A Review of Privacy-Preserving Decentralized Learning

In modern AI, one significant obstacle still exists that is the improvement in the security of data and privacy. Federated learning (FL) refers to a configuration in which several clients work together to solve machine learning issues while being managed by a central aggregator. Additionally, this configuration permits decentralization of training data, ensuring each device's data privacy. The original data stored by the client is preserved locally and cannot be switched or relocated. The learning objective is achieved through the use of FL, in which local training is provided on local data by devices and uploads the model to the server for aggregation, and the server then sends the participants the most recent model. Local computation and model transmission are the two main tenets of FL, which lowers some of the systematic privacy issues and costs associated with centralized machine learning techniques. In an endeavor to delve deeper into the intricacies of FL, this review paper comprehensively explores various aspects of the subject matter. It encompasses an elucidation of FL, its categorization into different types, a scrutiny of the associated challenges, an evaluation of the diverse platforms facilitating federated learning, and a review of its multifaceted applications across various domains.