Junhyeok Kang

I am an AI Research Scientist at LG AI Research, specializing in AI for time-series data. I received my Ph.D. degree from KAIST in August 2024, focusing on developing innovative methodologies for sequential data. I also bring extensive teaching expertise, having delivered over 300 hours of AI training for major industry and government institutions, including Samsung Electronics, the National Tax Service, and Statistics Korea.

Please feel free to contact me for research discussions or potential collaborations!

Work Experience

AI Research Scientist @ LG AI Research | Full-time | Nov. 2024 – Present

• Research on time-series foundation models and agentic frameworks.

PhD Intern @ LG AI Research | Internship | Aug. 2024 – Nov. 2024

• Research on time-series forecasting.

AI Instructor @ Elice | Freelance | Jul. 2019 – Sep. 2023

• Delivered 250+ hours of AI instruction at Samsung Electronics, including executive training.

AI Researcher @ Korea Telecom(KT) | Contract  | May. 2020 – Sep. 2020

• Develop a predictive model for the self-diagnosis of infectious diseases using cellular trajectories [news]

Graduate Student Researcher @ KAIST | Full-time | Feb. 2018 – Aug. 2024

• Research on time-series analysis and large-language models.

• Lab representative, Data Mining Lab (Apr 2022 - Jun 2024).

Projects

Foundation Models for Time-series @ LG AI Research | Jan 2026 - Current [news]

• EXAONE-Forecast 📈: Key contributor to the development of a zero-shot forecasting method with time-series foundation models, achieving #1🥇 on the GIFT-Eval zero-shot benchmark leaderboard (as of Aug 2026).

• EXAONE-Forecast-Agent 🤖: Key contributor to the development of  an agentic ensemble framework leveraging time-series foundation models, achieving #1🥇 on the GIFT-Eval benchmark leaderboard (as of Aug 2026).

Commodity Price Forecasting @ LG AI Research | Jan 2025 - Current

• Developed an explainable time-series model integrated with an agentic framework capable of event-driven analysis, advanced error calibration, and natural language explanations for lithium price predictions (for LG Energy Solution). [blog]

Education

Ph.D., School of Computing @ KAIST | Mar. 2020 – Aug. 2024

• Data Mining Lab (Advisor: Jae-Gil Lee) 

• Dissertation: Time Series Forecasting via Time-Frequency Domain Analysis

M.S., Graduate School of Data Science* @ KAIST | Mar. 2018 – Feb. 2020

• Data Mining Lab (Advisor: Jae-Gil Lee)

• *Formerly, Graduate School of Knowledge Service Engineering

B.S., School of Global Entrepreneurship and ICT @ Handong Global University  | Feb. 2012 – Feb. 2018

• B.S. in ICT Convergence and B.A. in Management (Double Major), Summa Cum Laude

Publications

International Conference & Journal

1. Channel-wise Retrieval for Multivariate Time Series Forecasting [paper]
   Junhyeok Kang†, Jun Seo, Soyeon Park, Sangjun Han, Seohui Bae, Hyeokjun Choe, Soonyoung Lee
   ICASSP 2026 Oral

2. Traceable Multi-Agent System for Knowledge-Based Forecasting [paper]
   Junhyeok Kang*, Sangjun Han*, Hyeokjun Choe*, Soonyoung Lee
   KDD Workshop on Enterprise AI Agents 2026 Oral - 🏆Best paper award

3. Universal Time-Series Representation Learning: A Survey  [paper]
   Patara Trirat, Yooju Shin, Junhyeok Kang, Youngeun Nam, Jihye Na, Minyoung Bae, Joeun Kim, Byunghyun Kim, Jae-Gil Lee
   ACM Computing Surveys 2026

4. CURE: Cumulative Knowledge Reuse for Efficient Device-Server Hybrid Inference in Vision-Language Models
   Youngjun Lee, Doyoung Kim, Junhyeok Kang, Hwanjun Song, Jae-Gil Lee
   ECCV 2026

5. VarDrop: Enhancing Training Efficiency by Reducing Variate Redundancy in Periodic Time Series Forecasting [paper, code, blog]
   Junhyeok Kang, Yooju Shin, and Jae-Gil Lee
   AAAI 2025

6. RA-TTA: Retrieval-Augmented Test-Time Adaptation for Vision-Language Models [paper, code]
   Youngjun Lee, Doyoung Kim, Junhyeok Kang, Jihwan Bang, Hwanjun Song, Jae-Gil Lee
   ICLR 2025

7. Mitigating Source Label Dependency in Time-Series Domain Adaptation under Label Shifts [paper, code, news]
   Jihye Na, Youngeun Nam, Junhyeok Kang, Jae-Gil Lee
   KDD 2025

