<i>Loop-the-Loops</i>: Fragmented Learning Over Networks for Constrained IoT Devices

In this work, we propose Timed Loop Gears (TLG), as a distributed method for enabling fragmented learning in Resource-Constrained networked IoT edge devices. TLG identifies atomic operations (gears), such feed-forward and back-propagation, necessary training Machine Learning (ML) models. Each of these gears executes on Fog Node (FN) exclusively each data point at time rather than the whole dataset its entirety. Additionally, Edge Devices (EDs) offload to fog layer using Message Queuing Telemetry Transport (MQTT) protocol that participating FNs subscribe incoming store them based topics, simplifying sharing. enables FN then transfer partially learned weights next suitable further training. This looping is repeated across until complete. Through extensive analysis, observe that, compared existing ML approaches, $n$ devices, reduces probability disruption due device failure by notation="LaTeX">$n^{2}$ times. Implementation results our fragmented learning demonstrate although negligibly increases memory consumption devices notation="LaTeX">$0.8\%$ , it CPU usage almost notation="LaTeX">$90\%$ . The proposed proves beneficial developing hosting models, even constrained contrast lightweight methods.