Distributed Neuromorphic Edge Computing: Theory and Applications in Environmental Monitoring

Neuromorphic computing has emerged as a promising paradigm for energy-efficient AI by mimicking brain-like event-driven processing. In parallel, geo-distributed sensor networks at the edge play a critical role in environmental monitoring and disaster response, requiring real-time, low-power analytics across remote locations.In this work, we combine these thrusts and propose the NeuEdge architecture for geo-distributed neuromorphic edge intelligence. We develop two distributed spiking neural network (SNN) paradigms: (1) Federated SNNs, where multiple edge SNN models are trained locally on sensor data and periodically synchronized, and (2) Split SNNs, where a single SNN is partitioned across networked edge nodes that communicate spikes. We formally define both models and present distributed learning algorithms for each. Theoretical results are derived on communication efficiency, convergence guarantees, and scalability with network size. We then implement simulations on synthetic and real environmental datasets to evaluate both approaches for tasks like environmental event detection and disaster prediction. The federated SNN approach is shown to achieve accuracy comparable to a centralized neuromorphic model while significantly reducing communication vs. raw data offloading. The split SNN approach enables real-time collaborative inference across sensor nodes, outperforming single-node baselines in prediction accuracy for geographically distributed phenomena.