Edgy Organism (2024-2026)
Project Overview
Edgy Organism develops electronic surveillance systems utilising spiking neural networks (SNNs) to detect and alert anomalous behaviour or identify individuals posing a threat in crowded public spaces. The system employs neuromorphic hardware such as Intel Loihi and True North, drawing inspiration from biological nervous systems to create autonomous, low-power, and efficient monitoring systems. The project demonstrates its capabilities through two operational scenarios: Stand-off, which involves satellite-based observation of patterns, and Leave-behind, which involves covert proximity surveillance. By leveraging neuromorphic computing architectures, the project aims to overcome the power and latency constraints of conventional deep learning approaches deployed on edge devices. The resulting system is designed for deployment in resource-constrained environments where continuous, real-time threat assessment is required without dependence on cloud connectivity.
The system applies principles from neuroscience to develop advanced anomaly detection capabilities. Drawing on the visual cortex and hippocampal-entorhinal grid-cell systems, it builds high-dimensional cognitive maps capable of representing and detecting anomalies in patterns of life. These biologically inspired representations enable the system to identify threats through spatial and temporal pattern recognition rather than conventional classification approaches.
This research is funded by EPSRC under the UK Defence and Security programme. It is conducted in collaboration with DSTL, HMGCC, and GCHQ, ensuring that the technology is developed with input from defence and intelligence stakeholders who understand the operational requirements of real-world surveillance and security scenarios.
Team
Funding
Partners
Repositories
ANTShapes
Anomalous Neuromorphic Tool for Shapes — a Unity-based simulator for configurable, labelled event-vision datasets used for neuromorphic anomaly detection and object-classification benchmarking.
snn-stdp-object-detection
NCARS STDP Reservoir — Unsupervised STDP feature extraction with spiking reservoir computing for event-based vision. Achieves 82.6% accuracy on NCARS dataset using convolutional STDP + ridge readout.