Transformer-Based Artificial Intelligence for Computer Networking: A Review of Network Traffic Modeling, Security, Prediction, and Automation

Abstract

Transformer architectures have transformed artificial intelligence by enabling effective learning of long-range dependencies in sequential and structured data. Their success in natural language processing and multimodal learning has motivated increasing research into their application to computer networking. Network traffic, logs, telemetry streams, configuration files, and security events contain temporal and structural dependencies that can potentially be modeled using transformer-based architectures. This review investigates the emerging use of transformers in computer networking, focusing on traffic prediction, anomaly detection, intrusion detection, traffic classification, network performance forecasting, configuration analysis, and network automation. Different transformer architectures and adaptation strategies are examined in relation to networking requirements such as low latency, high throughput, scalability, and continuous learning. The review compares transformers with recurrent neural networks, convolutional models, graph neural networks, and conventional machine learning techniques. Challenges involving computational complexity, model size, training data, interpretability, real-time inference, and adversarial robustness are analyzed. The paper concludes with future opportunities for lightweight transformers, graph-transformer hybrids, network foundation models, and AI-native autonomous networking.