Data and Vision Weekly Seminar, CS, UTSA

Fall 2005 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: 3:00-4:00 pm, Every Friday
Place: SB 4.01.20, CS Conference Room

Schedule:

09/23/05

Bayesian Inference of Genetic Regulatory Networks using Dynamic Bayesian Networks

09/30/05

Robust and Accurate Cancer Classification with Gene Expression Profiling

10/07/05

Incremental Semi-Supervised Subspace Learning for Image Retrieval

10/14/05

Automatic Linguistic Indexing of Pictures by a Statistical Modeling Approach

11/11/05

The factor graph network model for biological systems

11/18/05

Dimensionality Reduction Using Secant-Based Projection Methods

12/02/05

Map-Matching Vehicle Tracking Data

Schedule for Fall 2005

09/23/05 Bayesian Inference of Genetic Regulatory Networks using Dynamic Bayesian Networks
Speaker: Jia-Ying Wang
Presentation slides:
Abstract:
Inference of genetic regulatory networks (GRNs) of cell cycles based on microarray data is considered in this presentation. In particular, a dynamic-Bayesian-networks (DBNs) model and a full Bayesian learning scheme are introduced. The proposed DBN models directly the continuous expression levels and also is associated with parameters that indicate the degree as well as the types of regulations. To learn the network from data, the reversible jump Markov chain Monte Carlo (RJMCMC) algorithm is proposed. The RJMCMC algorithm can provide not only more accurate inference results than the deterministic alternative algorithms but also an estimate on the a posteriori probabilities (APPs) of the network topology. The estimated APPs provide useful information on the confidence of the inferred results and can also be used for efficient Bayesian data integration. The proposed approach was tested on yeast cell cycle microarray data and the results were compared with the KEGG pathway map.

References:
Back to top
09/30/05, Robust and Accurate Cancer Classification with Gene Expression Profiling
Speaker: Yijuan Lu
Presentation slides in [PDF]
Abstract:
Robust and accurate cancer classification is critical in cancer treatment. Gene expression profiling is expected to enable people to diagnose tumors precisely and systematically. However, the classification task in this context is very challenging because of the curse of dimensionality and the small sample size problem. In this paper, the author proposes a novel method to solve these two problems. Their method is able to map gene expression data into a very low dimensional space and thus meets the recommended samples to features per class ratio. As a result, it can be used to classify new samples robustly with low and trustable (estimated) error rates. The method is based on linear discriminant analysis (LDA). However, the conventional LDA requires that the within-class scatter matrix Sw be nonsingular. Unfortunately, Sw is always singular in the case of cancer classificati on due to the small sample size problem. To overcome this problem, the author develops a generalized linear discriminant analysis (GLDA) that is a general, direct, and complete solution to optimize Fishers criterion. GLDA is mathematically well-founded. It does not assume the nonsingularity of Sw, and thus naturally solves the small sample size problem. To accommodate the high dimensionality of scatter matrices, a fast algorithm of GLDA is also developed. Some extensive experiments on seven public cancer datasets show that the method performs well.

References:
Back to top
10/07/05, Incremental Semi-Supervised Subspace Learning for Image Retrieval
Speaker: Jerry Yu
Presentation slides [PDF]
Abstract: Subspace learning techniques are widespread in pattern recognition research. They include Principal Component Analysis (PCA), Locality Preserving Projection (LPP), etc. These techniques are generally unsupervised which allows them to model data in the absence of labels or categories. In relevance feedback driven image retrieval system, the user provided information can be used to better describe the intrinsic semantic relationships between images. A semi-supervised subspace learning algorithm is proposed to incrementally learn an adaptive subspace by preserving the semantic structure of the image space, based on user interactions in a relevance feedback driven query-by-example system. It is capable of accumulating knowledge from users, which could result in new feature representations for images in the database so that the system's future retrieval performance can be enhanced. Experiments on a large collection of images have shown the effectiveness and efficiency of the algorithm.
References:
Back to top
10/14/05, Automatic Linguistic Indexing of Pictures by a Statistical Modeling Approach
Speaker: Lije Zhang
Presentation slides in [PDF]
Abstract: Automatic linguistic indexing of pictures is an important but highly challenging problem for researchers in content-based image retrieval. So far, there have been three categories of technology trends to realize content-based image retrieval system. In this paper, the author introduces a 2D MHMM (2-dimensional Multiresolution Hidden Markov Model) modeling approach to solve this problem. Experiments have demonstrated the good accuracy of the complementation of this approach and its high potential in linguistic indexing of photographic images.
References:
Back to top
11/11/05, The factor graph network model for biological systems
Speaker: Isabel Tienda Luna
Presentation slides.
Abstract:
We introduce an extended computational framework for studying biological systems. Our approach combines formalization of existing qualitative models that are in wide but informal use today, with probabilistic modeling and integration of high throughput experimental data. Using our methods, it is possible to interpret genomewide measurements in the context of prior knowledge on the system, to assign statistical meaning to the accuracy of such knowledge and to learn refined models with improved fit to the experiments. Our model is represented as a probabilistic factor graph and the framework accommodates partial measurements of diverse biological elements. We develop methods for inference and learning in the model. We compare the performance of standard inference algorithms and tailor-made ones and show that hidden variables can be reliably inferred even in the presence of feedback loops and complex logic. We develop a formulation for the learning problem in our model which is based on deterministic hypothesis testing, and show how to derive p-values for learned model features. We test our methodology and algorithms on both simulated and real yeast data. In particular, we use our method to study the response of S. cerevisiae to hyperosmotic shock, and explore uncharacterized logical relations between important regulators in the system.
References:
1. I. Gat-Viks, A. Tanay, D. Raijman, and Ron Shamir, "The Factor Graph Network model for biological systems," In Proceedings of RECOMB, 2005. [PDF]
2. I. Gat-Viks, A. Tanay, and R. Shamir, "Modeling and analysis of heterogeneous regulation in biological networks," In Proceedings of the first Annual RECOMB Satellite Workshop on Regulatory Genomics, pages 21-31, 2004. Also J. Comput Biol in press.
3. S. Yedidia, WT. Freeman, and Y. Weiss, "Constructing free energy approximations and generalized belief propagation algorithms," Technical Report TR-2004-040, Mitsubishi electric research laboratories, 2004. [PDF]
4. D. J. C. MacKay, "Introduction to Monte Carlo methods," In M. I. Jordan, editor, Learning in Graphical Models, pages 175-204. Kluwer Academic Press, 1998. [PDF]
Back to top
11/18/05, Dimensionality Reduction Using Secant-Based Projection Methods

