PROJECTS
AI Fellow | Insight Data Science
Optical character recognition (OCR) is the technique that allows a computer to read static images of text, and convert them into editable, searchable data. Reading order is one of the difficult challenges in an OCR task. In order to improve the accuracy of from a scanned document, I applied the traditional image process to analysis layout and recover the reading order. This project is a consulting project with a startup company, Tolstoy.
R&D Intern | JD.com
As a R&D intern in Big Data group, I am responsible for researching several deep learning frameworks for mobile phone, including Caffe2 (Facebook), NCNN (Tencent) and MACE (Xiami). Our goal is trying to build a human segmentation deep learning networks and apply it to a mobile application.
Course | Purdue Univ.
This is the final project of the course "CS 57300 Data Mining" at Purdue University. In various areas of chemical analyses, toxicity prediction is one of the important research topics. Given several features of a molecule, our goal is using those features to predict whether a molecule is poisonous or not.
GIS Engineer | Garmin
This project can automatically extract Point of Interest(POI), such as a restaurant, from a Chinese blog article. The main challenge in this task is called Name Entity Recognition, which is a famous issue in Natural Language Processing. This project proposed a model based on Conditional Random Fields to solve NER problem. By combining the information from Google Map API and Facebook Graph API, the overall accuracy could be 97%.
Master Dissertation | National Tsing Hua Univ.
This thesis proposes a new indoor localization technique that fuses pedestrian dead reckoning with encounter information. This work builds on CRFs for the lower computational complexity than the traditional methods, such as particle filtering. Experimental results show that we further lower the complexity and expand location coverage.
Software Engineer | ITRI
This work describes an intelligent insulating shipping container (IISC) and the supporting backend for cold-chain logistics. Each IISC can monitor the interior temperature of the container during shipping of temperature-sensitive items and alerts the deliverer wirelessly via a Bluetooth 4.0 Low Energy connection on their smart mobile without infrastructure or dongle.
EPLab | National Tsing Hua Univ.
MobiRing is a compact, finger-worn wireless motion-tracking platform. It has been prototyped with an ultra-compact wireless sensing platform with an on-board triaxial accelerometer and a short-range wireless transceiver with an optional second accelerometer for enhanced accuracy.
EPLab | National Tsing Hua Univ.
BuddyGuard is a patient-care assistance system that can alert and documentation of body rotation for pressure sore and fall prevention. The system is built on TI CC2540 with several modules, such as a wireless charger. We also provide an iOS application with a user interface to display and record a patient's postures.
Course | National Tsing Hua Univ.
This app allow users to create their own AR business cards. Also, they can watch others‘ virtual content on their business cards through the app. While using the app to watch others’ business cards, users can add the card owner into their in-app card holder. In addition, users can also add annotation or classify the card they received in an AR way through the app.
Course | National Tsing Hua Univ.
AllWake allows users to create a group alarm for an event and invite their friend who is going to participate in the same event to join. Friends in the same group alarm can check other member‘s alarm setting and whether the alarm has been ignored or turned off in the next day morning.
Course | National Tsing Hua Univ.
This Android game is written by Scala, which is a functional programming language. Combine two famous games, Pokemon and Puzzle $ Dragon. We implemented the puzzle algorithm by ourselves and design the story of the game.