Edge end bias detection method in federated machine learning environment

Abstract

The invention discloses an edge end bias detection method in a federated machine learning environment. The method comprises the following steps of: screening to obtain ambiguous instances, increasingthe proportion of the ambiguous instances in a data set to construct a new data set, carrying out distributed training on a model by utilizing the new data set, obtaining the prejudice degree of eachmodel for sensitive attributes (prejudice information), removing prejudice by endowing each model with different attention weights according to the prejudice degree, and performing model aggregation after the prejudice is eliminated, thereby ensuring the fairness of an edge end in a federated machine learning environment.

Description

technical field [0001] The present invention relates to the field of machine learning, in particular to a method for edge bias detection in a federated machine learning environment. Background technique [0002] With the rise of IoT technology and edge computing, data is often not constrained to a single whole, but is distributed in many ways and exists in the form of isolated islands. The traditional data processing workflow is generally that one party collects data, then transmits it to the other party for data preprocessing and modeling, and finally generates a model and distributes it to a third party. In recent years, with the application of artificial intelligence in various industries, people's attention to user privacy and data security is gradually increasing. Users are beginning to pay more attention to whether their personal data is collected with personal permission. At the same time, with the gradual improvement and implementation of various laws and regulatio...

AI Technical Summary

Problems solved by technology

Existing debiasing methods are mainly aimed at a single training model. This type of machine learning model usually does not perform global parameter updates. Compared with the federated machine learning model, under the same parameters, the accuracy performance of a single training model is The performance is worse than that of federated learning; under different parameters, even if the model becomes worse, the accuracy loss of a single training model is significantly higher than that of the federated learning model

Images