Transformer-Based Multi-Source Transfer Learning for Intrusion Detection Models with Privacy and Efficiency Balance.

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Current intrusion detection systems struggle with cross-domain adaptability, privacy preservation, and detecting rare attacks.
• Existing methods often fail to generalize across different network environments and protect sensitive data.

Purpose of the Study:

• To propose a novel intrusion detection model framework, TrMulS, that addresses limitations in cross-domain adaptability, privacy, and minority-class attack detection.
• To develop a robust system for enhanced security intelligence analysis.

Main Methods:

• TrMulS integrates federated learning, generative adversarial networks (GANs) with multispace feature enhancement, and transformers with multi-source transfer.
• Local feature extraction via CNNs, subset construction, and transformer-based attention mechanisms are employed.
• Cross-domain transfer is achieved using an improved Exchange-GAN and Maximum Mean Discrepancy (MMD) to minimize feature distribution differences.
• A federated transfer learning strategy with encrypted parameter aggregation ensures privacy and global model optimization.

Main Results:

• Experiments on ISCX2012, KDD99, and NSL-KDD datasets demonstrate significant improvements in detection accuracy within cross-domain scenarios.
• The TrMulS framework shows superior performance compared to existing methods in challenging detection environments.
• The proposed method effectively balances efficiency, privacy, and detection accuracy.

Conclusions:

• TrMulS offers a novel paradigm for cross-domain security intelligence analysis, overcoming key limitations of current intrusion detection methods.
• The integration of advanced AI techniques provides a more adaptable, private, and effective solution for network security.
• This framework represents a significant advancement in building resilient and intelligent intrusion detection systems.