Publication Details

Abstract

Federated Learning (FL) is a recently proposed distributed machine learning methodology that protects data privacy by allowing clients to train models without disclosing raw data. The global model may be impeded by corrupted updates transmitted by malicious clients, which is a consequence of the decentralized character of federated learning. FEDGT is a unique architecture in federated learning that detects fraudulent clients and provides secure aggregation privacy. The proposed methodology utilizes trust rating instruments and gradient-based analysis to identify anomalous client behavior during training, all while ensuring the confidentiality of model modifications. FEDGT employs sophisticated detection methods and secure aggregation techniques to protect data privacy and prevent fraudulent donations. According to the results of the testing, the proposed strategy improves the robustness, reliability, and attack resistance of federated learning models, while simultaneously maintaining the accuracy and communication efficiency of the models.