Mingu Kwak

Mingu Kwak

I am Mingu Kwak, a postdoc researcher in Industrial and Systems Engineering at Georgia Institute of Technology, advised by Professor Jing Li. My research focus is on AI models, especially for medical data.

Projects

Research Projects (Georgia Institute of Technology)

  1. [NIH] Image-based Models of Tumor-Immune Dynamics in Glioblastoma
    • 2023.10 ~ Present
    • Developed conditional diffusion & transformer models for translating tumorous brain images into healthy brain images for reducing patient heterogeneity.
    • Trained the model on benchmark datasets and applied it to internal datasets.
    • Utilized generated normal and original images to construct an epidermal growth factor receptor (EGFR) classification model.
    • Developed a brain image preprocessing tool that transforms MRIs, region of interests (ROIs), and biopsy locations into BraTS atlas (Repo).
  2. [NIH] AIDen: An AI-Empowered Detection and Diagnosis System for Jaw Lesions Using CBCT
    • 2022.08 ~ Present
    • Integrated domain knowledge into a deep semantic segmentation model for precise lesion detection in 3D CBCT images.
    • Applied knowledge of lesion occurrence near tooth roots to regularize and guide the segmentation model.
    • Significantly improved the detection and segmentation performance of small lesions.
  3. [NIH] Multi-Modality Image Data Fusion and Machine Learning Approaches for Personalized Diagnostics and Prognostics of MCI due to AD
    • 2022.04 ~ Present
    • Developed a self-supervised contrastive model for classifying MCI converters and non-converters using 3D amyloid-PET images, incorporating label information during the pre-training step.
    • Created a mutual knowledge distillation model for handling incomplete multi-modal 3D image data (MRI and amyloid-PET) in MCI conversion classification.
    • Designed a novel teacher model that enhances knowledge distillation by focusing on modality-common representation.

Research Projects (Korea University)

  1. [Samsung Advanced Institute of Technology] Developing Non-Invasive Lipid Measurement Algorithm Based on 2D Array Sensor
    • 2020.05 ~ 2021.04
    • Developed a method to predict blood lipid concentrations using optical sensor data, eliminating the need for blood draws.
    • Designed a data preprocessing framework to predict lipid levels from sensor data.
    • Implemented a hybrid approach combining autoencoders and machine learning algorithms.
  2. [Samsung Electronics] Congestion-Aware Control of Overhead Hoist Vehicles in Semiconductor Fabrication Logistics
    • 2020.05 ~ 2020.12
    • Developed an adaptive agent to control transportation vehicles in semiconductor FABs, aiming to minimize traffic congestion.
    • Applied imitation learning and data augmentation techniques within a deep reinforcement learning framework.
  3. [Korea Institute of Startup and Entrepreneurship Development] Text Mining and Trend Analysis on Venture Companies and Startups
    • 2020.03 ~ 2020.07
    • Conducted web crawling to collect news articles on venture companies and startups and performed text mining analysis to identify key trends and keywords by year.
    • Used community detection algorithms to group keywords and built a pipeline for hierarchical trend analysis.
    • Performed sentiment analysis to evaluate positive and negative impacts of government policies, and included findings in government agency reports.
  4. [Hanwha ICT] Conversational Platform R\&D: Machine Reading Comprehension with Large Language Models
    • 2019.05 ~ 2019.12
    • Developed text question answering methods for both Korean news articles and in-house regulation documents to ensure compliance.
    • Trained large language models (e.g., BERT) and distilled them into smaller models for deployment (e.g., DistilBERT).
  5. [Hyundai Motors and DS-eTrade] Durability Monitoring System for Road Simulator
    • 2019.04 ~ 2019.12
    • Developed an algorithm to detect abnormal states and problematic parts of vehicles during road simulator operations.
    • Implemented a hierarchical feedforward attention network to detect abnormal states and explain the causes.
  6. [Hyundai Motors and DS-eTrade] Detecting and Categorizing Failure Patterns of EGR Valve
    • 2019.04 ~ 2019.12
    • Predicted failures in EGR valves of diesel cars and analyzed sensors causing these failures.
    • Implemented a hierarchical feedforward attention network to detect failures and identify critical sensors and time steps.
    • Utilized sensor-level attention scores to cluster failure patterns.
  7. [Samsung Electronics] Classification of Signal Patterns for Abnormal Cause Analysis of Semiconductor Logistics Systems
    • 2019.03 ~ 2019.11
    • Developed a framework consisting of anomaly detection, anomaly pattern clustering, and classification of logistics indices.
    • Discovered meaningful anomaly patterns by clustering channelwise reconstruction errors.
    • Employed an open-set model capable of classifying known classes and detecting unseen classes not present in training data.
  8. [Hyundai Heavy Industries and Youngshine D&C] AI-Based Smart Construction to Reduce Costs by 20%
    • 2016.09 ~ 2020.12
    • Conducted a study to predict construction equipment failures by analyzing sensor data.
    • Employed incremental PCA to adapt to changing data distributions over time, creating a lightweight model that can be easily embedded in equipment control systems.
  9. [Samsung Electronics] Deep Learning-Based Reliability Diagnosis Process Improvement
    • 2018.05 ~ 2019.04
    • Conducted a study on machine learning methods for early diagnosis and prediction of wafer quality in sub-10nm logic technology.
    • Applied domain knowledge-based and machine learning-based methods for missing data imputation.
  10. [Electronics and Telecommunications Research Institute] ICT-Based Crime Risk Prediction and Response Platform Development for Early Awareness of Risk Situations
    • 2018.04 ~ 2018.12
    • Developed a method combining criminal data with various public data to predict near-future crime reports.
    • Considered a combination of recurrent neural networks and ensemble algorithms as the prediction model.
    • Developed a software tool for visualizing crime report statuses.
  11. [Hyosung Heavy Industries] Advancing Health Index Module for 154kV Substation Facilities
    • 2018.09 ~ 2018.12
    • Conducted a study to improve the accuracy of a pre-developed health index module.
    • Applied the advanced module to various types of facilities to demonstrate its generality.
  12. [Hyosung Heavy Industries] Optimal Decision-Making System for Maintenance and Health Index Module for 154kV Substation Facilities
    • 2017.12 ~ 2018.03
    • Developed a system for optimal maintenance decisions for multiple facilities within a limited budget.
    • Applied dynamic programming and integer programming to optimize the decision-making process.
    • Designed a decision-making pipeline with gradient boosting machines to predict health indices and calculate feature importance for inspection items.
    • Implemented the pipeline into a monitoring system for real customers.
  13. [Hyundai Heavy Industries] Forecasting Demand for Construction Equipment Parts Using Big Data Analysis
    • 2016.10 ~ 2017.07
    • Predicted monthly demand for construction equipment parts and designed an objective dealer evaluation indicator by comparing predicted values to actual sales.
    • Developed software for visual comparisons of actual sales, predicted demands, and evaluation indicators by region, dealer, and part.