Changhun Kim

Changhun Kim

Machine Learning Researcher

AITRICS

I am a machine learning researcher at AITRICS, a healthcare AI startup in South Korea, where I am fulfilling my alternative military service under Prof. Eunho Yang. Before joining AITRICS, I obtained my master’s degree in artificial intelligence from KAIST, also under the guidance of Prof. Eunho Yang. I completed my bachelor’s degree in computer science and mathematics at KAIST as well. I am open to research collaborations globally, including remote opportunities. If you are interested in my work, please feel free to reach out!

My long-term research objective is to enhance the out-of-distribution generalization capability of machine learning models, thereby creating trustworthy AI systems that can be reliably deployed in new environments. This multifaceted goal involves ensuring robustness to distribution shifts (domain adaptation and generalization), handling unseen labels (zero-shot learning), and managing unseen tasks (cross-task generalization). During my master’s program, I focused on test-time adaptation to distribution shifts across various tasks, including 3D point cloud recognition, zero-shot generalization of vision-language models, automatic speech recognition, tabular learning, and time series classification. These experiences have also enabled myself to quickly adapt to new modalities and tasks.

At AITRICS, my research centers on improving the generalizability, robustness, and explainability of early prediction models for critical clinical outcomes such as cardiac arrests and in-hospital mortality. I am also interested in parameter- and data-efficient adaptation of deep generative models, such as diffusion models and large multimodal models, to downstream tasks. To achieve these goals, I work on developing practical algorithms that are empirically well-grounded or theoretically provable. Additionally, I am passionate about providing theoretical insights into machine learning models through the lens of probabilistic (Bayesian inference) and statistical (generalization bounds) frameworks.

Contact
  • 13F, 218, Teheran-ro, Gangnam-gu, Seoul, Korea
  • (+82) 10-3264-6509
  • changhun.a.kim@gmail.com
Education
  • M.S. in Artificial Intelligence, 2024

    Korea Advanced Institute of Science and Technology (KAIST)

  • B.S. in Computer Science and Mathematics, 2022

    Korea Advanced Institute of Science and Technology (KAIST)

News
See all
Professional Experience

AITRICS
AITRICS, Seoul, South Korea
Machine Learning Researcher, Nov 2023 – Present

KAIST Machine Learning and Intelligence Lab
KAIST Machine Learning and Intelligence Lab, Daejeon, South Korea
Master’s Student Researcher, Mar 2022 – Feb 2024
Undergraduate Researcher, Jun 2021 – Feb 2022

KAIST Applied Artificial Intelligence Lab
KAIST Applied Artificial Intelligence Lab, Daejeon, South Korea
Developer, Sep 2021 – Jan 2022

DeepNatural
DeepNatural, Seoul, South Korea
Machine Learning Engineer, Sep 2020 – Feb 2021

KAIST Vehicular Intelligence Lab
KAIST Vehicular Intelligence Lab, Daejeon, South Korea
Undergraduate Researcher, Oct 2019 – Aug 2020

Netmarble
Netmarble, Seoul, South Korea
Data Engineer, Jun 2019 – Aug 2019
Publications
Stable-TTS: Stable Speaker-Adaptive Text-to-Speech Synthesis via Prosody Prompting
Stable-TTS: Stable Speaker-Adaptive Text-to-Speech Synthesis via Prosody Prompting

Manuscript, 2024
AdapTable: Test-Time Adaptation for Tabular Data via Shift-Aware Uncertainty Calibrator and Label Distribution Handler
AdapTable: Test-Time Adaptation for Tabular Data via Shift-Aware Uncertainty Calibrator and Label Distribution Handler

Manuscript, 2024
arXiv
CloudFixer: Test-Time Adaptation for 3D Point Clouds via Diffusion-Guided Geometric Transformation
CloudFixer: Test-Time Adaptation for 3D Point Clouds via Diffusion-Guided Geometric Transformation

European Conference on Computer Vision (ECCV), 2024
Paper arXiv Code
SGEM: Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization
SGEM: Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization

Conference of the International Speech Communication Association (INTERSPEECH), 2023, Oral Presentation
Paper arXiv Code
Patents
Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization

US Patent Application No. 18/594,442 (Pending), Korea Patent Application No. 10-2024-0006413 and 10-2024-0023266 (Pending)
Honors

Dongwon Full Masters Scholarship
Dongwon Group, Spring 2022 – Fall 2023

Magna Cum Laude
KAIST College of Engineering, Feb 2022

Silver Prize
Korean Undergraduate Mathematics Competition, Korean Mathematics Society, Jan 2022

Overseas Exchange Scholarship
Mirae Asset, Dec 2019

Representative of Student Exchange Ambassador
KAIST, Nov 2019

Honor Student
KAIST College of Engineering, Sep 2019

KAIST Convergence AMP Scholarship
KAIST School of Computing, Mar 2019

Winner
Science Quiz, KAIST-POSTECH Science War, Sep 2018

National Full Undergraduate Scholarship
Korea Student Aid Foundation, Spring 2017 – Fall 2021
Teaching Experience

Teaching Assistant
Tabular Learning, Hanwha Ocean Capstone Project, Spring 2023

Teaching Assistant
AI Soccer Challenge, Bokja Girls’ High School AI Education Program, Fall 2020
Academic Service

Journal Reviewer

Conference Reviewer

Workshop Reviewer
Projects

Meta-Learning Applicable to Real-World Problems
Institute of Information & Communications Technology Planning & Evaluation (IITP), May 2023 – Feb 2024

Development of Plug and Play Explainable Artificial Intelligence Platform
Institute of Information & Communications Technology Planning & Evaluation (IITP), Mar 2022 – Apr 2023

Confidence Interval Estimation and Performance Relationship Analysis for Tire Performance Prediction Models
Hankook Tire & Technology, Nov 2023 – Jan 2024

Integrated Tire Performance Prediction Model Using Tire Pattern Features
Hankook Tire & Technology, Mar 2022 – Apr 2023

Convergence Analysis of Deep Learning Optimizers Under Generalized Smoothness
KAIST AI616: Deep Learning Theory, Sep 2023 – Dec 2023

How Many Times are We Going to Collaborate?
KAIST AI607: Graph Mining and Social Network Analysis, Sep 2022 – Dec 2022

Few-Shot Font Generation for Korean
KAIST AI604: Deep Learning for Computer Vision, Mar 2022 – Jun 2022

Issue Trend Analysis and Issue Tracking Analysis
KAIST CS474: Text Mining, Mar 2021 – Jun 2021

KAIST Educational Network System (KENSv3)
KAIST CS341: Introduction to Computer Networks, Mar 2021 – Jun 2021

TinyMutator: Mutation Testing Tool for Rust
KAIST CS453: Automated Software Testing, Mar 2020 – Jun 2020

Immersion Camp
KAIST CS496: Intensive Programming and Startup, Dec 2019 – Jan 2020

PintOS
KAIST CS330: Operating Systems and Lab, Mar 2019 – Jun 2019