Hi! I am Jaeyong Ko, an M.S. student in Industrial Engineering at Seoul National University, advised by Prof. Pilsung Kang in the Data Science & Business Analytics Lab.
My research focuses on reasoning in LLMs, VLMs, and unified models. Previously, I worked on LLM mathematical reasoning and financial-domain benchmarks. I am currently exploring visual latent reasoning, with the goal of bringing visual thinking from a unified model's generation branch into its understanding branch. Ultimately, I aim to develop unified reasoning models that surpass the combined capabilities of separate, specialized models.
http://".
However, if the current repository name is intended, you can ignore this message by removing "{% include widgets/debug_repo_name.html %}" in index.html.
baseurl ("_config.yml.
baseurl in _config.yml to "
Jaeyong Ko*, Hankyeol Kim*, Pilsung Kang (* equal contribution)
LVR @ WACV 2027 Workshop
A unified multimodal model framework that replaces computationally expensive intermediate image generation with compact latent tokens during interleaved text–visual reasoning.
Jaeyong Ko*, Hankyeol Kim*, Pilsung Kang (* equal contribution)
LVR @ WACV 2027 Workshop
A unified multimodal model framework that replaces computationally expensive intermediate image generation with compact latent tokens during interleaved text–visual reasoning.
Hangyeul Lee, Juyoung Oh, Jaeyong Ko, Sunmin Kim, Jaeik Park, Hyunkyu Kim, Jungmin Son, Pilsung Kang
ACM International Conference on AI in Finance (ICAIF 2026)
A benchmark for exhaustive life-event history extraction and financial-state reconstruction from longitudinal banking dialogues.
Hangyeul Lee, Juyoung Oh, Jaeyong Ko, Sunmin Kim, Jaeik Park, Hyunkyu Kim, Jungmin Son, Pilsung Kang
ACM International Conference on AI in Finance (ICAIF 2026)
A benchmark for exhaustive life-event history extraction and financial-state reconstruction from longitudinal banking dialogues.
Jaeyong Ko, Jinu Lee, Pilsung Kang, Yukyung Lee
arXiv preprint arXiv:2606.25524 2026
Identifies single-token triggers that can sharply change the success or failure of LLM mathematical reasoning, supporting fine-grained analysis of reasoning reliability.
Jaeyong Ko, Jinu Lee, Pilsung Kang, Yukyung Lee
arXiv preprint arXiv:2606.25524 2026
Identifies single-token triggers that can sharply change the success or failure of LLM mathematical reasoning, supporting fine-grained analysis of reasoning reliability.
Jaeyong Ko, Eunseo Kim, Doehyeon Lee
NeurIPS 2026 Education Track (Accepted)
A visual lecture and hands-on lab compare text-only, visual-only, and interleaved reasoning across four dimensions of unified multimodal models.
Jaeyong Ko, Eunseo Kim, Doehyeon Lee
NeurIPS 2026 Education Track (Accepted)
A visual lecture and hands-on lab compare text-only, visual-only, and interleaved reasoning across four dimensions of unified multimodal models.
Doehyeon Lee, Yoo-Min Jung, Jaeyong Ko
NeurIPS 2026 Education Track (Accepted)
An interactive lecture and lab explore a safety direction in vision-language models by detecting, bypassing, and restoring refusal with projection and activation addition.
Doehyeon Lee, Yoo-Min Jung, Jaeyong Ko
NeurIPS 2026 Education Track (Accepted)
An interactive lecture and lab explore a safety direction in vision-language models by detecting, bypassing, and restoring refusal with projection and activation addition.
