Data and Vision Weekly Seminar, CS, UTSA

Spring 2006 Data and Vision Weekly Seminar

Department of Computer Science
University of Texas at San Antonio

Organizers: Qi Tian, Kay Robbins, Weining Zhang, Yufei Huang (EE), Carola Wenk.
Time: 10:00am - 12:00am, Every Friday
Place: SB 4.01.20, CS Conference Room

Schedule:

02/10/06

at 11 am

Disease Gene Detection: Linkage and Microarrays

02/24/06

How to use grants.gov

03/3/06

Ranking metrics and evaluation measures

03/10/06

Knowledge Representation and Sharing Using Visual Semantic Modeling for Diagnostic Medical Image Data bases

at 11 am

Schedule for Spring 2006

02/10/06 at 11 am Disease Gene Detection: Linkage and Microarrays
Speaker: Christopher Jenkinson, UTHSCSA
Abstract:
The use of high throughput methods for detecting genetic variants and measuring gene expression has produced a recent dramatic increase in the abundance of data for the localization and detection of genes involved in human disease. The power of these approaches is attainable only with the application of sophisticated computerized data processing and statistical analysis. The increase in genetic "horsepower" is further enhanced by vast online databases of DNA sequence information, including multiple complete genome sequences and several million human single nucleotide polymorphisms (SNPs). With these tools at hand we can now begin to grapple with the most intractable of genetic problems - identification of genes for complex diseases such as cancer, atherosclerosis, diabetes and obesity. This talk will describe the application of these molecular genetic approaches to the dissection of complex traits, including type 2 diabetes in the San Antonio Mexican American community.
Back to top
02/24/06 How to use grants.gov
Speaker: Liz Garcia, UTSA Office of Research Development
Presentation slides

Abstract:
We will be providing a general overview of how the federal government is changing from hardcopy submissions of grant proposals to electronic submission. This change is and will be in effect for all 26 federal agencies by the end of 2007. Due to this change that is taking place which is mandated by federal law anyone interested in applying for a federal grant will be required to learn how to navigate and utilized the electronic GRANTS.GOV submission system. We will also be discussing how UTSA will be implementing new policies in order to comply with GRANTS.GOV rules and regulations. The agenda for our presentation will cover the following topics for discussion:
1. What is Grants.gov?
  Grants.gov History
2. Federal Agencies Current Situation
  NIH, NSF, USDA
3. Application Tips
  PureEdge software usage tips
4. Resources, Important Tips
  Grants.gov, R&D Webpage, Training Sessions
Participants will gain an understanding of how to get started and who to contact for information concerning GRANTS.GOV.