Speaker: Dragana Veljkovic
Presentation Slides in [PDF]
Abstract: High dimensional data that lies on a smooth low dimensional manifold is often found in natural processes. Finding a low dimensional representation of such data significantly improves the ability to extract knowledge. We are interested in finding good projections of the data, where a projection is considered to be good if its inverse mapping is well conditioned. Whitney's embedding theorem states that every smooth m-dimensional manifold can be projected into an (2m+1)-dimensional Euclidian space without loss of information. Based on the proof of Whitney's theorem, Broomhead and Kirby have developed a method for selecting good projection is developed [1, 2]. The projections are constructed by considering the effect on the set of unit secants calculated from the data. A method for constructing the inverse projection using radial basis functions is described as well [3].

References:
1. Broomhead, D. S. and M. Kirby (2000). "A new approach to dimensionality reduction: theory and algorithms." SIAM Journal on Applied Mathematics 60(6): 2114-2142. [PDF]
2. Broomhead, D. S. and D. R. Kirby (2005). "Dimensionality reduction using secant-based projection methods: the induced dynamics in projected systems." Nonlinear Dynamics 41(1-3): 47-67. [PDF]
3. Broomhead, D. S. and D. Lowe (1988). "Multivariable functional interpolation and adaptive networks." Complex Systems 2: 321-355. [PDF]
Back to top
12/02/05, Map-Matching Vehicle Tracking Data
Speaker: Randall Salas
Presentation slides
Abstract
Vehicle tracking data is used in many applications such as traffic control, routing, navigation, and vehicle location. A common method to obtain vehicle tracking data is using GPS. However, sampling a vehicle using GPS does not give an accurate data of the vehicle position since it suffers from noise and error sampling rate. Therefore, algorithms have to be implemented to match the vehicle to a road map. Six algorithms that provide map-matching techniques are presented; five of them using a variant of the Fr�chet distance method which guarantees a quality measure, and the other one with no quality measure. An incremental algorithm (heuristic) is presented that matches consecutive portions of the trajectory to the road network, effectively trading accuracy for speed of computation. On the other hand we propose algorithms which make a global decision and which provide good quality measures, such as: Integral Fr�chet, Sum Fr�chet, Weak Fr�chet (Using Depth First Search and Dijkstra) and Adaptive Weak Fr�chet.
Reference
1. S. Brakatsoulas, D. Pfoser, R. Salas, and C. Wenk, "On Map-Matching Vehicle Tracking Data", to appear in Proc. of the 31st VLDB Conference, Trondheim, Norway, 2005. [PDF]
2. C. Wenk, H. Alt, A. Efrat, L. Palaniappan, and G. Rote, "Finding a Curve in a Map," Video, 19th Ann. Symp. Computational Geometry (SoCG), 2003, San Diego, USA.
3. C. Wenk, R. Salas, and D. Pfoser, "Adressing the Need for Map-Matching Speed: Localizing Global Curve-Matching Algorithms," in preparation. [PDF]
4. H. Alt, A. Efrat, G. Rote, and C. Wenk, "Matching Planar Maps," Journal of Algorithms, 49: 262-283, 2003. [PDF]
Back to top

Questions and Comments?

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