Jaesin Ahn
Postdoctoral Researcher, Kyungpook National University
amoeba04 [AT] gmail.com
Bio
I am a Postdoctoral Researcher at the Advanced Technology Convergence Research Institute, Kyungpook National University. I received my Ph.D. in Artificial Intelligence from Kyungpook National University, advised by Prof. Heechul Jung, and also received my B.S. (2019) and M.S. (2021) in Electronics Engineering from the same university. During my Ph.D., I worked as a graduate research assistant at Memorial Sloan Kettering Cancer Center in New York, where I developed transformer-based methods for MRI reconstruction.
My current research focuses on understanding the internal mechanisms of AI systems. I leverage mechanistic interpretability and machine unlearning to pursue AI safety and trustworthiness, addressing challenges such as unsafe content generation, privacy concerns in both general and medical domains, and reliability issues in AI systems.
Research Interests
Publications
Diagnosing LLM De-Identification Failure in Clinical Text via Mechanistic Interpretability
Under Review
Ontology-Guided Semantic Mixing with Adaptive LoRA Experts for Long-Tailed Remote Sensing Recognition
Under Review
When Metadata Matters: Impact of Recording and Demographic Factors on Respiratory Sound Classification
Under Review
LEAF: A Lightweight Language-Enhanced Model for Forestry Analysis in Remote Sensing Imagery
Sanjar Karshiev, Faisal Saeed, Jaesin Ahn, Abdul Rehman, Muhammad Diyan, Shrooq Alsenan, Heechul Jung
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2026
Mitigating Sexual Content Generation via Embedding Distortion in Text-conditioned Diffusion Models
Jaesin Ahn, Heechul Jung
NeurIPS, 2025
A Self-Attention Classifier Head for Improved Image Classification and Interpretability of ViT
Jaesin Ahn, Heechul Jung
Electronics Letters, vol. 61, no. 1, 2025
GDoT: A Gated Dual Domain Transformer for Enhanced MRI Off-Resonance Correction
Jaesin Ahn, Heechul Jung
Neurocomputing, 129918, 2025
Redesigning Embedding Layers for Queries, Keys, and Values in Cross-Covariance Image Transformers
Jaesin Ahn, Jiuk Hong, Jeongwoo Ju, Heechul Jung
Mathematics, vol. 11, no. 8, p. 1933, 2023
Skip-StyleGAN: Skip-Connected Generative Adversarial Networks for Generating 3D Rendered Image of Hand Bone Complex
Jaesin Ahn, Hyun-Joo Lee, Inchul Choi, Minho Lee
MICCAI, 2020
Siamese U-Net with Healthy Template for Accurate Segmentation of Intracranial Hemorrhage
Doyoung Kwon, Jaesin Ahn, Jaeil Kim, Inchul Choi, Sungmoon Jeong, Young-Sup Lee, Jaechan Park, Minho Lee
MICCAI, 2019
Patents
Mitigating Sexual Content Generation via Embedding Distortion in Text-conditioned Diffusion Models
Jaesin Ahn, Heechul Jung
NeurIPS, 2025
Skip-StyleGAN: Skip-Connected Generative Adversarial Networks for Generating 3D Rendered Image of Hand Bone Complex
Jaesin Ahn, Hyun-Joo Lee, Inchul Choi, Minho Lee
MICCAI, 2020
Siamese U-Net with Healthy Template for Accurate Segmentation of Intracranial Hemorrhage
Doyoung Kwon, Jaesin Ahn, Jaeil Kim, Inchul Choi, Sungmoon Jeong, Young-Sup Lee, Jaechan Park, Minho Lee
MICCAI, 2019
LEAF: A Lightweight Language-Enhanced Model for Forestry Analysis in Remote Sensing Imagery
Sanjar Karshiev, Faisal Saeed, Jaesin Ahn, Abdul Rehman, Muhammad Diyan, Shrooq Alsenan, Heechul Jung
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2026
A Self-Attention Classifier Head for Improved Image Classification and Interpretability of ViT
Jaesin Ahn, Heechul Jung
Electronics Letters, vol. 61, no. 1, 2025
GDoT: A Gated Dual Domain Transformer for Enhanced MRI Off-Resonance Correction
Jaesin Ahn, Heechul Jung
Neurocomputing, 129918, 2025
Redesigning Embedding Layers for Queries, Keys, and Values in Cross-Covariance Image Transformers
Jaesin Ahn, Jiuk Hong, Jeongwoo Ju, Heechul Jung
Mathematics, vol. 11, no. 8, p. 1933, 2023
Diagnosing LLM De-Identification Failure in Clinical Text via Mechanistic Interpretability
Under Review
Ontology-Guided Semantic Mixing with Adaptive LoRA Experts for Long-Tailed Remote Sensing Recognition
Under Review
When Metadata Matters: Impact of Recording and Demographic Factors on Respiratory Sound Classification
Under Review
Challenges & Awards
Hosted by Ministry of Science and ICT (MSIT), among 630 teams
Organized by Google DeepMind, among 1,121 teams
Experience
Graduate Research Assistant
Memorial Sloan Kettering Cancer Center, New York, USA
Apr. 2022 – Mar. 2023
- Supported by IITP grant: Research on Unsupervised Domain Adaptation Technology Based on Deep Learning
- Developed a transformer-based correction network to mitigate off-resonance artifacts caused by accelerated MRI scanning with non-Cartesian sparse trajectories
- This research led to the publication of the "GDoT" paper in Neurocomputing
Projects
Developing the Next-Generation General AI with Reliability, Ethics, and Adaptability
- Designing a reliable AGI architecture by analyzing stability issues in multi-agent interactions and exploring model routing strategies for efficient inference
- Enhancing model capabilities and adaptability across diverse specialized domains, including medical and mathematical reasoning
Evaluation of Privacy Risk Mitigation Technologies for Generative AI
- Analyzed privacy leakage risks specifically in Korean Large Language Models (LLMs) and evaluated the effectiveness of mitigation strategies
- Verified the performance of privacy-preserving techniques, including deduplication, filtering, differential privacy, and machine unlearning