Jin Yeob Chung
About Me
As of February 2021, I am delighted to have joined Akros Technologies as one of the co-founding members of the company. We develop financial technologies for scalable and data-driven strategy for exceptional alpha. The kind of research that we are conducting at Akros is available at our tech blog.
Previously, I was a master's student reading Information Engineering at the University of Oxford. I am fortunate to be advised by Professor Paul Newman, Dr. Daniele De Martini, and Dr. Matthew Gadd at the Mobile Robotics Group on my master's thesis Multi-modal Vehicle and Pedestrian Detection.
My research interests are motivated by understanding how an intelligent system perceives the world and such system makes intelligible decisions. The considerations that I seek to address are:
(i) a consideration of means to efficiently distill features from streams of data
(ii) a consideration of multi-modal analysis and how to accurately embed different modes of information.
(iii) a consideration of the ability of an intelligent to interact with the environment and make intelligible decisions
Research Projects
Multi-Modal Vehicle and Pedestrian Detection (2021)
MEng Thesis
We introduce a framework that uses information from multiple modalities to perform pedestrian and vehicle detection. We use the quantified uncertainty from an image semantic segmentation model and a LiDAR semantic segmentation each respectively to give a more accurate, combined set of segmentation results.
PDF : Coming Soon | Code : Coming Soon
System Design for Mountain Search and Rescue (2020)
3rd Year Project #1
We provide a solution for the search and rescue (SAR) industry by proposing a system using autonomous UAVs designed to assist in high altitude SAR missions. My responsibility was to design an imaging system capable of detecting and tracking a human casualty close to real-time.
Jin Yeob Chung
PDF | Code : Coming Soon
Formal Verification of Neural Networks (2020)
3rd Year Project #2
We propose an algorithmic framework called branch and bound for the task of verification, one that provides sufficient lower and upper bounds on the outputs of a toy neural network. Novel branching strategy such as bisection in accordance with the highest gradient interval and
Jin Yeob Chung
Spherical Trigonometry (2013)
Tomline Essay Competition Runner-up
In this project, I introduce various concepts that exemplify what we call Spherical Trigonometry. From the basic axioms, I move onto the applications of Spherical Trigonometry on right-angled and oblique-angled triangles.
Jin Yeob Chung
PDF : Coming Soon
Work Experience
Integration of SiamRPN++ (2020)
OpenCV Student Developer
Funded by the Google Summer of Codes program, I exported the implementation of the SiamRPN++ into ONNX and imported it using OpenCV's ONNX importer. In the process, the support for convolution with inconstant weights was provided to account for the depth-wise correlation layer, a feat specific to the region proposal network used in the visual tracker.