Sparse-selective quantization for real-time cyber threat detection in large-scale networks

Area of Science:

• Computer Science
• Cybersecurity
• Network Engineering

Background:

• Real-time cyber threat detection in large-scale networks faces challenges in balancing computational efficiency and detection accuracy.
• Existing intrusion detection systems often struggle with high-dimensional network traffic data and identifying rare, severe threats.

Purpose of the Study:

• To propose a sparse-selective quantization framework for efficient and accurate real-time cyber threat detection.
• To enable analysis of high-dimensional network traffic data without losing critical attack signatures.
• To develop a scalable solution compatible with existing infrastructure and suitable for edge deployment.

Main Methods:

• The framework integrates sparsity-aware feature selection and dynamic precision quantization.
• A lightweight deep learning classifier with a GRU-attention mechanism is employed for threat classification.
• Optimization using TensorRT achieves inference latency below one millisecond; implementation with TensorFlow Lite facilitates edge deployment.

Main Results:

• The proposed method demonstrates significant improvements in both speed and accuracy compared to conventional intrusion detection systems.
• Effective detection of rare but severe cyber threats is achieved.
• The framework achieves inference latency below one millisecond, enabling real-time analysis.

Conclusions:

• The unified approach of sparsity analysis, adaptive quantization, and efficient deep learning offers a cohesive and scalable solution for real-time cyber threat detection.
• Linking feature-level sparsity patterns to dynamic quantization policies differentiates this work from prior methods.
• The framework is suitable for edge deployment in resource-constrained environments, advancing the state-of-the-art in network security.