Aggregation Techniques in Federated Learning: Comprehensive Survey, Challenges and Opportunities
Mukund Prasad Sah, Amritpal Singh · 2022 2nd International Conference on Advance Computing and Innovative Technologies in Engineering (ICACITE) · 2022
Federated learning is a new paradigm on the machine learning system that uses the traditional system of machine learning but implements privacy features on top of it. The implementation of federated learning is done in order to increase the privacy of the user and also give them access to their own rich private personalized data. But in the case of classical machine learning the implementation is that it needs all the data and model on one system and in that way we have to get the data of the user and store it in a certain place and then train the model there and get the updated model by training on that data. Let us look at the problems in this system first: the personalized data of the user is being sent to the servers and is used there. Even if the owners of the server are not exploiting the data of users it can be possible by any cyber attacks people can steal the data and use it for their own benefit and in that there is no one to blame for. In order to avoid these data breaches, there can be a solution where we instead to taking the data to the model send the model to the data, that is the core concept of the Federated Learning paradigm - in this way we can preserve the privacy of the user and also train our model on the rich personalized data set.