Federated machine learning method and device based on security computing, device and medium

Abstract

The invention discloses a federated machine learning method and device based on security computing, a device and a medium. The method comprises the steps that S1, a participant A initiates a joint modeling task to a participant B; S2, the participants A and B start the local sample data training according to task parameters and calculate local gradient values; S3, the participant A and the participant B calculate a synthesized gradient value by using a multi-party security protocol gradient synthesis function according to the local gradient value; S4, the participants A and B update the local model parameters respectively according to the synthesized gradient value; S5, the steps S2 to S4 are circularly executed according to the number of iterations of the task parameters, and a final model is obtained when a set condition is met; S6, the participant A sends a task ending instruction to the participant B; and S7, the participant B receives the information, finishes the task and destroys the local model information. According to the method, a network channel is constructed based on a point-to-point mode, the summary gradient is realized by adopting spdz for calculation, and data leakage is prevented.

Description

technical field [0001] The present application relates to the technical field of machine learning, in particular, to a secure computing-based federated machine learning method, device, device and medium. Background technique [0002] At present, the realization of secure machine learning is mainly based on the machine learning of trusted third parties, such as figure 1 As shown, the third-party organization is responsible for generating the public-private key pair, distributing the public key, encrypting, decrypting, and calculating. A and B; 2. Participants A and B calculate the gradient and loss respectively, and use the public key PK to encrypt the result and send it to the nemesis third party Z; 3. The trusted third party Z uses the private key SK to decrypt and summarize the calculated gradient, and send it to Participants A and B, update the local model. Due to the introduction of a third-party organization, the third-party organization can obtain additional calculat...

AI Technical Summary

Problems solved by technology

[0003] On the one hand, this application provides a federated machine learning method based on secure computing to solve the technical problems of data privacy leakage and model exposure risks in existing secure machine learning due to the introduction of third-party organizations

Images