Federated Learning Framework Based on Data Value Evaluation in Industrial IoT

With the continuous maturity and development of big data technology system, deep learning has been widely used in field Industrial Internet Things. However, traditional training model with centralized is prone to leakage private information industry, such as facial information. In recent years, federated solves problem privacy caused by sharing not only contributing local models. Federated does share data, which also makes it impossible evaluate contribution each client task. We propose a framework based on value evaluation. this method, under premise effectively completing task ensuring evaluator designed central server, uploaded evaluated obtain corresponding selection probability aggregation weight. Experimental results show that proposed method improves accuracy global obtained existing aggregation.