About
Hi, I am a research intern in Robotics and Computer Vision Lab (RCV) in UOS, advised by Prof. Jiyoung Jung.
Now I’m interested in robotics and I’m actively studying it (you can see it in my blog posts). In the future, I would like to research the application of computer vision, reinforcement learning and LLMs in robotics. In particular, I would like to use NVIDIA Isaac Gym for my research.
Education
B.S. in Artificial Intelligence, University of Seoul
Mar. 2021 ~ Aug. 2024 (expected)
I am currently a department topper with a 4.44/4.50 GPA [transcript]. Also, I am completely self-motivated person who loves to learn new things (I received a certificate of achievement for Particle Dynamics from KOOC, and now I am taking Rigid Body Dynamics). I have a positive attitude and enjoy debating with people who are passionate about their field of study.
Projects
Image Caption refinement via Feedback Mechanism (alias)
ECCV 2024, 2nd author
( Under Review … )Feb. 2024, RCV lab
AICOSS Hackathon Competition
We participated in a competition to develop an AI model that performs Multi-Label Classification based on satellite images containing various geographic and environmental features. We tried various models and losses, and we could get the 1st prize in the competition. And also I could get skills in multi-GPU training. [pdf] [code]
Jan. 2024, University of Seoul President’s Award
Proxy initialization for Deep Metric Learning
In deep metric learning, it is important to minimize intra-class variation while maximizing inter-class variation. There are many methods to deal with this problem, but I concentrated on initializing the proxy for each class. I found some interesting parts of initialization (e.g. initialize proxy for the mean of their classes), but unfortunately it couldn’t beat the SOTA method.
Jun. 2023, MIL lab
JMT : Korean Typo Correction based on JaMo-unit Tokenization
Unlike English, Korean characters are composed of 2-3 letters. So there can be many more combinations of typos in Korean than in English, and we cannot store all the typo words in our model’s vocabulary. So, we used “Jamo” unit tokenization. We used transformer for our model, and the performance was better than other tokenizations. [pdf]
Dec. 2022, NLP Final Project
Generate abbreviations to memorize words
In childhood, when we had to memorize some words for the exam, we would make a special word with the first character of the words. For example, when we have to memorize ‘부산’, ‘서귀포’ and ‘진천’, we can make a word ‘부서진’ for an abbreviation. We thought that it’s not so easy to make abbreviations when the given words get bigger, so we trained a model to make abbreviations for given words. [pdf]
Oct. 2022, deep daiv. 1st On-board Conference
Korea Expressway Corporation Complaint Classification
There were so many complaints for the company, so a lot of time was used for categorizing the region department. So we used deep learning to classify the region of the complaints. [pdf]
Aug. 2022, AICOSS Director’s Award
Awards
University of Seoul President’s Award for AICOSS Hackathon Competition
Jan. 2024
Grand Award for 2023 AI Department Programming Competition
Dec. 2023
Grand Award for 2022 AI Department Programming Competition
Feb. 2023
AICOSS Director’s Award for Hands-on Deep Learning Summer Bootcamp Competition
Aug. 2022
Scholarship
Academic Excellence Scholarship for 4 semesters
Spring 2024, Spring 2023, Fall 2022, Spring 2022
Seoul Mayor’s Scholarship for 1 semester
Fall 2023
University Business Scholarship for 5 semesters
Fall 2023, Spring 2023, Fall 2022, Fall 2021, Spring 2021
Tutor scholarship for 2 semester
Spring 2023, Spring 2022