[3L1-GS-11] AI and Society

[3L1-GS-11-01] MEVAL AI:Secure Computation AI Library for privacy-preserving Machine Learning

〇Ibuki Mishina¹, Koki Hamada¹, Dai Ikarashi¹, Ryo Kikuchi¹ (1. NTT Social Informatics Laboratories)

Keywords:Secure computation, Machine Learning

There is a technology called "Secure Computation AI" that performs machine learning training and prediction while keeping data encrypted. Using this technology, machine learning can be performed while keeping important data secure. However, Secure Computation AI is still in the category of cryptographic applications and is not widely used as a method of data analysis. One of the reasons for this is that secret computation AI does not have a general-purpose language interface that is easy to use for ordinary data analysts. In this paper, we have implemented a python library called MEVAL AI, which enables the use of secret computation AI with a language interface similar to common AI libraries such as scikit-learn and keras. MEVAL AI allows users to easily use the main methods of machine learning, such as regression, classification, clustering, and dimensionality compression, as a Python library while keeping the data encrypted.