Federal learning method based on dynamic adjustment model aggregation weight

Abstract

The invention discloses a federal learning method based on a dynamic adjustment model aggregation weight, and the method comprises the steps that: a cloud server receives a local training model and a data quality index from each client, and if a client weight updating condition is reached, a contribution score for each client is calculated according to the contribution of the data quality, the model precision and the model difference index to the model training precision, and weighted average is performed to generate a global model; and the cloud server issues the updated global model to the clients, each client carries out model training on the local training data after receiving the global model, and uploads the local model and the data quality index to the cloud server again after training is finished. According to the method, client training data such as data distribution, model precision and model difference are reasonably utilized, the dynamic aggregation weight is generated, available features in the client training process are fully mined, and a global model with higher quality is formed, so that the model precision and convergence efficiency are improved.

Description

technical field [0001] The invention relates to the field of distributed machine learning, in particular to a federated learning method based on dynamically adjusting model aggregation weights. Background technique [0002] As digital technology enters a period of rapid development, technologies such as big data and artificial intelligence usher in explosive development. On the one hand, this brings new opportunities for upgrading and transformation of traditional business formats; on the other hand, it inevitably brings new challenges to data and network security. a whole new challenge. In order to ensure data security and privacy, various companies are unwilling to share data, which makes the data of each enterprise can only be processed locally, resulting in the problem of data islands, and data islands prevent companies and researchers from repeatedly analyzing and mining data characteristics, hindering The development and application of big data and artificial intellig...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that existing federated learning methods cannot provide heterogeneous data solutions with good versatility, the present invention starts with model training information, mines client data and model features without revealing user privacy, and integrates model accuracy, data Quantitative representation of training information in three dimensions of quality and model difference, and three quantitative indicators are used to jointly model, and on this basis, the aggregation weight is dynamically set for the client to improve the accuracy and convergence speed of the model in non-IID data scenarios

Images