Jaedong Hwang

Master Student

Electrical and Computing Engineering

Seoul National University

: jd730snu.ac.kr

: jd730

About

I am an ECE master student at Computer Vision Lab (advisor: Prof. Bohyung Han), Seoul National University (SNU). Recently, I worked with as a research intern at Computer Vision Lab advised by Prof. Bohyung Han. Before joining Computer Vision Lab, I was a research intern at T-Brain and a visiting student at CLVR Lab at the University of Southern California advised by Prof. Joseph Lim. I also spent great time with Prof. Byoung-Tak Zhang at BioIntelligence Lab, SNU before enrollment.

My research interests come from a question, "why people can learn from imperfect data such as small, noisy or weak-labeled data and why machine cannot?" I am mainly interested in cognitive science (curiosity, procedural memory), computer vision (action localization, weakly-supervised instance segmentation) and reinforcement learning (exploration strategies) for applying cognitive science knowledge to machine learning.

News

  • [09/2019] I joined Computer Vision Lab @ Seoul National University as a master student advised by Professor Bohyung Han
  • [10/2018] One paper "Robust harmonic field based tooth segmentation in real-life noisy scanned mesh" is accepted to SPIE Medical Imaging (@San Diego, Feb 2019)
  • [09/2018] I joined Computer Vision Lab @ Seoul National University as a research intern advised by Professor Bohyung Han
  • [05/2018] I joined T-Brain @ SKT, Seoul, Korea, as a research intern
  • [02/2018] I joined CLVR @ University of Southern California as a visiting student advised by Professor Joseph Lim

Education

Seoul National University

 Master Student in ECE (Advisor: Bohyung Han)
Sep 2019 - Present

Seoul National University

 B.S. in Computer Science and Engineering (minor: Brain-Mind-Behavior)
(Summa Cum Laude)
Mar 2013 - Aug 2019
(Apr 2015- Jan 2017 : Military Service)

Academic Experience

Computer Vision Lab, Seoul National University

 Research Intern, Advisor : Bohyung Han

Developed an end-to-end weakly-supervised instance segmentation framework via object detection.

Sep 2018 - Aug 2019

CLVR Lab, University of Southern California

 Visitng Student, Advisor : Joseph Lim

Developed a demonstration quality assement framework via Meta-Learning.

Mar 2018 - May 2018

BioIntelligence Lab, Seoul National University

 Research Intern, Advisor : Byoung-Tak Zhang

Developed a discourse system acting as a cashier using Hidden Markov Model.

Jul 2014 - Apr 2015

Industrial Experience

T-Brain, SKT, Korea

 Research Intern
Developed a demonstration suitability assessment framework for imitation learning leveraging Model-Agnostic Meta-Learning (MAML).
Jun 2018 - Aug 2018

Cybercrime Investigation Unit, Seoul Metropolitan Police Agency

 Investigator (Auxiliary Police) - as a substitute of military service
Investigated and arrested criminals about illegal cyber gambling (approximatly $2.5 billion), malware, hacking, and illegal explosive maker.
Aug 2015 - Jan 2017

Publications

Robust harmonic field based tooth segmentation in real-life noisy scanned mesh
Jaedong Hwang*, Sanghyeok Park*, Seokjin Lee*, Yeong-Gil Shin.
SPIE 10949, Medical Imaging 2019: Image Processing. [paper], [poster]

An adaptive computational discourse system based on data-driven learning algorithm
 Seungwon Lee*, Jaedong Hwang*, Eunsol Kim, Byoung-Tak Zhang.
International Symposium on advanced Intelligent Systems (ISIS 2015) (Vol. 201). [paper]
Best Session Paper Awards

(*: equally contributed)

Awards & Honors

  • Best Paper of Brain-Mind-Behavior department
    2017
  • Seoul Metropolitan Police Agency Award Certificate
    2016
  • ISIS 2015 Best Session Paper Award
    2015
  • Jongha Fellowship
    2014

Scholarship

Seoam Scholarship Foundation (Merit-based)

A full scholarship for undergraduate studies and additional stipends of $200 each month (from 2018, stipends is increased to $350)
Sep 2013 - Aug 2019

Other Projects

Meta-Learned Suitability Assessor for Noisy Demonstration Sets

 Advisor : Joseph Lim
This project is for making a framework that selects suitable demonstration subsets in a noisy demonstraiton set including unsuitable demonstrations via meta learning. The model is able to automatically rank the suitability of given demonstrations so that it can learn from the most suitable subset. [paper], [project page]

Mar 2018 - Oct 2018

Multi Label Generative Adversarial Network

 Advisor : Byoung-Tak Zhang
This project was conducted in "Brain-Mind Behavior Individual Study(2071.416, prof. Sowon Hahn)" The model was combined Generative Adersarial Network part and traditional CNN Classifier. The data is made up of image, style labels, genre labels.

Sep 2017 - Dec 2017

Style Recognition of Artworks

This project is a final project of undergraduate course, "Computer Vision(M 1522.001000, prof. Gun Hee Kim)" from March to June. According to the project, Using SVM and PCA with featured layer of CNN is better than using Deep Neural Network only. Accuracy of style recognition is higher than human expert.

Mar 2017 - Jun 2017

Other Experience

Seoul National University Philharmonic Orchestra

 Server Administrator
Aug 2016 - Present
 Chief of department of Data Management
Oct 2014 - Mar 2015
 2nd Violinist
Mar 2013 - Aug 2019

National Museum of Modern and Contemporary Art, Seoul, Korea

 Volunteer
Mar 2017 - Oct 2017

Guahm Middle School, Seoul, Korea

 Mentor
Sep 2013 - Feb 2015

Online Coursework

Statistical Learning(Stanford Online) with Distinction
2016
Mining Massive Datasets (Coursera) with Distinction
2015
Machine Learning (Coursera)
2014