Hyunseo Kim

I am a PhD student in Interdisciplinary program in Neuroscience at Seoul National University, and a member of the biointelligence laboratory led by Byoung-Tak Zhang.

My research interests lie in the area of robotics, vision, imitation learning and representation learning.

Email  /  CV  /  Google Scholar  /  Github

I'm interested in computer vision, robotics, self-supervised learning, imitation learning, and representation learning. Much of my research is about inferring the object representation from related actions. Representative papers are highlighted.

EXOT is applied to the robot hand camera (wrist camera) and successfully detects the target object absence during object manipulation. EXOT is a single object tracker with an out-of-distribution classifier. It makes safe robotic manipulation possible even when the target moves.

MD-AIRL predicts a sequence of reward functions, which are iterative solutions for a constrained convex problem.

MPART suggests successful active online learning that selects representative queries and proceeds efficient model update that does not forget important info as soon as a new data sample is observed.

LPART suggests semi-supervised online learning for real-world problems where labels are rarely given and the opportunity to access the same data is limited.

We revealed a neural circuit that relay mechanosensory feedback from the digestive tract to the brain. Neurons in parabrachial nucleus that express the prodynorphin gene monitor the intake of both fluids and solids.