Intrusion Detection System with Domain-incremental Continual Learning

In the ever-evolving cybersecurity field, deep learning-based intrusion detection systems must continually adapt to new types of cyberattacks while preserving knowledge of previously encountered ones. This paper introduces the recently proposed CLiCK approach for domain-incremental cyberattack scenarios, where the characteristics of network traffic caused by emerging threats, evolving attack techniques, and shifting user behaviors are continuously changing. Experimental results demonstrated that the proposed method incrementally learns from new attack domains, with a low degree of catastrophic forgetting, allowing it to generalize across evolving cyberattacks.