Han-Jun Choi

About Me

Hi, I am an AI researcher passionate about solving practical challenges that arise when deploying AI models in real-world settings. I have been working at the Korea Electronics Technology Institute (KETI), where I focus on designing multimodal fusion frameworks that remain robust under incomplete and noisy data. Recently, I have been exploring Vision-Language-Action (VLA) modeling for immersive human–robot interaction. I received a B.S. and an M.S. in Computer and Communication Engineering from Kangwon National University, advised by Prof. Sang-Ki Ko.

News & Achievements

2025.08

I am really excited to share that our short paper "Imputing Multi-Agent Trajectories from Event and Snapshot Data in Soccer" (participated as a co-author) is accepted to CIKM (BK21 Top Conference).

2025.06

The paper "Multimodal Sentiment Analysis under Incomplete Modalities with Missing Text Localization and Confidence-Aware Fusion" is accepted at KIBME and received the Best Paper Award!

2025.05

The paper "Trajectory Imputation in Multi-Agent Sports with Derivative-Accumulating Self-Ensemble" is accepted at ECML PKDD (BK21 Top Conference).

2024.09

The paper "Analysis of Missing Modality Problems with Korean Multimodal Sentiment Data" is accepted at HCLT.

2024.02

The paper "Transformer-Based Multi-Agent Trajectory Imputation Model using Permutation Equivariance" is published at KCTP.

2024.01

I’m excited to share that I’ve started working at the AI Research Center at KETI! at KETI (Korea Electronics Technology Institute).

2023.06

The paper "Deep-learning-based Multi-Agent Trajectory Imputation" is accepted at KCC. Excellent paper award

2023.05

Our paper "Ball trajectory inference from Multi-Agent Sports Contexts Using Set Transformer and Hierarchical Bi-LSTM" is accepted at KDD (BK21 Top Conference). This is my first international publication!

2022.12

The paper "Optimal Data Generation Strategy for Training RNN-based Time-series Data Imputation Models" is published at KSC.

2022.06

The paper "Comparison Study of Deep Learning-based Spatiotemporal Data Imputation Algorithms" is published at KCC.

2022.02

The paper "Deep learning-based workout classification and repetition counting using IMU sensor data" is published at ICTC.

2021.12

The paper "Deep learning-based coordinate tracking algorithm for automated soccer video analysis" is accepted at KSC. Excellent paper award

2021.06

My first paper "Deep learning-based time-series wind speed data imputation algorithm using temperature data" is accepted at KCC!

Work Experience & Education

Jan 2024 - Present

Korea Electronics Technology Institute (KETI)

AI Researcher, AI Research Center

Physical AI-based Vision-Language-Action (VLA) Modeling
- Development of multimodal dialogue model for immersive robot-human interaction
- Review of VLA papers for 3d autonomous characters
- Implementation of VLA model based on Korean emotional dialogue dataset
Multimodal Fusion Modeling
- Multimodal Sentiment Analysis & Emotion Recognition framework development
- Conference paper submission to EMNLP (1st author, under review)
- Conference paper publication at KIBME (1st author, Best Paper Award)
- Conference paper publication at HCLT (1st author, oral presentation)
- Development of a multimodal data preprocessing pipeline (SW program registration)
- Journal submission in preparation (SCI(E), IF: 1.7, Q3)
Multi-Object Trajectory Modeling
- Conference paper publication at PKDD (1st author, International Second-tier conference)
- Conference paper publication at CIKM (3rd author, International Top-tier conference)
Gaze Estimation
- Data preprocessing and label validity verification
- Reviewing key papers in gaze estimation
MetaHuman Avatar Development
- Construction of MetaHuman avatar data for empathetic AI chatbots
- Preprocessing of MetaHuman avatars using 3D mesh-based tools
Other Activities
- ICT technology commercialization (2024 & 2025)
- Research trend seminar presentations

Master's Degree

Kangwon National University (KNU)

