Hidden Markov Model-based Cursive Handwriting Recognition


Multi-instance Boosting Tracker


Machine Learning Course Projects

Census data classification by logistic regression (report), dividing English words into syllables using Conditonal Random Field (report), document clustering using LDA (report), and sentence sentiment discovery using recursive auto-encoder (report).

Cross-lingual Synthesis of Wikipedia Articles

From summer 2009, I began a research internship at HKUST Human Language Technology Center on cross-lingual Wikipedia articles synthesis. I proposed a unsupervised approach that automatically generates Wikipedia articles for language versions with lower quality, using another well-written version of the same subject as a content template and using text segments in the target language retrieved from the Internet as building blocks. I further developed this idea in my Final Year Thesis (poster) and developed a text segment selection algorithm which features a scoring function that iteratively evaluates the goodness of candidate text segments. We generated the Chinese version of 70 Wikipedia articles based on their English counterparts, and subjective evaluation results show that the synthesized articles are generally richer in content than existing human-authored Chinese versions and are more readable than machine-translated texts. This system is described in the paper "Unsupervised Synthesis of Multilingual Wikipedia Articles" which was accepted by COLING 2010 for oral presentation (slides).

3D Laser Scanner Car

During sophomore year, I designed an autonomous mobile platform mounted with a laser beam emitter and stereo cameras that reconstructs and displays the 3D model of objects along the way in real-time. I developed the module that computes point clouds from stereo images of the laser beam and displays them in 3D with OpenGL. I also devised a dual-core pipeline that interleaves the frame buffer capture and data processing, which boosted the CPU utilization by 40%. This project was awarded a third prize at the Intel Cup embedded system design competition in Shanghai, 2008, among over 200 teams from the Asia-Pacific region. (report)

Rubik's Cube Solver

In a project that turned out receiving the highest mark in the course Microprocessor Experiment, I built an 8051-based automatic Rubik’s Cube solver that integrates pattern recognition with motor control. I also devised an innovative mechanical structure that was both practical and entertaining. (report) (demo)

Smart Robots using the VEX Toolkit

As a freshman I enrolled in a robotics course that uses VEX toolkit to build robots that perform intelligent actions such as pouring beer and following lines. In the final game which aimed at collecting point-carrying blocks, I built a wheeled robot that integrated various sensors and was able to change strategy in response to rivalries’ moves and current situation on the field.

Task 1 : Barrier Dodger (Trial 1, Trial 2, Trial 3)
Task 2 : Save the Internet! Save the world! (Demo)
Final competition : Block Grabber (Round 1, Round 2, Round 3, Round 4)