Ongoing Projects 

Sep. 2024 - Evaluating National Science and Technology Capabilities through Individual Researcher Modeling (국가미래전략원 연구과제)

This project aims to evaluate national capabilities in S&T (science and technology) by modeling the individual capabilities of researchers. By analyzing publication records and collaboration networks, we identify each researcher’s potential and aggregates it to assess national strengths. The findings will provide insights to improve S&T policies, with a focus on strengthening human resource development.

Sep. 2024 - Research on Federated Learning between Financial Institutions (금융 분야 연합학습 활용 연구)

This project investigates how federated learning can enable secure collaboration between financial institutions, especially focusing on fraud detection without exposing sensitive data. The purpose of this project is to build a foundation for enhancing collaboration in the financial sector, where inter-institutional collaboration is challenging due to strict security and data privacy requirements.

Aug. 2024 - Development and Demonstration of Federated Learning System for Predictive Modeling from Decentralized Clinical Data (분산 의료 데이터 대상 예측 모델 개발을 위한 연합학습 프레임워크 구축 실증 과제)

This project aims to design a framework for demonstrating federated learning algorithms on distributed clinical data. We develop a holistic system from a communication topology between physically isolated computing servers at different regions in South Korea to local & global computational schemes for collaborative training of a predictive clinical AI model.

Mar. 2024 - Understanding COVID-19 via deep learning-based single-cell sequencing data analysis  (딥러닝 기반의 단일세포 시퀀싱 데이터 분석을 통한 COVID-19 이해)

This project aims to develop a multi-model embedding fusion model consisting of graph neural networks and transformers for analyzing and understanding multi-omics single-cell data related to COVID-19. The model enable to extract rich single-cell representations that encapsulate comprehensive information about individual cells. These representations will be extended and applied to various tasks, such as understanding the biological characteristics of cells and predicting disease severity.

Sep. 2023 - Development of Homomorphic Encryption-friendly Deep Learning Model for Time Series Classification (IITP 연구과제)

This project aims to develop a HE (homomorphic encryption)-friendly deep learning model for time series classification. For HE implementation, a deep learning model should be ‘depth-efficient’ due to multiplicative depth constraints. We try to address this issue through various methods such as time-series imaging and nonlinear operation approximation.

Mar. 2022 – Machine Learning-based Personalized Postprandial Glycemic Response Prediction using Gut Microbiome and Continuous Glucose Monitoring (맞춤형 식후 혈당 반응 예측모델 개발 과제)

Developing a postprandial continuous blood glucose prediction model by integrating heterogeneous healthcare data such as clinical information, dietary records, gut microbiota data, and continuous blood glucose. Anticipating personalized blood glucose monitoring and tailored dietary management for individuals through this effort.

Apr. 2021 –  Smart and Advanced Environmental Public Health Surveillance System (환경부 연구과제)

It aims to develop a GIS-based environmental public health surveillance system using big data and machine learning. To be specific, we develop an environmental index, which can reflect environmental conditions and risks, by using environmental factors.

Projects Completed

Mar. 2019 – Feb. 2024 Development of Deep Learning based Privacy-Preserving Federated Learning Platform for Artificial Intelligence (신진연구과제)

Developing deep learning based privacy-preserving federated learning system for artificial intelligence. As privacy has become an important issue, federated learning (FL) is being noticeable as a remarkable solution of machine learning (ML) in recent years. FL is an emerging configuration of ML techniques, where collaborative learning of a global model between individuals or institutions is possible with no migration of data under the administration of a central server.

Jan. 2021 – Feb. 2023 Development of Intelligent Bio Multi-omics Analysis (게놈 특구 프로젝트)

Intelligent multi-omics analysis means an integrated analysis of biomedical data such as the genome, transcriptome, and epigenome with novel machine learning (ML) methodologies. Based on multi-omics data, it seeks to uncover disease mechanisms and develop more accurate ML models for disease prediction. It is anticipated to realize genome-based personalized and precision medicine and the expansion of the bio-health industry through the prediction, diagnosis, and treatment of hereditary diseases.

Feb. 2019 – Dec. 2023 Voice Activity Detection Model Development Project

Creating a Voice Activity Detection Algorithm aims to implement an algorithm that can analyze participation patterns in discussions and the interactions between participants by classifying whether or not they are speaking when several speakers are discussing. In particular, this research field plays the most fundamental building block of quantitative analysis when studying interactions between people in organizational behavior theory in business administration.

