Past Research Projects

(Course projects Not included)

1．M2M Computation Model。(Main contribution, 2007)

This model present a hierarchical data structure which is very suitable for many points set operations. And now, M2M model has successfully been applied on convex hull and path finding problem as well. Most impressive thing of M2M model is its novel properties including preprocessing sharing, high parallel processing, easily trade off between different merits. This project also wins the outstanding award in the Challenge Cup of Guangdong province recently and will present in the national Challenge Cup in October 2007. The paper about nearest neighbor searching base M2M model has been published in IMECS2007 and indexed by EI.

2． Ranking Potential Customers Based on GroupEnsemble Method (Team leader, 2007)

       The paper presents a solution for PAKDD competition 2007, which is about a cross-selling in credit card and house load. Data imbalance, value missing and time-variant distribution, which are common in the real world data, also occur in this problem. We establish a 3-layers machine learning model including Bagging, RankBoost and Expending Regression Tree, to handle this problem. More detail about the problem and competition will be http://lamda.nju.edu.cn/conf/pakdd07/dmc07/ .

3． Netflix Movie Rating Prediction (Team leader, 2007)

       We present two prediction models for the two problem “who will rate the movie” and “how many person will rate the movie” corresponding to the two tasks in KDD CUP 2007.  The main difficulties of those problems are huge amount of data and very weak relation in the given data. SVD-Decomposition, Co-Cluster, Graph-Based method, Naïve bayes method and adjusted collaborative filtering can well solve the first difficulty but the second difficulty may have to rely on other prior knowledge to improve it. More detail about the problem and competition will be http://www.cs.uic.edu/~liub/Netflix-KDD-Cup-2007.html.

4．Hand-writing Chinese Character Recognition System (Project leader,2006)

This project wins the third prize of Universities’ Software Competition of Guangdong Province 2006. On-line recognition has capacity of 400 Chinese characters and off-line 20 characters now. The recognitions both satisfy the invariants of zooming, transforming and rotating. The main aim of this system is providing a platform to test image processing algorithm and recognizing algorithm.

5．Distributed Firewall System Based on Intrusion Detection (Project leader, 2005)

       This project wins the second prize of Universities’ Software Competition of Guangdong Province 2005. The idea of this project comes from SOA and P2P service. Each agent cooperate to collect the information of the current status of network security and update the detect rule and detect components in various ways. The system combines misuse detection and anomaly detection, and supports novel concepts including dynamic action script, attack feature extraction, stateful inspection.

6．GA-ANN-GA Control Model in Space Travel (Main contribution, 2006)

This project wins the best demonstrating award of Universities’ Software Competition of Guangdong Province 2006. This project applies the genetic algorithm to train the forward neural network in order to make it become familiar with the complex situation of space travel and control the spaceship automatically. Several novel techniques are invented by us in this model. We prove adding linear layers in ANN can boost the GA training and GA can be a good supervisor when it is expensive to find a human trainer.  

7. Text-based Protocol Detection and Learning (Project leader, 2006)

       Many internet applications design their own communication protocols, and most of those protocols are based on text. It is an interesting machine learning question that whether our machines can detect a communication channel and learn the protocol in the channel. This project tries to handle this task with some text classification methods, and evaluate the performance by two factors: accuracy and classification delay.

8.  Organ Procurement and Transplantation Network(OPTN), Survey and Suggestion (Team leader, 2007)

The OPTN, a savior in the eyes of many diseased, saves and prolongs numerous lives. Although the OPTN is functioning, there are still approximately 95,000 candidates waiting for an organ. The efficiency and effectiveness of the network are in dire need of improvement. Our project demonstrates several solutions to the critical bottleneck in current OPTN system.