Federated Learning: A Survey of a New Approach to Machine Learning

Massive clients can use large-scale machine learning using federated learning without revealing their raw data to the outside world. It's capable of preserving client personal information while also achieving great learning performance for the client's benefit. Existing research on federated learning is primarily concerned with increasing learning efficiency and model accuracy. But in reality, customers are unwilling to take part in the learning process unless they are compensated for their time and effort consequently, it is critical to figure out how to get customers involved in federated learning by motivating them successfully. Other areas like crowdsourcing, cloud computing, smart grid, etc. are simpler than designing an incentive structure for federated learning. To begin, it's impossible to determine the exact worth of the training data collected from each individual client. Second, different federated learning algorithms' learning performance is challenging to model. This work examines the design of a federated learning incentive system. Before we evaluate and contrast different strategies, we provide taxonomy of existing federated learning incentive mechanisms. There have also been some innovative ideas for enticing customers to take part in federated learning.