I am a fullstack developer at Zalando. This is my personal website that I bult to store my personal projects.
During my leisure time, I learn various technologies that intersetd me and create small web app widgets that's gonna make my life a little bit easier.
The cell detector is based on GP regression model. The input of the program is histopathology image from bone marrow, which normally contains 300~400 cells, and the output is the predicted cell centers. The green dots represent correct prediction of the cell centers. The blue and red dots each represent misprediction and additional prediction.
The program achieve an average 92% accuracy in cell center prediction and the computational time for an image size of 1200x1200 is less that 5 mins.
The program is a result of collaboration. I designed the main algorithm for cell prediction based on GP regression. This approach decrease the complexity of implementaion significantly comparing to other common cell detection algorithm, but still be able to achieve a high accuracy in prediction.
MATLAB
This project utilize the open city data from Health Department of New York city, where records the sanitation inspection results from a total number of 24,000 restaurants.
From the inspection results, we were able to the patterns such as: ranking of the sanitary condition according to the type of cuisine, or which sanitary rule being violated the most. These patterns were hidden in the raw data if were not being organized and visualized.
Further, we combined the ZIP code of the restaurant with GPS coordinate to predict the sanitary condition for regions that without inspection results.
Python, MATLAB, HTML5 , Javascript, D3
The goal of the project was to create a scene containing a movable light source, shining light onto some geometric objects and their shadow, seen in figure. The main challenges encountered during the programming where the movement of the light, creating a light radius and creating dynamic shadow.
The website is a result of callaboration wtih Paul Stocker. I am responsible for the mouse interaction that allow user to move the light source as well as part of the coding in dynamic shadow.
HTML5 , Javascript, WebGL
The goal of this project was to develope an algorithm that recognize a certain object in a static scene. I adopted SIFT feature, a high dimension vector, to achieve high recognition accuracy.
In training stage, a traning image's keypoints detected by SIFT algorithm were stored ( top plot, a mobile phone ). In testing stage, all the keypoints in the static scene compared the similarity (Euclidean distance) with the keypoints in training image. Finally, the four most similar keypoints were selected, and were used by finding the coordinate transformation relation with the keypoint in training image ( bottom plot ).
MATLAB, VLfeat
We designed a classifier that was able to recognize hand-written digits from MNIST database.
Based on principle components of vectorized image of hand-written digits or spatial distribution of pixels as features, we trained the 3-layer feedforward neural network and using the back propagation algorithm to adjust the weight. We further examined the accuracy of using different feature sets. The classifier achieved a lowest 6.32% error rate.
I played a significant role in devising the algorithm of a 3-layer neural network that was the backbone of the project.
MATLAB
Exchange Student, Applied Mathematics and Computer Science
August 2015 — June 2016
Denmark Technical University
Copenhagen, Denmark
M.Sc. Electro-Optical Engineering
June 2013 — July 2015
National Chiao Tung University
Hsinchu, Taiwan
Thesis:
Separating iris texture and cosmetic contact lens by using independent component analysis
B.Sc. Electro-Optical Engineering
June 2009 — June 2013
National Taipei Univeristy of Technology
Taipei, Taiwan
| Category | Proficiency in approximate descending order from left to right |
|---|---|
|
Programming Languages |
Python, Javascript, C++ |
| Web Technologies | HTML5, CSS, D3, WebGL |
| Data Analysis/Visualization |
MATLAB, pandas, numpy, scikit-learn |
| Image Processing |
MATLAB, openCV |
| Languages | Mandarin, English, Japenese |
Spring 2016
Term project of DTU course "Advance Machine Learning". Devised the cell detection algorithm based on GP regression. Conducted different machine learning algorithm, such as feature extraction, recognition and clustering in non-trivial engineering task.
[MATLAB]
Term project of DTU course "Social Data Analysis and Visualization". Designed a webpage explaining the sanitation condition of restuarants in New York city. Examined over 24,000 restuarants data, and assessed importance of the variables based on statistic properties. Conducted machine learing algorithm, GP regression and K-mean clusting, to combine geometric information with the original data; futher, created sanitation prediction based on geomatric location.
[Python, numpy, sckit-learn, Javascript, HTML5, CSS, D3]
Fall 2015
Term project of DTU course "Computer Graphic". Created an interactable webpage for 3D scene. Achieved the effect of a point light source shinning on objects with self-shadow effect using cube map. Created cubemap by drawing the scene on framebuffer according to perspective of the light souce.
[Javascript, WebGL, HTML5, CSS]
Fall 2014
Term project of NCTU course "Computer Vision". Designed a program able to recognize objects based on their SIFT feature. Using the ratio of 1st and 2nd nearest neighbor to decrease the computational time of finding certain object in an unseen scene. Based on RANSAC approach to find the coordinate transformation from training image to testing image.
[MATLAB, VLfeat]
Spring 2013
Term project of NCTU course "Machine Learning". Devised a classifier capable of recognizing hand written digits. Evaluated performance of digit classifier when applying different features, principle components or spatial distribution of pixels.
[MATLAB]
2013 — 2015
place holder [place holder]
