Profile | JaeHyun Lim

Transcription

Profile | JaeHyun Lim
JAE HYUN LIM
[email protected]
[email protected]
Contact
Address
Positioning/Navigation Technology Research Section
Electronics and Telecommunications Research Institute (ETRI), 218 Gajeongno, Yueseong-gu,
Daejeon, Korea 305-700
Tel +82 42 860 5415
Mobile +82 10 8225 1401
Homepage
http://jhlim.com (personal)
http://kr.linkedin.com/in/jaehyunlim (linkedin)
EDUCATION
Korea Advanced Institute of Science and Technology
M.S. Candidate, Electrical Engineering
2012.Feb – 2014.Feb
Advisor: prof. Dae-Shik Kim
Daejeon, Korea
Korea Advanced Institute of Science and Technology
Bachelor of Science, Bio-Brain Engineering, Magna Cum Laude
2008.Feb - 2012.Feb
Daejeon, Korea
Korea Science Academy (high school)
2005.Mar - 2007.Feb
Busan, Korea
RESEARCH
INTERESTS
• Deep Learning
• Representation learning and probabilistic generative models for vision
• Machine learning
• Computer vision
• Bayesian methods and Bayesian brain
• Computational neuroscience
RESEARCH
PROJECTS /
WORKING
EXPERIENCE
Researcher at ETRI
2014.Aug-current
I started a researcher position at Electronics and Telecommunications Research Institute
(ETRI), a national research center in Korea. I’m participating in a project to develop indoor
navigation and human pose estimation systems with multimodal sensors.
Kaggle National Data Science Bowl competition
I participated in a data science competition, called Kaggle National Data Science Bowl
competition, classifying images of plankton. I finished 22nd position out of 1049 teams with
small difference from the winning solution.
My approach and codes: https://github.com/lim0606/ndsb
Competition site: https://www.kaggle.com/c/datasciencebowl
2015.Jan2015.Mar
Kaggle Dogs vs Cats competition
I join a team to implement deep learning algorithms for large scale image classification.
During the project I and my colleague just ranked 6th in Dogs vs Cats kaggle competition,
using deep convolutional neural network. We obtained 6th place against 215 other teams
with 98.1% accuracy.
http://www.kaggle.com/c/dogs-vs-cats/leaderboard
2014.Jan2014.Feb
DARPA Robotics Challenge, Team KAIST, Vision group leader (Advisor: Prof. DaeShik Kim)
As part of a team of team KAIST, which consists of several laboratories in KAIST and
participates in DARPA Robotics Challenge in Dec. 2013, I am leading the group of people
dedicating to design vision system for our robot (visit homepage).
SLAM via Structure from Motion
Sparse coding based 2d image tracking
Stereo vision with dense SIFT matching
Unsupervised 2d Image segmentation via deep learning and Bayesian nonparametric
Project page: http://drc.kaist.ac.kr
2012-2014.Jan
Multiple kernel learning with hierarchical feature Representations (Advisor: Prof.
Dae-Shik Kim) [1]
In this project, we suggest multiple kernel learning (MKL) with hierarchical feature
representations. Using MKL, we tried to find the best combination of different levels of
features learned by deep Boltzmann nets.
2013.Jan.2013.Oct
Learning spatio-temporally invariant representations from video (Advisor: Prof. DaeShik Kim) [2]
In this study, I designed an algorithm to discover individual identities from a video clip that
shows faces in motion displayed in random order. The idea behind this stems from the
possibility that ambiguity of different objects in data space can be elucidated by temporal
relationships corresponding to how data appear to the system.
2011.June–
2012.July
Predictive coding strategies for developmental neurorobotics (Advisor: Prof. DaeShik Kim) [3]
In this work, we conducted neurorobotics experiments to illustrate that even minimalist
prediction error-based strategies replicate key features observed during the emergence of
infant learning behaviors, such as action sequence generation, object permanence, and
imitation.
