Publication

Abstract : The traditional machine learning algorithms focus on centralised data repository where the aggregate data used for training is stored in a common location and processed. This approach is not suitable when data is stored in different locations and owned by different entities. Many crucial machine learning applications need computationally efficient and privacy-preserving solution. Also the central data repository has the risk of single point of failure. Federated learning is an emerging field in machine learning where the centralised concept is changed to distributed. Federated learning approach helps to train a model in machine learning without really sharing the data to a common server. In this approach, training is done locally at client side. A technique called federated averaging is applied at server side, where the model parameters from clients are aggregated and the updated parameters are computed. We propose a federated SVM architecture for solving a binary supervised classification problem. Here the experiments are done for MNIST dataset and COVID-19 dataset. Also the results are compared with centralised training approach.