8. Large Language Models are Zero-Shot Point-of-Interest Recommenders [paper, code]
   Joeun Kim, Youngjin Seo, Yeonsoo Kim, Junhyeok Kang, Jeeho Shin, Patara Trirat, Jae-Gil Lee
   Data Mining and Knowledge Discovery 2025

9. Multi-View POI-level Cellular Trajectory Reconstruction for Digital Contact Tracing of Infectious Diseases [paper]
   Dongmin Park, Junhyeok Kang, Hwanjun Song, Susik Yoon, and Jae-Gil Lee
   ICDM 2022

10. Hi-COVIDNet: Deep Learning Approach to Predict Inbound COVID-19 Patients and Case Study in South Korea [paper,  news]
    Minseok Kim, Junhyeok Kang, Doyoung Kim, Hwanjun Song, Hyangsuk Min, Youngeun Nam, Dongmin Park, Jae-Gil Lee
    KDD 2020 - selected as one of 2024 KAIST Top 10 Research Achievements[link]

11. Robust Sequential Variational Autoencoder for Multivariate Time Series Anomaly Detection
    Yao Yu, Junhyeok Kang, Jae-Gil Lee, Jonghwa Kim, Kyungdeok Seo
    IJCAI Workshop on AI for Internet of Things 2020

Tech Reports

1. FinVerse: Financial Time-Series Benchmark [paper]
   Jaehoon Lee, Jun Seo, Seunghan Lee, Tae Yoon Lim, Dongwan Kang, Hwanil Choi, Minjae Kim, Sungdong Yoo, Junhyeok Kang, Sangjun Han, Soonyoung Lee, Wonbin Ahn
   LG AI Research

Preprints

1. ReasonCast: Towards Explainable Time Series Forecasting with Reasoning [paper]
   Seunghan Lee, Jun Seo, Jaehoon Lee, Junhyeok Kang, Sangjun Han, Sungdong Yoo, Minjae Kim, Tae Yoon Lim, Dongwan Kang, Hwanil Choi, Soonyoung Lee, Wonbin Ahn
   arXiv:2608.01875, 2026.

2. Beyond Magnitude and Shape: A Direction-Aware Loss for Time Series Forecasting [paper]
   Seunghan Lee, Jaehoon Lee, Jun Seo, Junhyeok Kang, Sangjun Han, Sungdong Yoo, Minjae Kim, Tae Yoon Lim, Dongwan Kang, Hwanil Choi, Soonyoung Lee, Wonbin Ahn
   arXiv:2608.01857, 2026.

3. Adaptive Information Routing for Multimodal Time Series Forecasting [paper]
   Jun Seo, Hyeokjun Choe, Seohui Bae, Soyeon Park, Wonbin Ahn, Taeyoon Lim, Junhyeok Kang, Sangjun Han, Jaehoon Lee, Dongwan Kang, Minjae Kim, Sungdong Yoo, Soonyoung Lee
   arXiv:2512.10229, 2025.

Domestic Conference & Journal

1. Linear Probing without Pre-Training for Time Series Forecasting  [paper]
   Junhyeok Kang, Minyoung Bae, Yeonsu Kim, Jae-Gil Lee
   Korea Computer Congress, 2024.

2. Reinforcement Learning for Time Series Forecasting  [paper]
   Minyoung Bae, Jaehyun Park, Youngjin Seo, Junhyeok Kang, Jae-Gil Lee
   Korea Computer Congress, 2024.

3. An Empirical Investigation of Deep Learning Models for Defect Classification in Solar Cells Electroluminescence [paper]
   Haenara Shin, Junhyeok Kang, Irfan Akbar, Seola Choi, Youngeun Nam, Jae-Gil Lee
   Korea Computer Congress, 608-610, 2023.

4. Group Periodic Pattern Mining Considering Concurrency from Spatio-temporal Trajectories [paper]
   Jihye Na, Junhyeok Kang, Jae-Gil Lee, Wooshin Lee, Soyeon Jin, Sangmin Kim
   Database Research, Vol 38, No. 2, pp.3-17, 2022.

5. Contrastive Learning with Segment Supervision for Semi-supervised Time Series Classification [paper]
   Yooju Shin, Dongmin Park, Junhyeok Kang, Sejin Kim
   Korea Computer Congress, 842-843, 2022.

6. ActiveBoostThief: Model Extraction Attack Using Reliable Active Learning [paper]
   Youngeun Nam, Junhyeok Kang, Jae-Gil Lee
   Korea Computer Congress, pp. 594-596, 2021.

