Fair Rewards in Federated Learning: A Novel Approach with Adjusted OR-TMC Shapley Value Approximation Algorithm

Abstract

Federated Learning (FL), a new private and secure Machine Learning (ML) approach, faces a big difficulty when it comes to sharing profits with data producers. Shapley Values (SV) have been proposed as a fair incentive system to remedy this, but it is challenging to determine the SV with accuracy. Therefore, SV calculation is problematic since the number of necessary federated models rises exponentially with the number of data sources. As a result, an effective approximation approach is required. The One Round Model Reconstruction (OR) and Truncated Monte Carlo Shapley (TMC) approaches for SV approximation in FL are being improved and combined in this study. The proposed approach, Adjusted OR-TMC, combines TMC principles with OR and achieves a comparable level of accuracy over a shorter period. Because of this, Adjusted OR-TMC is the perfect OR replacement. The performance outcomes and underlying causes are covered in the study.