2011. Sept –
2012.May
Active learning using entropy measure based on Gaussian Process classifier
(Advisor: Prof. Soo-Young Lee)
The research suggested explicit probabilistic measures of uncertainty model, mainly in the
context of Gaussian process classification, including measure of entropy and expectation of
posterior distribution with Markov Chain Monte Carlo approximation. (pdf)
2011.Feb–
2011.Dec
17th LG Global Challenger (Advisor: Prof. Doheon Lee)
“LG Global Challenger” is the first and major overseas expedition program for
undergraduates with the competitive rate of 21:1. My excellent results and active
participation in Prof. Daeheon Lee's Bio Data Engineering course led me to be selected
amongst a group of students for visiting world leading institutes. I had the opportunity to visit
Microsoft's health group division in Seattle, Stanford University, patientlikeme.com's
headquarters, Google research center as well as Harvard Medical School. (homepage,
report)
2011
Self-segmentation (Advisor: Prof. Dae-Shik Kim) [4]
In this work, I conducted neurorobotics experiments to demonstrate that prediction errorbased strategies replicate key features observed during the emergence of infant learning
behavior, e.g. self-awareness behavior.
2010.Sept –
2011.June
Internship program at Keio University (Advisor: Prof. Yasuyo Minagawa)
The internship focused on research topics related to language developmental processes of
human infants and language ability of human. It let me know how to deal with NIRS,
functional MRI and other imaging tools for studying cognitive neuroscience.
2010.Jan.06 –
2010.Feb.07
A study on the role of thalamus in language processing with computational analysis,
2010 Winter/Summer URP Program (Advisor: Prof. Soo-Young Lee)
This study was to validate cortico-thalamo-cortico systems as a hypothesis of human
language processing with computation models cognitive experiments with functional
magnetic resonance imaging (fMRI). (pdf)
2009.Oct–
2010.May
KAIST Formula SAE Team Marketing Team Captain
As part of a team of academic formula racing car research group called K-alpha, designed
and tested our car to participate in the Formula SAE program. As a marketing manager,
2008-2009
I was in charge of arranging meetings with colleges and related institutions, making
promotion materials, advertising, and as well as accounting. (potfolio)
PUBLICATIONS
[1] J. Lee, J. H. Lim, H. Choi, and D.-S. Kim, “Multiple Kernel Learning with Hierarchical Feature
Representations,” in Neural Information Processing SE - 64, vol. 8228, M. Lee, A. Hirose,
Z.-G. Hou, and R. Kil, Eds. Springer Berlin Heidelberg, 2013, pp. 517–524. (pdf)
[2] J. H. Lim, H. Choi, J.-C. Park, J. Y. Jun, and D. Kim, “Learning spatio-temporally invariant
representations from video,” Neural Networks (IJCNN), The 2012 International Joint
Conference on. pp. 1–6, 2012. (pdf)
(IJCNN was be held jointly with the FUZZ-IEEE and the IEEE CEC as part of the 2012 IEEE World
Congress on Computational Intelligence (IEEE WCCI), June 10-15, 2012, Brisbane)
[3] J.-C. Park†, J. H. Lim†, H. Choi, and D.-S. Kim, “Predictive coding strategies for
developmental neurorobotics.,” Front. Psychol., vol. 3, p. 134, 2012. (†J.-C. Park and J. H.
Lim have contributed equally to this work.) (pdf)
[4] J. H. Lim, J. H. Yoo, S.-Y. Lee, D.-S. Kim, “Self-segmentation Based on Predictability
Measure in Multimodal Autonomous System”, accepted for poster session in Neural
Networks (IJCNN), The 2012 International Joint Conference on, San Jose, California (United
States), July 31-5 2011 (pdf)
TEACHING
EXPERIENCE
EE538 Introduction to Brain IT (Graduate course)
This course discussed the key differences in architecture and algorithms between
conventional information processing systems (e.g. von Neumann machines) and biological
brains. As a teaching assistant I provided general background knowledge about brain-like
information processing system, from single cell level neuron models to probabilistic neural
networks and to state-of the art machine learning technologies.
Assistant Teacher, International Students Science Fair 2006
The International Students Science Fair was initiated for the purpose of facilitating
international exchange and interaction in science education, based on science project
presentations delivered by high school students from all over the world.
Designed the class named “Introduction to robot soccer for young scientist” as an official
extracurricular activity of the fair.
SKILLS
MATLAB programming,
C/C++ Programming
Cuda programming in C/C++
Basic level web programming and database management
Java Programming
HONORS AND
AWARDS
National Research Fund Scholarship in KAIST
KAIST
Magna Cum Laude
Korea Advanced Institute of Science and Technology
KAIST Honor Program
Academic course transition program
REFERENCES
Prof. Dae-Shik Kim ([email protected])
Prof. Soo-Young Lee ([email protected])
Prof. Doheon Lee([email protected])
Prof. Kyung Dae Kim ([email protected])
Spring 2013
Summer 2006
Feb 2012 present
Feb 2012
2010 Sept – 2013
Aug