Edge Computing for Machine LearningJoan Clarke Stage
The landscape of big data applications is changing rapidly: large centralized datasets are replaced by high volume, high velocity data streams generated by a vast number of geographically distributed, loosely connected devices such as mobile phones, autonomous vehicles or industrial machines. Current parallel learning approaches are not designed for such highly distributed systems. Therefore, a new paradigm for parallelization is emerging that treats the learning algorithm as a black box, training local models and aggregating them into a single strong one. The approach is highly scalable, communication-efficient, and privacy-preserving, since it does not require to exchange local data. It enables novel applications by minimizing computation and communication costs, both highly relevant for autonomous driving and learning on mobile phones. It also allows to learn from privacy-sensitive data that is otherwise protected by corporate secrecy, such as sensor data from industrial machines.
Cédric Villani, Mathematician - member of the Academy of Sciences - 1st vice-president of the Parliamentary Office For Scientific and Technological Assessment (OPECST) – Member of Parliament Fields Medal,