privacy protection method for deep learning based on local random nodes
Panfeng Zhang, Wenyong Zhang · 2022
Deep learning algorithm needs a lot of data for training to achieve better training results, in practical applications often need to multi-partner data for training. But the centralized training does not meet the standards of privacy protection. Therefore, a cooperative training scheme is proposed. This scheme can achieve privacy protection at a small computational cost. The main idea is to transfer the training process between the first layer and the second layer of the traditional neural network to local, and make the process between them irreversible so as to guarantee the privacy of the training data. The experiment shows that the accuracy of the model can achieve good results.