Federated K-Means Clustering: A Novel Edge AI Based Approach for Privacy Preservation

Edge AI is the result of a more fundamental thought, “bring code to data, not data to code”. The assumption here is that the data is stored on devices which are computationally powerful and are therefore `edge devices'. Our paper deals with the technique of Federated Averaging applied to `Federated k-Means' algorithm. To the best of our knowledge, at the time of writing, there is no published research outcome of applying Federated k-Means, a computing paradigm aligned towards this concept of Edge AI. The data generated at the edge device never leaves it, thus ensuring privacy preservation. Each device has a ML model on it, thus reducing the latency associated with a centralized model deployed on the cloud. Our paper aims to provide a holistic analysis of the performance metrics at the edge devices as well as at the central server after each federation step. Ours is an early publication of the research results, aiming to excite the researcher community on the immense possibilities to explore this new area, a possible solution to the concerns of user data privacy and the inherent latency associated with traditional ML approaches. The findings of our paper can potentially be applied to a privacy-preserving real-world use case, such as EMR (Electronic Medical Record) based ML applications.