The embodiment of the invention provides a joint training method and device for a
service model, and the method and device enable the nonlinear complex operation to be distributed to a
third party forprocessing in multi-party safety calculation, and greatly reduce the complexity of the joint training of a plurality of service parties for a
nonlinear model. Meanwhile, each service party holding the
feature data or the
label data of the training sample generates a preset number of random numbers in an agreed
random number generation mode, the sample data of the training samples in one-to-one correspondence in sequence is obtained, the consistency of the data is kept, and since the
third party does not participate in the
random number generation process, other service parties effectively guarantee data privacy for the
third party. And other service parties interact with each other through a
secret sharing method, so that data privacy is ensured among the service parties. In conclusion, according to the method, on the basis of
privacy protection, the operation complexity of a
nonlinear model jointly trained by a plurality of service parties is greatly reduced.