Mixed privacy protection image classification method based on federal model distillation

Abstract

The invention discloses a mixed privacy protection image classification method based on federal model distillation. The method comprises the following steps: 1, pre-training a local neural network by a client; 2, the central server obtains a public data set, randomly samples the public data set and then issues the public data set to the client; 3, the client predicts the issued data; 4, the client performs norm cutting on the prediction result, randomly splits the prediction result into prediction fragments, and sends the prediction fragments to the corresponding client according to numbers; 5, the client aggregates the prediction fragments to obtain confusion prediction and uploads the confusion prediction to the central server; 6, the central server aggregates the confusion predictions and adds noise to obtain global predictions, and then issues the global predictions to the client; 7, training the local neural network by the client through global prediction distillation, and then reviewing and training the local neural network; and 8, the client performs image classification by using the local neural network. According to the invention, based on a federated model distillation algorithm, privacy-protected image classification is realized by using a secret sharing thought and a differential privacy technology.

Description

technical field [0001] The invention relates to the field of anomaly detection, in particular to a hybrid privacy-preserving image classification method based on federated model distillation. Background technique [0002] Image classification is to assign a corresponding class label to each image from a given classification set. It is very easy for the human visual system to identify the category of an image. However, for a computer, it cannot immediately obtain the semantic information of an image like a human and then complete the image classification task. At present, the implementation of image classification is mainly based on deep learning methods. A deep learning model trained on high-quality data can accurately classify images, and today's high-quality data is difficult to integrate due to privacy issues. [0003] The emergence of federated learning (FL) has alleviated the data integration difficulties caused by privacy issues to a certain extent. By uploading the ...

AI Technical Summary

Problems solved by technology

[0004] However, existing image classification methods based on federated model distillation still have the risk of local data privacy leakage

Images