NOTE : Blog updates may occasionally not be reflected, so we recommend performing a FORCE RELOAD (Windows: Ctrl + Shift + R, Mac: Command (⌘) + Shift + R)
Topics I keep returning to across current projects, papers, and experiments.
Learning from Human Demonstrations, Vision-Language-Action model
I keep studying how human demonstrations, language, and perception can be aligned into robot policies.
Learning to generate novel data while preserving task-relevant semantics
I am participating in research utilizing diffusion models, which can be described as physics-informed techniques for nonequilibrium stochastic dynamical systems.
Cross-domain adaptation between heterogeneous visual spaces
I am interested in how representations move across domains without losing task-relevant semantics.
Nature-inspired technology
I enjoy experimenting with applying methodologies used in one task or domain to other areas.
Selected main projects and representative work.
Enhanced the tumor segmentation performance of the MedCLIP-SAM framework by integrating DPO-based optimization with diffusion-driven counterfactual image generation.
Developed an end-to-end pipeline to construct physically constrained EgoVLA training data represented through MANO hand meshes.
A seminar session that introduces EgoVLA and EgoBridge through the lens of shared ideas and representation learning.
A seminar session that introduces the Olaf robot implementation process based on reinforcement learning and human-robot interaction principles.
Experiment with DHN-NCE loss function to improve hard negative discrimination
A generative design project exploring efficient font creation for minority languages with environmental awareness.
A demand forecasting project for restaurant menus using operational data and machine learning modeling.
Community activity around AI/ML study, technical sessions, and collaborative growth through GDGoC.
Participated in the Korean translation of the PyTorch documentation and translated the Knowledge Distillation Tutorial document.
Development and deployment of Softie, a routine recommendation application for mental health improvement.
Experiments under active iteration, experimentation, and repeated attempts.
External activities, communities, and small projects.
Conducted an iOS frontend development study at UMC.