Privacy protection method and system for incomplete data in neural network training

The invention provides a privacy protection method and system for incomplete data in neural network training. Participators comprise a trusted third party TA, a data provider DP and a cloud server CS. It is assumed that there are T data providers (DPs) participating in training, a trusted authority (TA) generates and distributes keys for each DP without participating in any computations. The DP has a special data set, hopes to carry out collaborative neural network learning with other participants on the randomly divided data set, and keeps online by accessing the cloud. Individual privacy data is not disclosed between the DPs to any participant other than the TA. The DP also does not allow the CS to know sensitive information such as an original data set and a training intermediate result. According to the method, the encrypted data set is filled by the cloud server CS, and training is carried out in the neural network until the termination condition is met. According to the method, the privacy of data durin