Elevated few-shot network intrusion detection via self-attention mechanisms and iterative refinement

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Traditional network intrusion detection systems (NIDS) struggle with adaptability due to large sample requirements, especially for novel and rare attacks.
• Existing few-shot NIDS often fail to capture long-range dependencies in network traffic, limiting their effectiveness against complex threats.

Purpose of the Study:

• To propose a novel elevated few-shot network intrusion detection method.
• To enhance the adaptability and accuracy of NIDS in rapidly changing network environments and complex attack scenarios.

Main Methods:

• Utilized self-attention mechanisms for effective feature extraction and capturing long-range dependencies in network traffic.
• Incorporated positional encoding to preserve temporal sequences and enhance temporal dynamics processing.
• Employed meta-learning with multiple update strategies for initial general training and few-shot fine-tuning.

Main Results:

• Achieved high detection rates of 99.90% on CICIDS2017 and 98.23% on CICIDS2018 datasets.
• Demonstrated effectiveness using only 10 samples, significantly reducing data dependency.
• Showcased improved adaptability and prediction accuracy compared to traditional methods.

Conclusions:

• The proposed elevated few-shot NIDS effectively addresses sample scarcity and improves detection of novel network intrusions.
• Self-attention and iterative refinement offer a robust solution for enhancing network security against evolving cyber threats.