| Abstract |
The Internet of Things (IoT) produces an unprecedented amount of data, generated by billions of connected devices. Due to the distributed nature of IoT devices, datasets are distributed and it is often infeasible to move all the locally collected data to a centralized location. Bandwidth and storage are too limited for the transmission of raw data, or such transmission can be prohibited due to privacy constraints. Due to these constraints, distributed machine learning algorithms which work on local datasets with limited global coordination are needed. In this demonstration, we present an distributed learning system that enables edge devices to collaboratively learn a shared model while keeping all the raw data stored distributedly at the edge. The system estimates parameters related to data distribution and resource consumption, and adapts the learning process based on these estimations in real time. |