7. Deep Grid Embedding using a Road Network Topology for Route Prediction [paper]
   Junhyeok Kang, Jae-Gil Lee, Byeongjin Kim, Suwon Lee
   Database Research, Vol 36, No. 1, pp.57-73, 2020.

8. Anomalous Trajectory Detection Based on Seq2Seq Auto-Encoder [paper]
   Hyunsoo Kim, Junhyeok Kang, Howon Moon, Jae-Gil Lee
   Journal of Korean Society for Geospatial Information Science, Vol. 28, No. 1, pp. 35-40, 2020.

9. Latent Representation Learning for Autoencoder-based Top-K Recommender System [paper]
   Dongmin Park, Junhyeok Kang, Jae-Gil Lee
   Journal of KIISE, Vol. 47, No. 2, pp. 207-215, 2020.

10. Analysis of Trajectory Encoding Methodology Analysis Based on Experimental Evaluation for Deep Learning [paper]
    Junhyeok Kang, Minseok Kim, Jae-Gil Lee
    KIISE Transactions on Computing Practices, Vol. 25, No. 8, pp. 402-406, 2019.

11. A Survey on Deep Learning-based Anomaly Detection Models for Time Series Data [paper]
    Yao Yu, Junhyeok Kang, Jae-Gil Lee
    Korea Computer Congress, pp. 919-921, 2019.

12. Experimental Comparison of Trajectory Encoding Methods for Deep Learning [paper]
    Junhyeok Kang, Jae-Gil Lee, Yoomi Park
    Korea Software Congress, pp. 204-206, 2018.

Patents

1. Method and Apparatus of Variate Tokenization for Time Series Forecasting
   Lee, J., Kang, J., and Shin, Y.
   Korean Patent Application No: 10-2024-0152425, Oct. 31, 2024.

2. Method for Generating Pattern Information and Electronic Apparatus Supporting Thereof
   Lee, W., Kim, S., Na, J., Kang, J., and Lee, J.
   Korean Patent Publication No: 10-2024-0057816, May. 03, 2024. 

3. Trajectories Embedding Method for Deep Learning and Route Prediction Method Using the Same
   Lee, J., Kang, J., Kim, M., and Lee, J.
   Korean Patent No:10-2647224-0000, Mar. 08, 2024.

4. Method and Apparatus for Predicting Imported Infectious Disease Information Based on Deep Neural Networks
   Lee, J, Kim, M., Kang, J., Kim, D., Song, H., Min, H., Nam, Y., and Park, D.
   US Patent Registration No: US11557401B2, Jan. 17, 2023.

5. Method and Apparatus for Predicting Confirmed Patients of Infectious Disease Based on Deep Neural Networks
   Lee, J, Kim, M., Kang, J., Kim, D., Song, H., Min, H., Nam, Y., and Park, D.
   Korean Patent No: 10-2349270-0000, Jan. 05, 2022.

Invited Lectures & Industry Training

• AI, From Research to Product @ CJ OliveNetworks | 2 hours | Jul. 2025

• Introduction to Time-series Forecasting @ AI Camp at Handong Global University | 8 hours | Jul. 2025

• Digital Transformation Training for Experienced Employees @ Samsung Electronics Co., Ltd.  | 84 hours | Jan. 2023 - Apr. 2023

• Digital Agility Training for Executives @ Samsung Electronics Co., Ltd. | 16 hours | Nov. 2022

• Digital Transformation Training for New Employees @ Samsung Electronics Co., Ltd.  | 178 hours | Oct. 2022 - Sep. 2023

• Machine Learning @ Statistics Korea | 7 hours | Jul. 2022

• Regression Analysis @ National Tax Service | 12 hours | Jun. 2022

• Cluster Analysis @ National Tax Service | 10 hours | May. 2022

• Building Chatbots in Python @ Samsung Software Academy for Youth (SAFFY) | 40 hours | Jul. 2019

Teaching Experiences

• Teaching Assistant of KAIST AIB (AI Business Transformation Program for CEOs; 2nd and 3rd period) @ KAIST | 2021-2022 [news]

• Python Programming @ KAIST IT academy | Summer 2022

• Teaching Assistant of KSE525: Data Mining and Knowledge Discovery @ KAIST ISysE | Spring 2020

• Teaching Assistant of Programming II @ Handong Global University | 2017

• Teaching Assistant of Programming I @ Handong Global University | 2016-2017

Awards & Honors

• 1st Prize, Biomedical Object Detection Contest (USD 1,500) @ AIFactory | 2020

• Best Presentation Paper Award @ Korea Software Congress | 2018

• Summa Cum Laude @ Handong Global University | 2018