Jul. 2019 –  Dec. 2022 Prediction of Diabetes Causes Based on Genetic and Epidemiological Survey Data

This study developed a machine learning model to predict type 2 diabetes caused by multiple causes such as genetic and environmental factors. We constructed a model that can predict type 2 diabetes incidence with high performance by reflecting information on family history, lifestyle, genetic factors, and metabolites that can cause diabetes. In particular, to reflect genetic factors, a new indicator called genome-wide polygenic risk score (gPRS) was developed, thereby significantly improving type 2 diabetes predictability.

Jul. 2019 –  Dec. 2022 Leukemia Drug Prediction Project

It aims to develop a machine learning model for recommending optimal drugs to leukemia patients. The project is joint work with the Seoul St. Mary's Hospital.

Feb. 2020 – Dec. 2022 Households' Financial Status Analysis (가계 금융 상태 분석)

It aims to develop an overall financial health score based on the data-driven approach. By suggesting the household finance score, we can expect that the individuals can monitor their positions and manage to prevent or overcome the financial risk.

Oct. 2021 –  Oct. 2022 Multi-modal Data-based Outfit and Style Recommendation (무신사 산학과제) 

It aims to generate compatible outfits and recommend those styles. An outfit is defined as a set of different types of goods. The model extracts multi-modal embedding via image and text data of goods and learns a latent space such that embeddings of goods belonging to the same outfit are close.

Jan. 2021 – June 2022 Development of Big Data Processing Techniques and Diagnostics for Commercial Air Conditioners (LG전자 산학과제)

Development of anomaly detection and failure cause diagnosis algorithm using air conditioner sensor data.

Jan. 2021 – Dec. 2022 Smart Insole Advancement Project Based on Gait Data Analysis (중기부 연구과제)

It aims to improve smart insoles' performance by developing an artificial intelligence model based on smart insoles' gait data. Starting with the development of a weight prediction model through sensor data of smart insoles, a model applied to various fields such as 'Disease Prediction' and 'Fitness Training Monitoring' will be developed.

Jul. 2020 – Jan. 2021 Manufacturing Data Analysis and AI Model Development (KPX 케미컬 과제)

The purpose of this project is to predict defects and determine the cause of faults using the chemical mechanical polishing pad (CMP pad) manufacturing process data for semiconductors produced by KPX Chemical.

Mar. 2018 – Feb. 2019 Sensor-Based Gas Detection Algorithm Development Project (산학융합원 과제)

It aims to develop a machine learning model for identifying the type of gas in mixed gas. It is possible to grasp the type of gas without human intervention through this model, which can learn from the data obtained through multiple sensors.

Apr. 2020 – Dec. 2022 Retail Recommendation System Development Project (중기부 연구과제)

This study developed a sequential recommendation system for retail. The sequential recommendation extracts user preference from a series of users' actions. However, the purchasing sequence is hard to collect in the offline market. Therefore, the sequential data is determined by a shopping list and shopping routes and then is used for item recommendation.

May 2019 – Development of a Prediction Model for Crude Oil Import Volume (울산항만공사 연구과제)

Developing an accurate prediction model for crude oil import volume for Ulsan Port Authority (UPA). The model will assist UPA in terms of facility expansion plans of crude oil processing units and decision support.

Apr. 2019 – Mar. 2022 SK Lubricants Process Optimization (SK 루브리컨츠 산학과제)

Developing optimized prediction models for lubricant refinery process for SK Energy. In the process of refining crude oils to make Lubricants, many sub-processes are manually controlled. We try to improve such unstable processes using machine learning models and prediction model tools.

May 2019 – Dec. 2019 Development of New Biological-Age Model (바이오에이지 산학과제)

A health index is a useful tool for people as it can work as a blueprint for managing their own health. To estimate one of the beneficial health indices, the Biological Age (BA), we developed a new BA algorithm based on representation learning to obtain better accuracy than existing methods.

Sep. 2018 – Aug. 2020 Development of Privacy-preserving and Secure Machine Learning-based Federated Prediction Models for Distributed Data (생애첫 연구과제)

Developing a novel strategy for handling sensitive data in a distributed setting by combining privacy methods and security techniques to achieve the best of both worlds.

Dec. 2018 – Aug. 2019 Development of Prediction Model for Thick-Plate Roughing Mill in POSCO (포스코 산학과제)

Improving current prediction model of roughing mill in POSCO. Steel plates are manufactured by hot rolling, which requires an adequate amount of torque applied on the slab. We applied machine learning models to improve the previous prediction model based on physical equations of roll-torque.