PROJECT TITLE :

FedMarket: A Cryptocurrency Driven Marketplace for Mobile Federated Learning Services

ABSTRACT:

Federated learning (FL) makes it possible to train a shared collaborative machine learning model while maintaining the confidentiality of all training data stored across multiple devices. A monopolist FL task publisher is something that the current state of the art for FL takes into consideration. On the other hand, we offer a FL marketplace where a number of different FL task publishers and mobile devices can coexist in order to complete a variety of distinct and distinct learning tasks. Mobile devices that take part in the training of FL models offer pay-as-you-go (i.e. using cryptocurrencies based on blockchain technology) FL training services to FL task publishers. The proposed framework allows for multiple FL task publishers to compete against one another, and the participating workers (i.e. mobile devices) have the ability to select one FL task publisher over another in order to take part in the training of a global model. We make use of code offloading in order to enable customized FL pipelines in mobile devices and reduce the model heterogeneity that is inherent in the varying and changing FL tasks that are published by task publishers. The effectiveness of the proposed framework was demonstrated by the results of the experiments.