M.S. Degree · GPA: 4.38/4.5 · Advisor: Professor Sang-Ki Ko

Research Topics
- Multivariate Time-Series Analysis
- Time-Series Imputation
- Multi-Object Detection
- Multi-Object Tracking
- Multi-Object Imputation
- Spatiotemporal Data Analysis
Research Papers
- [KTCP 2024] Transformer-Based Multi-Agent Trajectory Imputation Model using Permutation Equivariance (1st author, KCI Journal)
- [KDD 2023] Ball trajectory inference from Multi-Agent Sports Contexts Using Set Transformer and Hierarchical Bi-LSTM (2nd author, International Top-tier conference, oral presentation)
- [KCC 2023] Deep-learning-based Multi-Agent Trajectory Imputation (1st author, domestic conference, Excellent Paper Award)
- [KSC 2022] Comparison Study of Deep Learning-based Spatiotemporal Data Imputation Algorithms (2nd author, domestic conference, oral presentation)
- [ITCT 2022] Deep learning-based workout classification and repetition counting using IMU sensor data (1st author, domestic conference, oral presentation)
- [KSC 2022] Optimal Data Generation Strategy for Training RNN-based Time-series Data Imputation Models (1st author, domestic conference, oral presentation)
Teaching Assistant
- Understanding of Artificial Intelligence
- Programming Languages
- Compiler Design

Bachelor's Degree

Kangwon National University (KNU)

B.S. Degree · GPA: 3.88/4.5

Research Topics
- Multivariate Time-Series Analysis
- Time-Series Imputation
- Multi-Object Detection
- Multi-Object Tracking
- Sports Analytics
Research Papers
- [KSC 2021] Deep learning-based coordinate tracking algorithm for automated soccer video analysis (1st author, domestic conference, Excellent Paper Award)
- [KSC 2021] Deep learning-based time-series wind speed data imputation using temperature data (1st author, domestic conference, Oral Presentation)
Other Activities
- Participation in the DACON AI Competition: Solar Power Generation Forecasting
- Participation in the DACON AI Competition: AI-Based Camera Image Quality Enhancement
- Participation in the Joint Hackathon for SW-University
- Participation in the AI Grand Challenge: AI system for mathematical problem solving
- Project: Developing exercise classification and counting model

2024.01 - 현재

한국전자기술연구원 (KETI)

AI 연구원, AI연구센터

Physical AI 기반 Vision-Language-Action (VLA) 모델링
- 로봇-인간간 몰입형 상호작용을 위한 멀티모달 대화 모델 개발
- 3D 캐릭터 생성을 위한 VLA 연구 동향 분석
- 한국어 감성 대화 데이터셋 기반 VLA 모델링(진행 중)
멀티모달 융합 모델링 (Multimodal Fusion Modeling)
- Multimodal Sentiment Analysis & Emotion Recognition 프레임워크 개발
- EMNLP 학술대회 논문 투고 (1저자, 심사 중)
- KIBME 학술대회 논문 게재 (1저자, 최우수논문상)
- HCLT 학술대회 논문 게재 (1저자, 구두 발표)
- 한국어 멀티모달 데이터 전처리 파이프라인 개발 (SW 프로그램 등록)
- 저널 투고 준비 중 (SCI(E), IF: 1.7, Q3)
다중 객체 궤적 모델링 (Multi-Object Trajectory Modeling)
- PKDD 학술대회 논문 게재 (1저자, 국제 우수학술대회)
- CIKM 학술대회 논문 게재 (3저자, 국제 최우수학술대회)
시선 추정 (Gaze Estimation)
- 데이터 전처리 및 라벨 유효성 검증
- 시선 추정 핵심 논문 리뷰
메타휴먼 아바타 개발 (MetaHuman Avatar Development)
- 공감형 AI 챗봇을 위한 메타휴먼 아바타 데이터 구축
- 3D 메쉬 기반 도구를 활용한 메타휴먼 아바타 전처리
기타 활동 (Other Activities)
- ICT 기술 사업화 (2024 & 2025)
- 최신 연구 동향 세미나 발표

Master's Degree

Kangwon National University (KNU)

M.S. Degree · GPA: 4.38/4.5 · Advisor: Professor Sang-Ki Ko

연구 주제 (Research Topics)
- Multivariate Time-Series Analysis
- Time-Series Imputation
- Multi-Object Detection
- Multi-Object Tracking
- Multi-Object Imputation
- Spatiotemporal Data Analysis
연구 논문 (Research Papers)
- [KTCP 2024] Transformer-Based Multi-Agent Trajectory Imputation Model using Permutation Equivariance (1저자, KCI 저널)
- [KDD 2023] Ball trajectory inference from Multi-Agent Sports Contexts Using Set Transformer and Hierarchical Bi-LSTM (2저자, 국제 최우수 학술대회, 구두 발표)
- [KCC 2023] Deep-learning-based Multi-Agent Trajectory Imputation (1저자, 국내 학술대회, 최우수논문상)
- [KSC 2022] Comparison Study of Deep Learning-based Spatiotemporal Data Imputation Algorithms (2저자, 국내 학술대회, 구두 발표)
- [ITCT 2022] Deep learning-based workout classification and repetition counting using IMU sensor data (1저자, 국내 학술대회, 구두 발표)
- [KSC 2022] Optimal Data Generation Strategy for Training RNN-based Time-series Data Imputation Models (1저자, 국내 학술대회, 구두 발표)
조교 활동 (Teaching Assistant)
- 인공지능의 이해
- 프로그래밍 언어
- 컴파일러 설계

