A Review on Federated Learning approach in Artificial Intelligence

In today's era, there is exponential increase in generation of data due to high speed internet availability, use of various electronics gadgets and IOT Use of that data for decision making has been increased proportionally. Typically, Machine Learning (ML) process has been done by collecting the data centrally on the cloud or server from each connected client nodes for training a model of machine learning which will be referred as global model. Then it is distributed and applied to all devices for generating expected throughput. The issue is that, user data is personal and confidential. As there is increase in volume, velocity and veracity of data that directly increases the concerns of data privacy. Collection of accurate data is the major concern for the classical centralized machine learning process of training and building ML prototypes with high efficiency, because in this method client must have to provide their private information to server. So here we are proposing an approach of federated learning which will be the interchange of local learning update parameters with the global learning model. Federated Learning is based on machine learning in which it trains an algorithm over multiple decentralized devices using local data, without transferring the data from client to server. The main proposition behind federated learning implementation is to bring the centralized model parameter to the connected decentralized devices and so it abolish the need of uploading user data on server. As users personal data detain on the client node itself, this helps protect data from its misuse and ensure security, as only the prototype result parameter are shared with the global model. This machine learning privacy enhancement is groundbreaking and opens up new ways for machine learning systems to handle sensitive data.