• Home
• Resume
• Research
• Publications
• Teaching
• Coursework
• Activities
• Useful Links

Yanhua Chen

Home  Research

I. Postdoc Research Summary More than 150,000 individuals in the USA are living with a kidney transplant. There are a long waiting list: 82,966 patients on the list as of January 23, 2010, and in 2009, there were 28,381 patients have been added to the list, while only 14,060 patients have been luckily received the transplant kidneys. Deceased donation and living donations are the two resources of organs for transplantation, and living-donor transplant has a higher chance of success. However, often times willing donors (even within the same family) are incompatible with recipients due to various biological factors, such as ABO blood type mismatch and human leukocyte antigen (HLA) mismatch. The National Kidney Paired Donation (KPD) transplants program has been established as a novel clinical solution to overcome the shortage of donors. The essential idea of the KPD program is to exchange living kidney donors between two recipient/donor pairs. The fundamental question in the KPD program is how to make an optimal decision of organ exchanges that benefit patients the best. We propose to take a graph-based approach, Optimal Graph Crossmatch Model (OGCM), for the KPD program to provide innovative statistical tools to design clinical studies and to analyze data from clinical studies, and hence improving patients’ well-being as well as quality of life. OGCM will enhance existing algorithms to analyze both static and dynamic large graphs to identify all compatible matches between any number recipient/donor pairs, leveraging the efficiency of graph matching strategy and the optimization of maximum number and quality of transplants. The goal of the project is to integrate high-level user interactivity into OGCM through software and real-time computing to find the set of mutually exclusive kidney exchanges that achieves the maximum number of transplants offering the highest quality transplants. Top External Projects Optimal Graph Crossmath Model for Kidney Paired Donation (KPD) Program, NSF/CRA Top SOFTWARE/CODE AND DATA KPD Software Top

II. PhD Research Summary Generally, data mining (sometimes called data or knowledge discovery) is the process of analyzing data from different perspectives and summarizing it into useful information - information that can be used to increase revenue, cuts costs, or both. The most commonly tasks in data mining include: classification, clustering, visualization, estimation and so on. Clustering or unsupervised learning is a generic name for a variety of procedures designed to find natural groupings or clusters in multidimensional data based on measured or perceived similarities among the patterns. The purpose of clustering is to extract useful information from unlabeled data, and it plays a very important role in data mining. Moreover, with the fast growth of Internet and computational technologies in the past decade, many data mining applications have advanced swiftly from the simple clustering of one data type to the co-clustering of multiple data types, usually involving high heterogeneity. Applications of data clustering/co-clustering are found in many fields, such as information discovery, text mining, web analysis, image grouping, medical diagnosis, and bioinformatics. In many practical learning domains (e.g., text processing, bioinformatics), there is a large supply of unlabeled data but limited labeled data, and in most cases it might be expensive to generate large amounts of labeled data. Traditional clustering/co-clustering algorithms completely ignore these valuable labeled data and thus are inapplicable to these problems. Consequently, semi-supervised clustering, which can incorporate the domain knowledge to guide a clustering algorithm, has become a topic of significant recent interest. We first develop a Non-negative Matrix Factorization (NMF) based framework to incorporate prior knowledge into data clustering. Later, we extend SS-NMF to do heterogeneous data co-clustering. From a theoretical perspective, SS-NMF for data clustering/co-clustering is mathematically rigorous. The convergence and correctness of our algorithms are proved. In addition, we show that our work provides a unified view for data clustering/co-clustering. Some well-established approaches can be considered as special cases or variations of our models. Experiments performed on various publicly available data sets demonstrate the superior performance of our work. On the other hand, we further propose a novel model for exemplar-based clustering. Exemplar-based clustering is to find the representative "exemplars" from the actual data points and simultaneously cluster them into meaningful groups characterized by exemplars. Instead of "centroids" derived from the traditional clustering, exemplars are more useful since they enable us to better summarize and visualize the data.

Top

• CRI: IAD Acquisition of Research Infrastructure for Knowledge-enhanced Large-scale Learning of Multimodality Visual Data, NSF



• HyperEye: Susceptibility Weighted Imaging-based Informatics Tools for Brain Tumor Studies, State of Michigan, 21st Century Jobs Fund



• Collaborative Research: Integrated Modeling and Learning of Multimodality Data across Subjects for Brain Disorder Study, NSF

Top

• SS-NMF for Homogeneous Data Clustering Sample Code Download
• SS-NMF for Heterogeneous Data Clustering Sample Code Download
• Exemplar-based Visualization Data and Software Demo Download

Top

• Semi-supervised NMF for Homogeneous Data Clustering




Traditional clustering algorithms are inapplicable to many real-world problems where limited knowledge from domain experts is available. Incorporating the domain knowledge can guide a clustering algorithm, consequently improving the quality of clustering. We propose SS-NMF: a semi-supervised non-negative matrix factorization framework for data clustering. In SS-NMF, users are able to provide supervision for clustering in terms of pairwise constraints on a few data objects specifying whether they "must" or "cannot" be clustered together. Through an iterative algorithm, we perform symmetric trifactorization of the data similarity matrix to infer the clusters. Theoretically, we show the correctness and convergence of SS-NMF and SS-NMF provides a general framework for semi-supervised clustering. Through extensive experiments conducted on publicly available datasets, we demonstrate the superior performance of SS-NMF for clustering. Further details of this work are available here.

Top




• Semi-supervised NMF for Heterogeneous Data Co-clustering




Co-clustering heterogeneous data has attracted extensive attentions recently due to its high impact on various important applications, such us text mining, image retrieval and bioinformatics. However, data co-clustering without any prior knowledge or background information is still a challenging problem. We propose a Semi-Supervised Non-negative Matrix Factorization (SS-NMF) framework for data co-clustering. Our method computes a new pairwise relational matrix by incorporating user provided constraints through distance metric learning. Using an iterative algorithm, we perform tri-factorization of the new matrix to infer the clusters of two data types. Through extensive experiments conducted on publicly available data sets, we demonstrate the superior performance of SS-NMF for data co-clustering. Further details of this work are available here.

Top




• Exemlpar-based Visualization of Large Document Corpus




With the rapid growth of the World Wide Web and electronic information services, many data, such as text corpus, bioinformatics, and so on, are becoming available on-line at an incredible rate. By displaying them in a logical layout (e.g., color graphs), data visualization presents a direct way to observe the data as well as understand the relationship between them. In this work, we propose a novel technique, Exemplar-based Visualization (EV), to visualize an extremely large data sets. Capitalizing on recent advances in matrix approximation and decomposition, EV presents a probabilistic multidimensional projection model in the low-rank data subspace with a sound objective function. The probability of each data sample proportion to the class is obtained through iterative optimization and embedded to a low dimensional space using parameter embedding. By selecting the representative exemplars, we obtain a compact approximation of the data. This makes the visualization highly efficient and flexible. In addition, the selected exemplars neatly summarize the entire data set and greatly reduce the cognitive overload in the visualization, leading to an easier interpretation of large data sets. Empirically, we demonstrate the superior performance of EV through extensive experiments performed on the publicly available data sets. Further details of this work are available here.

Top