Bachelor's Degree

Kangwon National University (KNU)

B.S. Degree · GPA: 3.88/4.5

연구 주제 (Research Topics)
- Multivariate Time-Series Analysis
- Time-Series Imputation
- Multi-Object Detection
- Multi-Object Tracking
- Sports Analytics
연구 논문 (Research Papers)
- [KSC 2021] Deep learning-based coordinate tracking algorithm for automated soccer video analysis (1저자, 국내 학술대회, 최우수논문상)
- [KSC 2021] Deep learning-based time-series wind speed data imputation using temperature data (1저자, 국내 학술대회, 구두 발표)
기타 활동 (Other Activities)
- DACON AI 경진대회 참가: 태양광 발전량 예측
- DACON AI 경진대회 참가: AI 기반 카메라 이미지 화질 향상
- SW중심대학 연합 해커톤 참가
- AI 그랜드 챌린지 참가: 수학 문제 풀이 AI 시스템
- 프로젝트: 운동 분류 및 카운팅 모델 개발

Research Papers

* Equal Contribution 🏆 Award 📌 Key Paper

2026

📌

TCMR: Semantic-aware Text Completeness for Missing Semantic Reconstruction in Incomplete Multimodal Sentiment Analysis

Han-Jun Choi, Byunggill Joe, Saim Shin, Jin Yea Jang

[Under-review] · EMNLP Submission

2025

Imputing Multi-Agent Trajectories from Event and Snapshot Data in Soccer

Geonhee Jo, Miru Hong, Han-Jun Choi, Pascal Bauer and Sang-Ki Ko

[CIKM] · 2025년 8월 | 최우수 국제 학술대회

📌

Trajectory Imputation in Multi-Agent Sports with Derivative-Accumulating Self-Ensemble

Han-Jun Choi*, Hyunsung Kim*, Minho Lee, Minchul Jeong, Chang-Jo Kim, Jinsung Yoon, Sang-Ki Ko

[PKDD] · 2025년 6월 29일 | 우수 국제 학술대회

🏆

Multimodal Sentiment Analysis under Incomplete Modalities with Missing Text Localization and Confidence Aware Fusion

Han-Jun Choi, Byunggill Joe, Jin Yea Jang | 최우수 논문상 🏆

[KIBME] · 2025년 2월 29일

2024

Analysis of Missing Modality Problems with Korean Multimodal Sentiment Data

Han-Jun Choi, Saim Shin, Jin Yea Jang

[HCLT] · 2024년 10월 17일

Transformer-Based Multi-Agent Trajectory Imputation Model using Permutation Equivariance

Han-Jun Choi, Hyun-Woo Park, Sang-Ki Ko

[KTCP] · 2024년 2월 1일

2023

📌

Ball trajectory inference from Multi-Agent Sports Contexts Using Set Transformer and Hierarchical Bi-LSTM

Hyunsung Kim, Han-Jun Choi, Chang-Jo Kim, Jinsung Yoon, Sang-Ki Ko

[KDD] · 2023년 8월 6일 | BK21 최우수 국제 학술대회

🏆

Deep-learning-based Multi-Agent Trajectory Imputation

Han-Jun Choi, Hyun-Woo Park, Sang-Ki Ko

[KCC] · 2023년 6월 / 우수발표 논문상 🏆

2022

Comparison Study of Deep Learning-based Spatiotemporal Data Imputation Algorithms

Park Hyun Woo, Han-Jun Choi, Sang-Ki Ko

[KSC] · 2022년 6월 30일

Deep learning-based workout classification and repetition counting using IMU sensor data

Han-Jun Choi, Min Gi Jeon, Su-Hyeon Kim, Sang-Ki Ko, Hal-Gyun Jeong

[ITCT] · 2022년 2월 9일

Optimal Data Generation Strategy for Training RNN-based Time-series Data Imputation Models

Han-Jun Choi, Park Hyun Woo, Sang-Ki Ko

[KSC] · 2022년 1월 19일

2021

🏆

Deep learning-based coordinate tracking algorithm for automated soccer video analysis

