Joint training method and device for business models

Abstract

The embodiment of this specification provides a joint training method and device for a business model. In the process of multi-party joint training of the business model, in order to use the Newton method to train the business model with a faster convergence speed, when determining the inverse matrix of the Hessian matrix, introduce For the auxiliary matrix generated by a trusted third party, each business party of the joint training business model will determine the product of the Hessian matrix and the auxiliary matrix in a secret sharing manner, and each get a corresponding share. Afterwards, the combined result of the product of the Hessian matrix and the auxiliary matrix is ​​disclosed, so that each business party can obtain the inverse matrix of the above combined result, and then determine the corresponding share of the inverse matrix of the Hessian matrix. According to this specific business-party interaction setting, it is feasible to use Newton’s method to adjust model parameters in multi-party secure computing under the premise of ensuring data privacy, thereby improving the joint training efficiency of business models.

Description

Technical field

[0001] One or more embodiments of the present specification, embodiments relate to computer technology, and in particular relates to a multi-party secure computing, the method of using the computer to the business model and joint training device. Background technique

[0002] Big Data era, a lot of data may be stored in dispersed in different businesses, based on considerations of privacy protection, and can not fully share their data business to business, and therefore there may be islands of data.

[0003] In some cases, enterprises and between enterprises often require cooperative security model, in order to fully protect the premise of enterprise data privacy, data processing model for collaboration trained to use the data of the parties, so as to more accurately and efficiently handle business data. In this case, for data processing model training data can be dispersed parties in cooperation modeled. So how in the training process model based on the protectio...

Examples

Embodiment Construction

[0040] The schemes provided herein will be described below with reference to the drawings. The following describes the technical terms embodiment according to the present embodiment description.

[0041] Multi-party Secure Computing (Secure Muti-Party Computation, MPC) is a secure algorithms to protect data privacy. Secure Computing enables multi-party participation in multiple data calculated collaborate calculated without leakage own data.

[0042] Secret Sharing (Secret Sharing, SS) is a secure algorithms to protect data privacy. A plurality of data side may be made without leaking own data, using the secret sharing method for collaborative computing, to obtain secret information. Each data share may obtain a copy of the confidential information. Single party can not recover the data secret information. Only the parties have a number of data integration in order to restore the share secret information. For example, the data held by the data x square P1 1 The data side P2 holds ...

Technical field

[0001] One or more embodiments of the present specification, embodiments relate to computer technology, and in particular relates to a multi-party secure computing, the method of using the computer to the business model and joint training device. Background technique

[0002] Big Data era, a lot of data may be stored in dispersed in different businesses, based on considerations of privacy protection, and can not fully share their data business to business, and therefore there may be islands of data.

[0003] In some cases, enterprises and between enterprises often require cooperative security model, in order to fully protect the premise of enterprise data privacy, data processing model for collaboration trained to use the data of the parties, so as to more accurately and efficiently handle business data. In this case, for data processing model training data can be dispersed parties in cooperation modeled. So how in the training process model based on the protectio...

Examples

Embodiment Construction

[0040] The schemes provided herein will be described below with reference to the drawings. The following describes the technical terms embodiment according to the present embodiment description.

[0041] Multi-party Secure Computing (Secure Muti-Party Computation, MPC) is a secure algorithms to protect data privacy. Secure Computing enables multi-party participation in multiple data calculated collaborate calculated without leakage own data.

[0042] Secret Sharing (Secret Sharing, SS) is a secure algorithms to protect data privacy. A plurality of data side may be made without leaking own data, using the secret sharing method for collaborative computing, to obtain secret information. Each data share may obtain a copy of the confidential information. Single party can not recover the data secret information. Only the parties have a number of data integration in order to restore the share secret information. For example, the data held by the data x square P1 1 The data side P2 holds ...

Images