👋🏼Jinsu KIM

🤳🏼Phone: 010-8994-5376

🏠 Location: 서울특별시 관악구 봉천동

📚 Major:

• Bachelor’s degree
  • Nuclear Engineering
  • Dep. of Physics and Astronomy
• Master’s degree
  • Nuclear Engineering: Fusion Plasma

✉️e-mail: [email protected]

🏫 School: Seoul National University, Republic of Korea

📝 GPA:

• Bachelor’s degree
  • 3.98 / 4.3, summa cum laude (rank : 1/30)
• Master’s degree
  • 4.15 / 4.3

Introduction

✨ About me

<aside> 🎤 I’m a graduate school student in the Department of nuclear engineering at Seoul National University. I’m currently interested in AI applications for fusion plasma. Currently, I have researched disruption prediction using deep learning and tokamak plasma operation control using reinforcement learning in KSTAR. The results can be seen in my github repository.

Before entering graduate school, I worked for a startup company as a front-end developer and machine learning scientist in the semiconductor AI area. Recently, I’m proceeding with a side project with my team members, for implementing a web service that helps detect molecular images in pharmaceutical research patent documents and convert them to the molecular structure with SMILES format.

Now, I’m focused on research topics as seen below.

• Physics-informed Neural Network: Fast and efficient Grad-shafranov solver
• Disruption physics and prediction based on the Bayesian approach
• Multi-agent RL for multi-objective tokamak plasma control in KSTAR

Here are the links to my workbench and job career information. If you have any questions, please contact Gmail ([email protected]) or SNU-email([email protected]).

1. Code repository : github-zinzinbin
2. Job career info: linkedin-zinizinbin
3. Research career info : Research porfolio

</aside>

Research

🔍 Research topics

<aside> 📓 Recently, I focus on AI application in fusion plasma area. The research topics that i conducted in a graduate school are listed below.

• Disruption physics
  • Disruption prediction using multimodal learning
  • Enhancement of the disruption prediction based on Bayesian approach
• Tokamak plasma operation control
  • Application on reinforcement learning
    • 0D parameter control
    • Shape parameter control
    • Multi-objective reinforcement learning : 0D parameter + Shape parameter control </aside>

<aside> 👉🏼 These toggle lists below contain some details of my research including experimental results and analysis.

Disruption prediction and its analysis using multimodal learning in KSTAR

Tokamak plasma autonomous control based on Reinforcement Learning

</aside>

✏️ Topics for studying

<aside> 🔥 Additionally, i’m interested in computational physics combined with neural networks and Bayesian application. Since Bayesian approach can provide explainability and robustness compared with frequentist approach, I’m trying to use Bayesian based neural network models for accurate and reliable disruption prediction. The combination of the computational physics and neural networks, which is so called as ‘Physics-informed neural network’, is a minor area for now, but the benefits for using PINN are attractive to me. Thus, i am recently studying PINN and related computational methods. Below is the list of the current topics that i have an interest in.

• Physics-informed neural network
  • Application to the Grad-Shafranov solver : GS solver for implementation of the virtual KSTAR environment for shape parameter control
  • Analysis for alternative transport code for MHD regime
• Bayesian machine learning
  • Application to the disruption prediction model in KSTAR
  • Bayesian based imputation for missing data in KSTAR </aside>

Development

📱 App development : WelfareForEveryone (모두의 복지)

<aside> 🤖 Application for welfare information notifying service targeting to vulnerable populations

• Details

  • Chatbot service with SST + TTS

  • Recommendation of welfare info based on user information

  • Map services : Notifying the location of welfare service centers

  • Categorization of the information list

  • Example

    

• Project performance

  • Github repository : official repository

    GitHub - 21WelfareForEveryone/WelfareForEveryOne: 복지사각지대 해소를 위한 맞춤 복지 추천 앱: 2021 프로보노 공모전 대상(과학기술정보통신부장관상) 수상

  • Google play store enrollment

    모두의 복지 - Apps on Google Play

</aside>

📱 Web service development : K-MolOCR

<aside> 🤖 Web service for detecting and converting molecular structure images to SMARTS text

• Details
  • Deep Learning based object detection model for localizing the molecular structure images in PDF files
  • Molecular structure image ↔ SMARTS text including Markush data
  • Free web services for researchers in chemistry
• Project performance

  • Project page (Notion)

    Project : K-MolOCR

</aside>

Skill