Jeongbin Sohn

Cumulative GPA: 4.38/4.5

• Developed SIGNN++, a graph neural network-based star identification framework robust to terrestrial noise such as light pollution, clouds, and motion blur.
• Achieved a 98.5% solved rate on real night-sky images, compared with 79.1% for the strongest existing method.
• Built a real-world evaluation setup using a Rocket Lab ST-RT star tracker as the attitude reference.
• Adapted star identification algorithms to lunar crater identification in collaboration with the Korea Institute of Energy Technology (KENTECH).

Analyzed recorded videos of individuals with C5–C7 spinal cord injuries using the dorsal grasper and estimated its 3D pose with the PnP algorithm.

Studied the Robot Operating System (ROS) and path-planning algorithms.

Ported existing code to Python and implemented a domain-change algorithm for indoor localization.

Led training sessions and social events, fostering teamwork among club members.

Founding member of ELITE (English for Learning, Inquiry, Thought, and Expression), an English study club supervised by Prof. Ellis Lee; planned club activities and led group studies.

Member of the Korean-American Scientists & Engineers Association (KSEA) at UC Berkeley; participated in the academic journal club.

Learned English and taught Korean with a UC Berkeley student.