BrainShield: A Neuromorphic Intrusion Detection System Inspired by the Drosophila melanogaster Whole-Brain Connectome

Abstract

Abstract—This paper presents BrainShield, an Intrusion De- tection System whose architecture is derived from the whole- brain connectome of Drosophila melanogaster. The system models 9,987 Izhikevich spiking neurons across 64 populations in 12 neuropil regions. We propose Top-4 Attentive Aggregation, a scoring mechanism that selects the four most anomalous features per analysis window to resist signal dilution in concentrated attacks. The system was evaluated on raw PCAP traffic from three days of the CIC-IDS2017 dataset (over 22 million packets), achieving 99.9% precision on volumetric attacks and successfully flagging brute-force campaigns that conventional threshold-based detectors missed. Analytical energy estimates suggest the topology could run at roughly 0.6 nJ per inference window on Intel Loihi 2. Index Terms—Neuromorphic Computing, Spiking Neural Networks, Intrusion Detection, Drosophila Connectome, Intel Loihi 2, Adversarial Robustness, Energy-Efficient AI

Abstract of "BrainShield: A Neuromorphic Intrusion Detection System Inspired by the Drosophila melanogaster Whole-Brain Connectome" by Mohammed Saddam Ruwayd, https://doi.org/10.5281/zenodo.18986956, used under its Creative Commons Attribution license (CC-BY); markup removed, otherwise unmodified.