a promising solution. However, large and complex DL models can be computationally expensive for real-time applications. This paper investigates the potential of shallow Machine Learning (ML) models for 5G/B5G tasks. We compare the performance and training/inference time of shallow ML models with state-of-the-art DL models in two key tasks: network slicing attribution and Intrusion Detection Systems (IDS). The results demonstrate that shallow models achieve comparable performance with significantly faster training and prediction, leading to an acceleration of over 90% in most cases.