Han-Jun Choi, Hyun-Woo Park, Hyun-Sung Kim, Dong-Wook Chung, Jongeun Chung, Sang-Ki ko and Jongeun Chung

[KSC] · 2021년 12월 20일 / 우수 논문상 수상 🏆

Deep learning-based time-series wind speed data imputation algorithm using temperature data

Han-Jun Choi, Sang-Ki Ko, and Jongwoon Lim

[KSC] · 2021년 6월 23일

Awards

2025

Best Paper Award

1st Place

KIBME 2025 (Korean Institute of Broadcast and Media Engineers)

Multimodal Sentiment Analysis under Incomplete Modalities with Missing Text Localization and Confidence Aware Fusion

February 29, 2025

2023

Excellent Paper Award

Outstanding

KCC 2023 (Korea Computer Congress)

Deep-learning-based Multi-Agent Trajectory Imputation

June 30, 2023

2021

Excellent Paper Award

Outstanding

KSC 2021 (Korea Software Congress)

Deep learning-based coordinate tracking algorithm for automated soccer video analysis

December 20, 2021

Resume / Portfolio

Resume

Curriculum Vitae · Updated April 2026

Open / Download

Portfolio

Research & Project Highlights · June 2026

📄 Click to View PDF

Open / Download

Selected Papers

[Under-review] · EMNLP Submission

TCMR: Semantic-aware Text Completeness for Missing Semantic Reconstruction in Incomplete Multimodal Sentiment Analysis

Han-Jun Choi, Byunggill Joe, Saim Shin, Jin Yea Jang

연구 목표

데이터의 노이즈·결측 문제로 인한 감정 인식 성능 저하 문제 해결
불완전 데이터 환경에 강건한 멀티모달 융합 모델 설계

제안 방법

텍스트 신뢰도 기반 멀티모달 융합 프레임워크 TCMR 제안
결측 텍스트 내 남은 감정 관련 정보량 추정 방식 제안
신뢰도 추정 네트워크 학습을 위한 정답 레이블 생성 기법
신뢰도 기반 손실된 감정 정보 복원 모듈 구현
멀티태스크 학습 안정화를 위한 최적화 기법 설계

실험 결과

벤치마크 데이터셋(MOSI, MOSEI, SIMS)에 대한 성능 평가
다양한 결측 비율(0~100%)에서의 성능 변화 분석
기존 방법론 대비 일관된 성능 우위 달성
예측 정확도(Acc) 약 6.68% 향상, 오차(MAE) 약 6.37% 감소
실세계 데이터 환경에서 신뢰도 있는 감정 예측이 가능함을 입증

성과

EMNLP 2026 제출 (국제 최우수 학술대회)
1차 심사 결과 발표: 2026년 5월 예정
최종 결과 발표: 2026년 7월 예정

Research Objectives

Mitigate performance degradation under noisy & missing data
Develop robust multimodal fusion for real-world settings

Proposed Method

Confidence-aware multimodal framework (TCMR)
Sentiment-relevant information estimation
Pseudo-labeling for confidence supervision
Confidence-guided reconstruction
Stabilized multi-task optimization

Experimental Results

Three benchmark datasets (MOSI, MOSEI, SIMS)
Evaluated across various missing rates (0–100%)
Acc ↑ 6.68%, MAE ↓ 6.37%
Demonstrating robustness to noise and missing data

Outcome

EMNLP 2026 (under review)
First-round decision: May 2026 (expected)
Final decision: July 2026 (expected)

ECML PKDD 2025

Trajectory Imputation in Multi-Agent Sports with Derivative-Accumulating Self-Ensemble

Han-Jun Choi*, Hyunsung Kim*, Minho Lee, Minchul Jeong, Chang-Jo Kim, Jinsung Yoon, Sang-Ki Ko

연구 목표

다중 객체 궤적 데이터의 결측치로 인한 분석 성능 저하 문제 해결
생체역학적 제약을 반영한 결측 궤적 보간 프레임워크 설계

제안 방법

다중 객체간 복잡한 상호작용 모델링
궤적 데이터의 순열 동변성을 고려한 임베딩 제안
위치·속도·가속도간 물리적 관계식 기반 궤적 예측 모듈 구현
Soft Voting 기반 동적 융합 모듈 설계

