Secret calculation is a method of depositing data in another server with the data encrypted, analyzing the data, and having only the calculation result returned.Homomorphic encryption is a general term for cryptographic systems that enable addition and multiplication while the input information is encrypted, and the method that can be realized for the first time was proposed in 2009.However, a method for performing various statistical analyzes while guaranteeing practical calculation time and calculation accuracy for large-scale data has not been realized so far.

　The research group focused on the fact that many statistical calculations are described by matrix operations and magnitude comparison operations.We have developed an algorithm for efficient matrix operations and magnitude comparison operations using fully homomorphic encryption, and an operation method for realizing high-precision numerical operations required for statistical analysis on ciphertext.By combining these methods, evaluation of standard descriptive statistics, predictive statistics, statistical tests, etc. can be performed in a few seconds to 10 minutes for encrypted data of tens of thousands of records including numerical attributes, order attributes, and discrete attributes. I succeeded in realizing it.

　Until now, it has been difficult to integrate individual medical information, genetic information, behavior history, purchase history, etc. held by multiple organizations between organizations in order to protect privacy.By utilizing the technology developed this time, it can be expected to realize statistical analysis and machine learning that transcends organizational barriers without compromising privacy at all.