Xue-wen Chen, Ph.D. - US grants

Abstract

Program: NSF 04-588 CISE Computing Research Infrastructure
Title: CRI: Collaborative Research: Reconfigurable Computing Cluster

Lead Proposal: CNS 0551688
PI: Sass, Ronald
Institution: University of Kansas Center for Research Inc

Collaborative Proposal: CNS 0551678
PI: Chatha, Karamvir
Institution : Arizona State University

Researchers at the University of Kansas, Arizona State University and Clemson University will acquire a 64-node experimental Reconfigurable Computing Cluster to support on-going research projects investigating High-Performance Computing. They will use off-the-shelf components including field programmable gate arrays (FPGA's) and an inexpensive custom network adapter to enable new research that is a natural out-growth of established research programs. The experimental cluster will enable the researchers to conduct their research projects and to evaluate two novel aspects of the proposed architecture: (a) using the FPGAs as the primary processor on the node, and (b) integrating the network switching components into the configurable resources of the FPGA. If successful, this approach could lead to significant advances in high performance computing. Projects supported will include: network scalability to assess how the number of nodes and distance between nodes affects performance; assessing use of these FPGA based clusters for common applications such as computation fluid dynamics; and use of these models for RC-BLAST, an important bio-informatics application. The investigators will develop undergraduate research projects using these resources.

NSF-0644366
Chen, Xue-Wen

The objectives of this research program are (1) to develop and apply novel computational
approaches for uncovering genome-wide networks of interactions between genes and proteins, and (2) to conduct related educational activities in a newly established bioinformatics program in the Department of Electrical Engineering and Computer Science at the University of Kansas. Specifically, built upon reconstructing biological networks of moderate size, the new research will computationally uncover genome-wide biological networks and map interactions of genes and proteins across a variety of organisms. The research directions include: Simultaneously integrating multiple biological knowledge into dynamic Bayesian networks for learning networks of gene interactions; learning networks of protein interactions from heterogeneous data; learning integrated networks of gene and protein interactions; learning genome-wide networks of gene and protein interactions; and cross-species network learning. It will advance the state of the art by developing machine learning methods for effectively integrating multiple prior knowledge from different sources of data, including learning for highly heterogeneous data and large-scale network. The research will also produce new methods and user-friendly software that can be applied by molecular biologists to gain insight into diverse biological problems, such as how biological processes are regulated on a genome scale and how individual bio-molecules interact with one another in the cell.

Learning with prior knowledge and highly heterogeneous data sources are fundamental to computational biology, information theory, machine learning, data mining, and other areas. Thus, the proposed research will benefit a variety of application domains including research in biology and medicine. The biological discovery derived from this project will also contribute to a variety of fields that include agriculture development, rational drug design, and health care. The research program will foster and facilitate collaborations between biologists and the PI. The educational components are closely tied to the research activities, which include (1) developing and improving bioinformatics courses that are closely related to the research outlined here and integrating them into the core bioinformatics curriculum, and (2) providing special training opportunities in the interdisciplinary area of bioinformatics for a wide-range of students, from high school through graduate school, including groups typically underrepresented in the field of science and technology.

The Third Annual IEEE International Conference on Bioinformatics and Biomedicine provides an open, interactive forum to promote multi- and interdisciplinary research and education in bioinformatics, computing science and biomedicine, facilitating development of new ideas for bridging gaps. In addition to contributed talks and special publication in top scientific journals, the meeting provides focused workshops, tutorials and posters. BIBM has made a significant effort to engage your researchers in the meeting organization and also provides mentoring activities in the program. There is particular support for both early career researchers and graduate students, providing both training and broadening the overall scientific impact of the conference.

The Treatise on Invertebrate Paleontology, founded in 1948 by an international consortium of paleontological societies, is considered to be the most authoritative compilation of data on invertebrate fossils. It holds an almost biblical significance and is to be found in every good library. The Treatise has found applications in many areas, such as understanding evolution, studying climate change, and finding fossil fuels.

In order to maximize the possible benefit from this landmark effort, there is a strong desire in the paleontological community to make this vast repository of paleontological data available in electronic form for current and future scientists and laypeople. The object of the proposed research is to facilitate knowledge discovery activities in invertebrate paleontology by providing scientists with a general framework that takes advantage of the rich information extracted from the Treatise. The first mission is to turn the Treatise data into available knowledge. The second mission is to develop computational tools for analyzing, modeling, and visualizing paleontological data in order to facilitate knowledge discovery.

The knowledgebase we propose to develop will play a central role in paleontological data management. It will facilitate paleontologists to further explore unexploited areas and to raise and answer research questions that are unsolvable under the current paradigm. Furthermore, it will provide a paradigm shift from the book-based knowledge system, which is perceived as supporting evidence of mainstream research and provides little knowledge regarding the patterns and relationships that are embedded within them, to a unified framework in which computational analysis and modeling are integrated with knowledge to derive a new era of paleontological research. Consequently, the knowledgebase will likely open transformational opportunities in scientific discovery to help understand the complexity of nature. Additionally, the website we will create will enable anyone to explore the world of fossils on the world-wide web, and to link with the University of Kansas Natural History Museum and their outreach programs for K-12 students and the general public.

This project develops algorithms for learning from images and video with noisy labels. The overwhelming amounts of images and video freely available online present unprecedented challenges for machine learning and computer vision research communities. They also bring tremendous opportunities and great potentials for addressing human-machine semantic gaps in image understanding and for revolutionizing our ways to index, retrieve, and interact with images and video. Inaccurate labels and mislabeled data are common problems for image and video datasets. Noisy labels would cause problems with the existing learning algorithms. This project can have broad impacts on other big data problem. The project is integrated with education by training students, ensuring broad participation of underrepresented groups, and outreaching general public.

This research exploring distributed learning methods for large-scale images and video with noisy labels. The PI investigates the learning problem of loss functions with both smooth and non-smooth regularization terms, and accordingly develops new distributed learning algorithms that are capable of leveraging the abundance of images that are too large to fit into a single machine. The research has an immense potential in image and video analysis, and computer vision applications. Specifically, this research emphasizes both algorithmic and theoretic aspects by (1) developing distributed learning based approaches for optimization and learning of noisy labels; and (2) investigating issues such as guaranteed convergence, convergence rate, and scalability. This work provides new methods that are widely applicable to many economically, medically and scientifically important large-scale datasets for novel discoveries across many domains.