실험 결과

팀 스포츠(축구, 농구, 미식축구) 데이터셋에 대한 성능 평가
서로 다른 세 가지 결측 시나리오에 따른 성능 변화 분석
State-of-the-art 성능 달성
위치 오차 최대 37% 감소, 궤적 현실성 지표 약 36.7% 향상
현실 객체 움직임과 가장 유사한 궤적을 생성할 수 있음을 입증

성과

ECML PKDD 2025 게재 (국제 우수 학술대회)

KDD 2023

Ball trajectory inference from Multi-Agent Sports Contexts Using Set Transformer

Hyunsung Kim, Han-Jun Choi, Chang-Jo Kim, Jinsung Yoon, Sang-Ki Ko

연구 목표

축구 경기 자동 중계를 위한 공 궤적 취득 연구
선수의 궤적 정보를 활용한 공 궤적 추론 프레임워크 설계

제안 방법

경기 내용 전반의 컨텍스트 표현을 위한 순열 불변 임베딩 제안
자연스러운 궤적 예측을 위한 물리적 제약 기반 손실 함수 설계
도메인 지식 학습을 통한 계층적 추론 구조 설계
룰 기반 후처리 알고리즘을 통한 예측 궤적 보정

실험 결과

축구 시뮬레이션 데이터셋에 대한 성능 평가
State-of-the-art (SOTA) 성능 달성
기존 SOTA 모델 대비 궤적 예측 정확도 약 25.6% 향상
예측 궤적을 활용한 경기 자동 중계, 공 소유권 예측 문제 적용
현실 세계에 다양한 실용적인 태스크에 적용이 가능함을 입증

성과

KDD 2023 게재 (국제 최우수 학술대회)

Recommendations

Thank you for your interest. Below are my recommendation letters from distinguished professionals.

About Me

Publications

Experiences

Awards

Resume / Portfolio

Selected Papers

Recommendations

About Me

News & Achievements

Work Experience & Education

Korea Electronics Technology Institute (KETI)

Kangwon National University (KNU)

Kangwon National University (KNU)

한국전자기술연구원 (KETI)

Kangwon National University (KNU)

Kangwon National University (KNU)

Research Papers

TCMR: Semantic-aware Text Completeness for Missing Semantic Reconstruction in Incomplete Multimodal Sentiment Analysis

Imputing Multi-Agent Trajectories from Event and Snapshot Data in Soccer

Trajectory Imputation in Multi-Agent Sports with Derivative-Accumulating Self-Ensemble

Multimodal Sentiment Analysis under Incomplete Modalities with Missing Text Localization and Confidence Aware Fusion

Analysis of Missing Modality Problems with Korean Multimodal Sentiment Data

Transformer-Based Multi-Agent Trajectory Imputation Model using Permutation Equivariance

Ball trajectory inference from Multi-Agent Sports Contexts Using Set Transformer and Hierarchical Bi-LSTM

Deep-learning-based Multi-Agent Trajectory Imputation

Comparison Study of Deep Learning-based Spatiotemporal Data Imputation Algorithms

Deep learning-based workout classification and repetition counting using IMU sensor data

Optimal Data Generation Strategy for Training RNN-based Time-series Data Imputation Models

Deep learning-based coordinate tracking algorithm for automated soccer video analysis

Deep learning-based time-series wind speed data imputation algorithm using temperature data

Awards

Best Paper Award

Excellent Paper Award

Excellent Paper Award

Resume / Portfolio

Resume

Portfolio

Selected Papers

TCMR: Semantic-aware Text Completeness for Missing Semantic Reconstruction in Incomplete Multimodal Sentiment Analysis

연구 목표

제안 방법

실험 결과

성과

Research Objectives

Proposed Method

Experimental Results

Outcome

Trajectory Imputation in Multi-Agent Sports with Derivative-Accumulating Self-Ensemble

연구 목표

제안 방법

실험 결과

성과

Research Objectives

Proposed Method

Experimental Results

Outcome

Ball trajectory inference from Multi-Agent Sports Contexts Using Set Transformer

연구 목표

제안 방법

실험 결과

성과

Research Objectives

Proposed Method

Experimental Results

Outcome

Paper Title Goes Here

Keywords

Recommendations

Recommendation Letter #1:서울시립대 고상기 교수님

Recommendation Letter #2:한국전자기술연구원 장진예 선임님

Recommendation Letter #3:한국전자기술연구원 조병길 선임님

Recommendation Letter #1:
서울시립대 고상기 교수님

Recommendation Letter #2:
한국전자기술연구원 장진예 선임님

Recommendation Letter #3:
한국전자기술연구원 조병길 선임님