Department of Computer Science
University of Texas at San Antonio
Schedule:
Abstract:
As the size of information repositories increases, the importance of retrieval operations rises. This partly explains the success of internet search engines. The type of information object gradually changes and becomes more complex. Starting from huge text collections, we have moved to large image collections. More recently, aided by the proliferation of three-dimensional scanners and modeling software, collections of three-dimensional models (referred to as 'objects' here) are also expanding in size. Some collections contain objects of various classes (e.g. furniture) while some are more specialized and contain objects of a single class (e.g. human faces for biometric applications).
In this talk we present recent work on accurate and efficient inter-class object retrieval, based on both two-dimensional (depth buffers) and three-dimensional (spherical harmonics) characteristics of objects. Experiments show that when such inter-class retrieval methods are applied to intra-class problems, the results are quite poor. In intra-class retrieval it is necessary to exploit the specific characteristics of the class in order to distinguish the small differences that exist between objects belonging to the same class. An intra-class retrieval method, based on a parameterized and annotated class model, is also presented. Its very encouraging results in face and ear recognition are given. The datasets used include the Face Recognition Grand Challenge (FRGC) 2.1. Our intra-class method is a contestant in both the FRGC and the Face Recognition Vendor Test (FRVT) and boasts the top published accuracy to date.
Bio
Theoharis Theoharis received the BSc. degree in Computer Science from
the University of London in 1984, the MSc. Degree in Computation from
the University of Oxford in 1985 and the Ph.D. degree in Computer
Graphics and Parallel processing from the University of Oxford in 1988.
He served as a Research Fellow at the University of Cambridge between
1987 and 1990 and as a Consultant with Andersen Consulting between 1992
and 1993. He is currently Associate Professor at the Department of
Informatics, University of Athens. He has extensively published in
journals and conferences in the areas of Biometrics, Computer Graphics,
Visualisation and Parallel Processing.
References:
Abstract:
Clustering methods are widely used on gene expression data to
categorize genes with similar expression profiles. Finding an appropriate
similarity measure is critical to the analysis. The authors developed a
new technique for clustering the genes when the key factor is the shape of
the profile, and when the expression magnitude should also be accounted
for in determining the gene relationship. This is achieved by modeling the
shape and magnitude parameters separately in a gene expression profile,
and then using the estimated shape and magnitude parameters to define a
metric in a new feature space. Several different transformation schemes to
construct the feature spaces are explored, including a space whose
features are determined by the mutual differences of the original
expression components, a space derived from a parametric covariance
matrix, and the principal component space in traditional PCA analysis. The
former two are the newly proposed and the latter is explored for
comparison purposes. The new measures were employed in a k-means
clustering procedure to perform analyses. Application of these algorithms
to a simulation and experimental datasets shows that the algorithm
associated with the first feature space, named TransChisq, showed
clear advantages over other methods.
References:
Abstract:
Automated annotation of digital pictures has been a highly challenging
problem for computer scientists since the invention of computers. The
capability of annotating pictures by computers can lead to breakthroughs in a wide
range of applications including Web image search, online picture-sharing communities,
and scientific experiments. The ALIPR (Automatic Linguistic Indexing of Pictures - Real Time) system of
fully automatic and high speed annotation for online pictures has been constructed.
Thousands of pictures from an Internet photo-sharing site,
unrelated to the source of those pictures used in the training
process, have been tested. The experimental results show
that a single computer processor can suggest annotation
terms in real-time and with good accuracy.
References:
Abstract:
Gene expression programs depend on recognition of cis elements in
promoter region of target genes by transcription factors (TFs), but how TFs regulate
gene expression via recognition of cis elements is still not clear. To study this issue, developing a
cis-regulatory circuit of a gene is very useful. The authors have
proposed a novel cross-gene identification scheme to infer how multiple TFs
coordinate to regulate gene transcription in the yeast cell cycle and to
uncover hidden regulatory functions of a cis-regulatory circuit. They have applied this method to
cell cycle genes because the available expression profiles for these genes are long
enough. This method not only can quantify the regulatory strengths and synergy of the TFs but also
can predict the expression profile of any gene having a subset of the cis elements studied.
References:
Abstract:
We have developed a visualization methodology, called a Cluster
Overlap Distribution Map (CODM), for comparing the clustering results of
time-series gene expression profiles generated under two different
conditions. Although various clustering algorithms for gene expression
data have been proposed, there are few effective methods to compare
clustering results for different conditions. Using CODM, the
utilization of three-dimensional space and color allows intuitive
visualization of changes in cluster set composition, changes in the
expression patterns of genes between the two conditions, and relationship
with other known gene information, such as transcription factors.
References:
Abstract:
Rekindled by a boost in demand for highly accurate and non-repudiative identification and verification
solutions, biometrics research has attracted more and more focus. In this talk, three biometric modalities will be studied, namely fingerprint,
iris and palmprint. I will try to probe behind the scene to expose what kinds of pattern recognition techniques have been
proven to be successful in state-of-the-art commercial applications, and what kinds of issues are deterring the technology from
becoming more widespread.
References:
Abstract:
The existing methods focus on a universal approach that exerts the same
amount of preservation for all persons, without catering for their
concrete needs. The consequence is that we may be offering insufficient
protection to a subset of people, while applying excessive privacy control
to another subset.
From this paper, I will introduce a new generalization framework based on
the concept of personalized anonymity, which is to satisfy everybody's
requirements, and thus, retains the largest amount of information from the
microdata. I will also introduce a careful theoretical study that leads to
valuable insight into the behavior of alternative solutions, such as
K-Anonymity and L-Diversity.
References:
Please send emails to qitian@cs.utsa.edu, or seminar co-organizers: Kay Robbins, Weining Zhang, Yufei Huang, Carola Wenk, and Qi Tian.