References:
- Grants.gov pages of the Office of Research Development
- grants.gov
Back to top
03/3/06 Ranking metrics and evaluation measures
Speaker: Jerry Yu
Abstract:
In this work, we present a general guideline to establish the relation between a distribution model and its corresponding similarity estimation. A rich set of distance metrics, such as Harmonic distance and Geometric distance, is derived according to Maximum Likelihood theory. These metrics can provide a more accurate model than the conventional Euclidean distance and Manhattan distance. Because the feature elements are from heterogeneous sources and may have different influence on similarity estimation, the assumption of single isotropic distribution model is often inappropriate. We propose a novel boosted distance metric that not only finds the best distance metric that fits the distribution of the underlying elements but also selects the most important feature elements in respect to similarity. We experiment with different distance metrics for similarity estimation and compute the accuracy of different methods in two applications: stereo matching and motion tracking in video sequences. The boosted distance metric is tested on fifteen benchmark data sets from the UCI repository and two image retrieval applications. In all the experiments, robust results are obtained based on the proposed methods.
References:
[1] J. Yu, N. Sebe and Q. Tian, "Ranking Metrics and Evaluation Measures," invited book chapter, Advances in Imaging and Electron Physics, Elsevier, to appear in 2006, pdf
[2] Presentation slides: pdf
Back to top
3/10/06 Knowledge Representation and Sharing Using Visual Semantic Modeling for Diagnostic Medical Image Data bases
Speaker: Lijie Zhang
Abstract:
Information technology offers great opportunities for supporting radiologists expertise in decision support and training. However, this task is challenging due to difficulties in articulating and modeling visual patterns of abnormalities in a computational way. To address these issues, well established approaches to content management and image retrieval have been studied and applied to assist physicians in diagnoses. Unfortunately, most of the studies lack the flexibility of sharing both explicit and tacit knowledge involved in the decision making process, while adapting to each individuals opinion. In this paper, they propose a knowledge repository and exchange framework for diagnostic image databases called evolutionary system for semantic exchange of information in collaborative environments (Essence).This framework uses semantic methods to describe visual abnormalities, and offers a solution for tacit knowledge elicitation and exchange in the medical domain. Also, they provides a computational and visual mechanism for associating synonymous semantics of visual abnormalities. Another contribution is that they model the user preference in a computational way so that the framework is able to serve according to user profile.
References:
[1] Adrian S. Barb, Chi-Ren Shyu, Member, and Yash P. Sethi, "Knowledge Representation and Sharing Using Visual Semantic Modeling for Diagnostic Medical Image Databases", IEEE Trans. on Inf. Technology in Biomedicine, Vol. 9, No. 4, 2005, pages 538-553, pdf
Back to top
3/31/06 at 11 am An Interactive Self-Supervised Learning Framework for Classifying Microarray Gene Expression Data
Speaker: Yijuan Lu
Abstract:
With the development of microarray technology, it provides massive amounts of high dimensional gene expression data simultaneously and most of their functions are unknown. Computational methods that can effectively resolve high dimensionality and small sample size problems for the high throughput data are valuable in systems biology. In this presentation, I introduce a self-supervised learning algorithm for classifying microarray gene expression data: Kernel Discriminant-EM (KDEM) algorithm. This algorithm efficiently utilizes a large set of unlabeled data to compensate for the insufficiency of a small set of labeled data and it extends linear algorithm in DEM to kernel algorithm to handle nonlinearly separable data in a lower dimensional space. More importantly, to bridge the gap between the gene expression data and the associated functions, Relevance Feedback is introduced to microarray analysis to incorporate the expert knowledge into the decision module. KDEM and relevace feedback together construct an efficient and effective learning framework. Extensive experiments on the yeast cell cycle regulation dataset and Plasmodium falciparum red blood cell cycle dataset show the promise of the approach.
References:
[1] An Interactive Self-Supervised Learning Framework for Classifying Microarray Gene Expression Data, by Y. Lu, Q. Tian, F. Liu, M. Sanchez, Y. Wang, accepted to International Workshop on Bioinformatics Research and Applications (IWBRA), pdf
Back to top
4/7/06 Applications of Subspace Distance Measures
Speaker: Dragana Veljkovic
Abstract:
Finding a good representation of a high-dimensional data set can be crucial to understanding its intrinsic structure. Whether the data is clustered or distributed throughout its space-time domain, information about the local subspace structure and the relationship in the whole data set can assist in learning the global data structure. In some cases, local neighborhoods that appear distant in the Euclidian space can have very similar intrinsic structure. Similarly, some clusters with distant centroids can lie on parallel planes. Unless specifically analyzed, this local behavior can remain unnoticed. We will present several subspace distance metrics and show how they can be applied to existing problems in image processing, dimension reduction, clustering and other data analysis problems.
References:
[1] D. Q. Huynh, A. Heyden, "Outlier detection in video sequences under affne projection," Proc. IEEE Conf. Comput. Vision Pattern Recog., vol.2, pp.695-- 701, Kauai, HI, U.S.A., Dec. 2001.
[2] D. Q. Huynh, A. Heyden, "Robust Factorization for the Affine Camera: Analysis and Comparison," the Seventh International Conference on Control, Automation, Robotic and Vision (ICARCV'02), Dec 2002.
[3] T.-J. Chin, J. U, K. Schindler, D. Suter, ? Face Recognition from Video by Matching Image Sets,? Proc. Digital Image Computing: Techniques and Applications, 2005.
Back to top
4/21/06 Using modeling and abstraction to develop a flexible, domain specific XML data representation
Speaker: Ken Holladay
Abstract:
The benefits of using XML as a platform independent data description language are well established. However, many domain specific XML formats (typically expressed using XML Schema) are either too general or too rigid. Very general formats can be expressed using XML constructs like "any" so that the XML data file will validate with any tag. However, this type of description conveys minimal domain specific information and requires significant discovery efforts by a data consumer. The opposite extreme is a rigid description of every data field. While this may be adequate for a simple problem domain, it is problematic for complex domains where integrating new data fields or collections of fields would require frequent schema revisions. This talk describes a metamodel approach to solving this dilema. First, a meta schema defines how to describe data elements and model components for the domain. A data file based on the meta schema contains descriptions for specific instances of data elements and model components. A second schema defines how to assemble the data elements and model components into a specific data description instance. The technique is illustrated using non-cooperative signal analysis as the problem domain.
References:
[1] Zaijun Hu; Vollmar, G., "Towards XML metamodel patterns for XML data modeling" 12th International Workshop on Database and Expert Systems Applications, Sept. 2001.
Abstract:
Modelling sequential data is important in many areas of science and engineering. Hidden Markov models (HMMs) and Kalman filter models (KFMs) are popular for this because they are simple and flexible. For example, HMMs have been used for speech recognition and bio-sequence analysis, and KFMs have been used for problems ranging from tracking planes and missiles to predicting the economy. However, HMMs and KFMs are limited in their "expressive power". Dynamic Bayesian Networks (DBNs) generalize HMMs by allowing the state space to be represented in factored form, instead of as a single discrete random variable. DBNs generalize KFMs by allowing arbitrary probability distributions, not just (unimodal) linear-Gaussian. In this talk, I will lead you to review Dr. Kevin Murphy's thesis and discuss how to represent many different kinds of models as DBNs, how to perform exact and approximate inference in DBNs, and how to learn DBN models from sequential data.
Back to top
4/28/06 Dynamic Bayesian Networks: Representation, Inference and Learning
Speaker: Tao Wei
Abstract:
Modelling sequential data is important in many areas of science and engineering. Hidden Markov models (HMMs) and Kalman filter models (KFMs) are popular for this because they are simple and flexible. For example, HMMs have been used for speech recognition and bio-sequence analysis, and KFMs have been used for problems ranging from tracking planes and missiles to predicting the economy. However, HMMs and KFMs are limited in their "expressive power". Dynamic Bayesian Networks (DBNs) generalize HMMs by allowing the state space to be represented in factored form, instead of as a single discrete random variable. DBNs generalize KFMs by allowing arbitrary probability distributions, not just (unimodal) linear-Gaussian. In this talk, I will lead you to review Dr. Kevin Murphy's thesis and discuss how to represent many different kinds of models as DBNs, how to perform exact and approximate inference in DBNs, and how to learn DBN models from sequential data.
Back to top

Questions and Comments?

Please send emails to carola@cs.utsa.edu, or seminar co-organizers: Kay Robbins, Weining Zhang, Yufei Huang, Carola Wenk, and Qi Tian.