Examination of deep learning methods in encrypted data

Today, one of the most important problems in data storage and processing processes is to ensure the security of data. While various encryption algorithms are used to ensure data security, accuracy and loss values and performance in calculations on encrypted data also appear as separate problems. In this research, the steps of analyzing image data encrypted using deep neural networks are presented practically. For this, first of all, raw data was taken and encrypted, and by creating a model, the data was trained with this model and calculations and analyzes were made on this data. Properties such as data type, size, encryption key were used as parameters in the calculations and analyzes. In addition, VGG16, VGG19, ResNet50 deep learning models were applied to the cifar10 data set, a model was created by applying the DES encryption algorithm for MNIST image data and the probability of predicting encrypted data according to the unencrypted data was calculated. During this period, the performance of the network was tried to be measured and the accuracy and loss values were shown graphically in these calculations.