Jinho Kim

- Designed and trained deep neural networks (UNet, ResNet, unrolled architectures) for solving large-scale ill-posed inverse problems under limited supervision - Built 3D segmentation models for structured object extraction in volumetric data - Implemented distributed training workflows on NVIDIA DGX systems (Kubernetes) and HPC clusters (Slurm) - Designed reproducible training pipelines using Docker-based environments - Skills: PyTorch, CUDA, Docker, Kubernetes, Slurm, HPC, experiment tracking, CNNs

- Developed deep learning-based motion correction models for dynamic image reconstruction - Applied clustering algorithms (k-means) for retrospective signal gating and data-driven preprocessing - Evaluated model robustness under motion and incomplete data constraints - Built containerized research pipelines for scalable experimentation - Skills: PyTorch, Python, Docker, HPC

- Developed a GUI-based automated design system for semiconductor layout generation - Built Python-based simulation post-processing tools for performance analysis - Reduced manual design effort via rule-based automation pipelines - Skills: Python, Ruby, GUI development, automation scripting

- Implemented classical and deep learning-based super-resolution algorithms - Optimized real-time image enhancement pipelines \item[*] Break from Nov.2015 to Jun.2017 due to study abroad in the United States - Skills: Image processing, Deep learning, Super-resolution, Image enhancement