- Applies a federated learning method between individuals without sharing their own data
- Demonstrates AI performance through personalized federated learning by sharing AI between individuals and enterprises
Harex InfoTech unveiled the results of its research and development project on "Blood Pressure Management System" using User Centric Artificial Intelligence (UCAI) with personalized federated learning method at the session of the Korea Management Information Society of the "25th Korea Management Association Convergence Conference" at BEXCO in Busan, South Korea in recent. Kyung Yang PARK, CEO of Harex InfoTech presented the results of the joint research with Harex InfoTech's UCAI Research Center and Professor Kyoung Jun LEE's team at Kyung Hee University.
The study focused on the "Digital Me" service, which is an AI-based product-service system designed to manage and improve personal aspects such as health, finance, psychology, and knowledge in real time. The research team developed a novel Aim-Measure-Predict-Evaluate-Recommend (AMPER) framework based on the Digital Me concept, which replaces the traditional approach.
The AMPER sets M (Measure) and evaluates the current state of the user based on the user-centric A (Aim). P (Predict) predicts the user's future state exclusively based solely on data. E (Evaluate) evaluates the predicted state, followed by R (Recommend) to suggest appropriate actions that optimize the user's progress toward the target state, thereby maximizing state improvement.
Despite the potential benefits of data aggregation, the research team believes that combining or sharing user data is inadvisable due to the sensitive nature of healthcare services within the Digital Me domain. Therefore, they developed a healthcare-specific blood pressure management system using personalized federated learning. This innovative method enables AI services without compromising the security of user data.
Harex InfoTech extends UCAI services to enterprises and small businesses across multiple industries. This enables them to quickly deliver highly customized services and marketing to consumers without the need for in-house AI development. By facilitating personalized services and marketing through AI federated learning, data remains decentralized, ensuring both security and data sovereignty. This collaborative approach fosters synergy between companies, enabling small businesses to harness the power of AI services.
Personalized federated learning is a technique that trains a single AI model using only the model's training values, without centralizing data. This method is particularly effective when data is distributed and belongs to different people.
The research team created a model to predict the next blood pressure reading from the blood pressure data of 9,908 people, and applied the personalized federated learning method, achieving a 31.36% improvement over the average of 9,908 individual models. This is the first case of applying personalized federated learning to blood pressure prediction, which is expected to be an important milestone in leveraging data while protecting personal medical information.
"Healthcare is essential to prevent the tragedy of a long and unhealthy life," said Soo Hyun Kim researcher at Harex InfoTech. She added, "We applied the Digital Me service we have been researching to the healthcare (blood pressure) domain, and since health-related data is sensitive information, we will apply technologies such as federated learning to build a UCAI healthcare system that everyone can easily use without personal information issues."
In addition, the research team achieved not only federated learning between individuals but also federated learning between enterprises. Four individual recommendation models were created by dividing two different data in the field of commerce, and the whole model was learned and compared with the federated learning method, and the whole model applied with the federated learning method showed an improved accuracy of up to 33.6% of the whole individual models. Individuals and organizations alike have demonstrated the power of federated learning. The results of this study are seen as an important achievement proving that even individuals and companies that do not want to directly share data can build models with improved performance through federated learning.
Harex InfoTech is a first-generation fintech company established in 2000 that announced the world's first mobile card payment service in Silicon Valley (San Jose, CA) that same year. It won the Gold Medal in Best Technology at "Innovation Projects 2013" held by PYMNTS.com at Harvard University. Harex InfoTech shares its platform with all businesses to build a User Centric hyper-connected network capable of delivering hyper-personalized services driven by User Centric Artificial Intelligence (UCAI). These hyper-personalized services are heralding a new shared economic order, optimizing benefits to users and businesses.
