Research Interests
- Machine Learning and Neural Networks
Neural networks, competitive learning, classifiers, clustering, Minimax Probability Machines, SVM, kernel methods, data mining, machine learning techniques for information retrieval, relevance feedback, social computing, and multimedia processing, etc.
- Social Computing and Web Intelligence
Social computing, social media, social networks, search engine, crawler, recommender systems, collaborative filtering, link analysis, similarity rank, ranking algorithms, learning algorithms, web document classification, web mining, web information retrieval, web knowledge management, wiki, etc.
- Multimedia Information Processing and Retrieval
Information retrieval, text processing, Chinese text processing, plagiarism detection, readability analysis, multimedia content analysis, content-based information retrieval, relevance feedback, P2P information retrieval, face recognition, object tracking, high-dimension indexing, etc.
Projects
- Social Computing: Models and Systems
In this project, we seek to formulate a formal framework to model human/social computation. Moreover, we plan to design and implement at least one language game with detailed analysis that is based on the proposed formal model. By studying the properties of the model, we can improve existing implementations and benefit future development of similar systems.
- Collaborative Filtering and Social Networks
We work on missing value prediction algorithms and similarity computation between users and items in recommender systems.
- Link-based Ranking Algorithms
We examine ways to rank objects, i.e., pages, relations, documents, etc., using link relationship that exists between these objects. The result can be combined with content-based approach to rank web sites according to link similarity.
- Extending the Minimum Error Minimax Probability Machine
This project plans to extend a distribution-free Bayes optimal classifier called the Minimum Error Minimax Probability Machine (MEMPM) by formulating a unified general framework for a family of classifiers.
- Exploration of Low Density and Manifold Assumptions in Semi-supervised Learning
This project proposes a novel classification methodology that uses the Gaussianity assumption to derive quadratic discriminant functions for classification based on the Maximum a Posteriori (MAP) value in the feature space.
The VeriGuide System. VeriGuide is a plagiarism identification engine for detecting similar text among English and Chinese documents. This project involves text processing, web mining, information retrieval, matching algorithms, readability analysis, performance issues, etc.
