Recent Challenges in a New Distributed Learning Paradigm

Abstract

The majority of online transactions between the parties can be reported in real-time publicly by relying on smart contracts (SCs) and federated learning (FL). They are both decentralized and reinforced. By aggregating the gradient values from client devices, FL enables a large number of clients to create deep learning models anonymously. Yet, it lacks an incentive system for the contributing clients. Conversely, since self-executing contracts with immutable data records are resilient to failure, the virtues of SCs can be a tenable solution as an incentive mechanism in the FL system. The clients can claim the rewards by providing a proof transaction function and stating their contribution arbitrarily through SCs. However, because the transactions are made public, directly implementing SCs in the CL system could jeopardize the users’ privacy. The observer can deduce the characteristics of the client’s resources. Therefore, in this research, we elaborate on the critical points to be taken into account in adopting SCs as an incentive mechanism for the FL environment. We also state an empirical investigation and the open challenges that can address the aforementioned issues and concerns. Eventually, we recapitulate the essential points to be considered in developing a new distributed learning paradigm.