Comparative analysis of methods for building encrypted data processing systems and their comparison for solving machine learning problems

The article discusses two methods for constructing encrypted data processing systems based on homomorphic encryption and separation calculations. Each method differs from each other in the presented algorithm and data processing model. Basic operations on encrypted data are considered, namely, multiplication and addition. The following libraries were used for the study: tenSEAL, tensorflow and PyTorch. The tenSEAL library is a fork of the full homomorphic encryption tool created for the C++ programming language, but adapted for the Python programming language used in the study. This library allows you to use the method of full homomorphic encryption in constructing a computational model, the task of which will be to process encrypted data. To implement the method of separation calculations, better known as multi-party computation, the tensorflow library will be used, which allows you to create several tensors and train them simultaneously, which in turn makes it possible to implement the principle of separation calculations.