Secure learning and prediction using confidential dataSecure Computation AINTT Social Informatics Laboratories

Secure computation AI provides an environment that enables you to perform AI operations—everything from data registration to preprocessing, building predictive models and predicting new data—without having to decrypt data even once.This will enable AI analysis using data from multiple companies, which is normally difficult to use effectively due to concerns about leakage of knowledge and data, and enable various types of AI analyses using private personal information.

In recent years, digital transformation has come on in leaps and bounds in various fields. One such area is AI analysis of a wide variety of big data, such as customer information held by companies, online shopping history, and sensor data collected by the Internet of Things. This is expected to lead to the development of various industries and services. However, leaks and misuse of corporate confidential data and private personal data is a major concern and has inhibited the collection and use of such data.

• The world's first technology that enables standard deep learning processes, including optimization, using data that remains encrypted
• Data never needs to be decrypted  Enot during communication and storage of training data, the computation stage of learning processing, or decision making with predictive data
• By implementing function approximation algorithms that speed up and maintain accuracy in calculations such as reciprocals and square roots, essential for neural networks, prediction is as fast and accurate as normal computation while also being secure
• This technology is equipped with several neural network models, such as FFNN (feedforward neural network), CNN (convolutional neural network), and RNN (recurrent neural network), as well as various functions such as optimization algorithms and preprocessing
• Data can be processed while encrypted, so more advanced AI using data from multiple companies can be built without the companies viewing each other's data

• We believe it will be possible to create services that use data from within the same industry and brand new services that use data from across different industries.
• As an example from the healthcare sector, the risk of specific diseases will be able to be predicted by using information held by multiple hospitals and institutions on the diagnosis and treatment of rare diseases and data on patients after treatment and medication.
• By learning from a combination of personal vital data, exercise data from the gym, and dietary data from restaurants, we can create a service that provides personalized food and exercise plans.

Secure computation AI is a technology in which secure computation based on secret sharing, which NTT has been researching for many years, is applied to AI machine learning and prediction algorithms. As well as training data, predictive models can also be encrypted, enabling a variety of AI processing to be performed without disclosing the data to third parties such as data analysts and system operators.

NTT Social Informatics Laboratories - Social Information Sharing Research Project