My name is Kyungjin Im.
I am a undergraduate researcher at the Brain and Machine Intelligence Lab. (BMIL.), Sungkyunkwan University, Republic of Korea.
My research topic is LLM, VLA, model merging, and federated learning.
Contact: kyungjin DOT im03 AT gmail DOT com
[May. 2026] Our paper "Compact Merging of Vision-Language-Action Models: Shared Cores and Spectral Residuals" has been accepted to the International Conference on Machine Learning (ICML) WiML Workshop.
[May. 2026] Our paper "Post-Hoc Merging is Not Enough: Many-Shot Model Merging with Loss-Gap Balancing" has been accepted to the International Conference on Machine Learning (ICML).
[Apr. 2026] Our paper "Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift" has been selected as a Highlight paper (Top 10% of accepted papers) at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
[Feb. 2026] Our paper "Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift" has been accepted to the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
[Dec. 2025] Our domestic journal article "Personalized Federated Post-training for Real-time Adaptation to Label Distribution Shifts," has been chosen as the best paper in J-KICS.
[Dec. 2025] Won the Best Paper Award at the Graduation Thesis Presentation, Soongsil University.
[Nov. 2025] Won the Excellence Award of Soongsil Patent Competition, Soongsil University.
[Nov. 2025] Won the Excellence Award at the SW-Centered University Essay Contest.
[Aug. 2025] Our domestic journal article "Personalized Federated Post-training for Real-time Adaptation to Label Distribution Shifts," has been accepted to The Journal of Korean Institute of Communications and Information Sciences (J-KICS).
[Aug. 2025] Our domestic journal article "Implementation of a Real-Time Pothole Detection and Depth Estimation System Using YOLOv5 and MiDaS with User Alert Functionality," has been accepted to The Journal of the Korea Contents Association.
[Aug. 2025] Our domestic conference paper "FIM-based Personalized Federated Post-training Solution Robust to Data Distribution Shift" has been chosen as the best paper in KCC.
[Jul. 2025] Our domestic conference paper "FIM-based Personalized Federated Post-training Solution Robust to Data Distribution Shift" has been presented to Korea Computer Congress (KCC).
[Apr. 2025] Our domestic conference paper "Federated Post-training for Personalized Client Models" has been presented to the Joint Conference on Communications and Information (JCCI).
M.S., Department of Electrical and Computer Engineering (Sep. 2026 -)
B.S., School of AI Convergence (Mar. 2022 - Aug. 2026)
Kyungjin Im, Heewon Park, Minhae Kwon, "Compact Merging of Vision-Language-Action Models: Shared Cores and Spectral Residuals" International Conference on Machine Learning (ICML) WiML Workshop, July 2026.
Kyungjin Im*, Miru Kim*, Chanin Eom*, Minhae Kwon, “Post-Hoc Merging is Not Enough: Many-Shot Model Merging with Loss-Gap Balancing”, International Conference on Machine Learning (ICML), July 2026. (*equal contribution) (BK21+ IF:4, Acceptance rate: 26.6%)
Heewon Park*, Mugon Joe*, Miru Kim*, Kyungjin Im, Minhae Kwon, “Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift,” IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2026. (*equal contribution) (Highlight Paper- Top 10% of accepted papers) (BK21+ IF:4)
Kyungjin Im, Somin Im, Minhae Kwon, Anonymous submission – In Context Learning, Top AI Conference, under review.
Heewon Park*, Kyungjin Im*, Minhae Kwon, Anonymous submission –VLA merging, Top AI Conference, under review.
Kyungjin Im, Heewon Park, Miru Kim, Mugon Joe, Minhae Kwon, “Personalized Federated Post-training for Real-time Adaptation to Label Distribution Shifts,” Journal of Korean Institute of Communications and Information Sciences (J-KICS), vol.50, no.12, pp.1811-1821, Dec 2025. (Best Paper Award)
Jiyun lee, Yunseo Heo, Kyungjin Im, Jiwon Seo, Iksu Kim, “Implementation of a Real-Time Pothole Detection and Depth Estimation System Using YOLOv5 and MiDaS with User Alert Functionality,” Journal of the Korea Contents Association, vol.25, no.11, pp.1-12, Nov 2025.
Kyungjin Im, Heewon Park, Miru Kim, Mugon Joe, Minhae Kwon, “FIM-based Personalized Federated Post-training Solution Robust to Data Distribution Shift,” Korea Computer Congress (KCC), July 2025. (Best Paper Award) [Video]
Kyungjin Im, Heewon Park, Miru Kim, Minhae Kwon, “Federated Post-training for Personalized Client Models,” the Joint Conference on Communications and Information (JCCI), July 2025. [Video]
Heewon Park, Miru Kim, Kyungjin Im, Minhae Kwon, "Learning method and system for post-learning based on federated learning and a computing device for performing the same" US Patent, filed on January 12, 2026 (Application no. US 19/445,750)
Kyungjin Im, Heewon Park, Miru Kim, Minhae Kwon, "Learning method and system for post-learning based on federated learning and a computing device for performing the same" Korean Patent, filed on September 23, 2025 (Application no. 10-2025-0137594)
Kyungjin Im, Heewon Park, Miru Kim, Minhae Kwon, "Personalized Federated Learning System with Parameter Alignment and Contribution-weighted Model Update," Korea Copyright Commission, Sep 5, 2025. (Registration no. C-2025-035247)
Kyungjin Im, Heewon Park, Miru Kim, Minhae Kwon, "Federated Learning System Utilizing a Pretrained Model Trained in an Imbalanced Data Label Distribution Environment," Korea Copyright Commission, Aug 12, 2025. (Registration no. C-2025-030852)
Kyungjin Im, Heewon Park, Miru Kim, Minhae Kwon, "Personalized Federated Learning System Considering Data Bias with Weight Adjustment," Korea Copyright Commission, Aug 12, 2025. (Registration no. C-2025-030850)
Kyungjin Im, Heewon Park, Miru Kim, Minhae Kwon, "Decentralized Training System with Local Data," Korea Copyright Commission, Aug 12, 2025. (Registration no. C-2025-030851)
Kyungjin Im, Heewon Park, Miru Kim, Minhae Kwon, "Robust Personalized Federated Post-Training System for Imbalanced Data Environments," Korea Copyright Commission, Aug 12, 2025. (Registration no. C-2025-030853)
Best Paper Award, J-KICS (2025)
The Journal of Korean Institute of Communications and Information Sciences
Best Paper Award, KCC (2025)
Korea Computer Congress Conference
Best Paper Award, Soongsil University (2025)
Graduation Thesis Presentation
Excellence Award, Soongsil University (2025)
Soongsil Patent Competition
Excellence Award, National Program of Excellence in Software (2025)
SW-Centered University Essay Contest
National Science and Engineering Excellence Scholarship (2022)
Korea Student Aid Foundation (KOSAF)