Abstract
Intrusion detection systems (IDS) monitor network traffic for malicious activity. However, many AI-based IDSs are too resource-heavy for edge deployment, leaving resource-constrained portions of networks unguarded. The main objective of this study are to develop, optimise and investigate several lightweight AI-IDSs models for resource-constrained network environments, while ensuring that the IDSs maintain a strong detection performance to both known and unknown attacks. The majority of the proposed models, along with their quantised variants, outperformed the baseline 1D CNN model, with several displaying improvements across all evaluation metrics. The CNN-GRU, CNN-LSTM and 2D CNN models proved to be the most viable for real world deployment.
Videos 1
Watch presentations, demos, and related content
Overview of the LAIDS study’s objectives, methodology, experimentation, and results
Like, comment, and subscribe on YouTube to support the creator!
Overview of the LAIDS study’s objectives, methodology, experimentation, and results
Gallery 1
Explore the visual story of this exhibit
Title Image