I am Junha Song, a graduate with an M.S. degree advised by Prof. In So Kweon, and currently a Ph.D. student under the supervision of Prof. Jaegul Choo from KAIST. I was a research engineer with industry-university scholarship at Hyundai Mobis in 2021-2022 and a research intern at Qualcomm AI Research in 2022 and Lunit Global in 2023. I invite you to explore my blog from which you find that I am a highly self-motivated researcher. My ultimate goal is to develop AI that benefits all individuals, regardless of their socioeconomic status.
[News] I will be studying at Carnegie Mellon University as a Korean government-sponsored exchange student. Please refer to Education.
Research Experiences
- Lunit Global
Jun 2023 - Dec 2023
AI Research Intern
Mentors: Tae Soo Kim and Thijs Kooi - Qualcomm AI Research
Jul 2022 - Dec 2022
AI Research Intern
Mentor: Sungha Choi - Hyundai Mobis
Mar 2021 - Jun 2022
Research Engineer with industry-university scholarship
Autonomous driving advanced development team
Publications
-
Test-time Adaptation in the Dynamic World with Compound Domain Knowledge Management
Junha Song, Kwanyong Park, Inkyu Shin, Sanghyun Woo, Chaoning Zhang, and In So Kweon
In IEEE Robotics and Automation Letters (RA-L, ICRA), 2024
[pdf], [article], [presentation] -
A Survey on Masked Autoencoder for Self-supervised Learning in Vision and Beyond
Chaoning Zhang, Chenshuang Zhang, Junha Song, John Seon Keun Yi, and In So Kweon
In International Joint Conference on Artificial Intelligence (IJCAI), 2023.
[pdf], [slide] -
EcoTTA: Memory-Efficient Continual Test-time Adaptation via Self-distilled Regularization
Junha Song, Jungsoo Lee, In So Kweon, and Sungha Choi
In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023.
[pdf], [project page]
Research Interests
- Adaptation with foundation models
- Powerful foundation models, such as CLIP, SAM, and Stable Diffusion, are trained on mountains of data and thus possess a remarkable capacity for understanding a wide range of images. I am captivated by their potential and eager to leverage them so as to develop and adapt ML models for real-world products. This endeavor holds immense promise in mitigating the out-of-distribution generalization problem, ensuring that AI systems can reliably perform across diverse scenarios.
- Exploring the opportunities of foundation models (papaer from CRFM at the Standord).
- On-device adaptation frameworks
- Despite advances in deep learning, the AI model often struggles with performance degradation due to environmental changes. For example, the cognitive ability of self-driving cars can change depending on time, weather, and city-state. To address this, I am fascinated with Test-time Adaptation, where we allow the model to adapt itself to a new environment. I believe this technique would be key to ensuring robust performance of the model and ultimately building reliable AI applications.
- Test-time adaptation / Federated learning with client-provided weak labels (like HFRL).
Education
- Carnegie Mellon University (CMU)
Aug 2024 - Mar 2025
Exchange student in Computer Science
Supported by a Korean government - Korea Advanced Institute of Science and Technology (KAIST)
Aug 2023 - Present
Ph.D student in Graduate School of AI
Advisor: Prof. Jaegul Choo - Korea Advanced Institute of Science and Technology (KAIST)
Feb 2021 - Feb 2023
M.S. degree in the Division of Future Vehicle
Advisor: Prof. In So Kweon
Grade: 3.9 / 4.3 (Percent: 95.56/100) - Kookmin University (Seoul, South Korea)
Feb 2015 - Feb 2021
B.S. degree in IT and Automobile Engineering
Grade: 4.39 / 4.5 (Rank: 1/121 | Percent: 98.7/100 | Major: 4.43)
National Science and Engineering Scholarship (Full tuition) from Korea Student Aid Foundation
Mandatory Military Service for 21 Months
Awards and Honors
- Best Master's Thesis Award, Korea Advanced Institute of Science and Technology (KAIST) (2023)
- Lecture planning consultant, Fast Campus (2022)
- National Science and Engineering Scholarship (Full tuition for a BS.D. student) , Korea Scholarship Foundation
- Future Transport Design Award and Honorable Judge Award, 'Vehicle monitoring over internet toward digital twins', Cloud Programming World Cup, Japan (2019)
- Capstone Awards, Korean Society of Automotive Engineers (2019)
Projects
- Development of real-time masking/unmasking system for personal video information for public services such as CCTV (article), Korea Ministry of Science and ICT (2021 - 2023)
- Development of segmentation networks robust to environment variance, Hyundai Mobis (2021)
- Satellite image precision object detection, Korea Agency for Defense Development (ADD) (2020)
- Detection of Surrounding Vehicles using Deep Neural Network and Fusion of Panoramic Camera and Lidar Sensor, Korea Foundation for the Advancement of Science and Creativity (KORAC), Korea (2019)
B.S. Research Experiences
- "Style Transfer Maps from Satellite Images by using Generative Model", Korean Institute of Communications and Information Science (KICS) (2020)
- "Improvement of LiDAR and IMU-based autonomous driving performance in right-angle corner situations", Korean Sociey of Automotive Engineers (2019)
- Research Intern at Machine Intelligence Lab, Kookmin University (Dec 2019 - Oct 2020)
- Research Intern at Intelligence and Interaction Lab, Kookmin University (Feb 2019 - Nov 2019)
Skills
- Programming language: Python, C++
- Machine Learning Librarie: Pytorch, Tensorflow
- Application development: Robot Operating System (ROS)
- Sensor utilization: Camera, RGB-D Camera, LiDAR, GPS/IMU