Abstract:

Since 2007, high performance computing has been at the beginnings of the most dramatic change in form and function in the last decade and a half. Since the advent of the killer micro and the MPPs and commodity Clusters it spawned supported by message-passing programming techniques, most notably MPI, HPC has been on an exponential curve augmenting performance at historic rates through incremental changes to feature size, clock rate, and architectural complexity. But as always happens with S-curves, HPC is turning towards its final asymptote and is undergoing what may prove to be its 6th and potential final phase change. Most visible is the adoption of multicore heterogeneous system architectures driven by constraints in power, complexity, clock rate, and reliability while continuing to exploit improvements in feature size to achieve growth in performance. To realize this goal and the achievement of Exascale performance by the end of the next decade within practical limitations critical advances in efficiency, scalability, energy, and programmability will be required. In all previous such metamorphoses in HPC, the underlying principles of a new execution model was used to guide the co-design of new architectures, programming methods, and system software. Such is the case for the emerging HPC Phase VI. This presentation will discuss the likely elements the new execution model based on the exploratory ParalleX model of computation, and describe key attributes of architecture, operating, and runtime system software, and programming methods that are likely to gain ascendency over the next decade. Results from recent experiments with HPX prototype runtime system will be presented.

About the speaker:

Dr. Thomas Sterling is a Professor of Computer Science at Louisiana State University, a Faculty Associate at California Institute of Technology, a Distinguished Visiting Scientist at Oak Ridge National Laboratory, and a Fellow of Computer Science Research Institute at Sandia National Laboratory. He received his PhD as a Hertz Fellow from MIT in 1984. Dr. Sterling is probably best known as the “father” of Beowulf clusters and for his research on Petaflops computing architecture. He was one of several researchers to receive the Gordon Bell Prize for this work on Beowulf 1997. In 1996, he started the inter-disciplinary HTMT project to conduct a detailed point design study of an innovative Petaflops architecture. He currently leads the MIND memory accelerator architecture project for scalable data-intensive computing and is an investigator on the DOE sponsored Fast-OS Project to develop a new generation of configurable light-weight parallel runtime software system. Thomas is co-author of five books and holds six patents.

Abstract:

Tracking of precipitation events in weather imagery is important for several weather-related applications including forecasting and clutter/rain classification. In particular, the path characteristics of tracked events in weather imagery can be used to classify the event as precipitation, which usually moves quickly, or clutter which usually appears to be relatively static. In order to facilitate tracking, precipitation can be modeled as a mixture of localized functions, whose shape, position, and intensity change in time. Tracking of precipitation events can be performed via tracking of the localized function parameters. Recently,Radial Basic Function neural network (RBFNN) techniques have been used for nowcasting (forecasting a few minutes ahead). These techniques are effective, yet not efficient. In this talk, we describe how rain events can be modeled using RBFNN, and we also discuss possible solutions for modeling rain events using RBFNN in an efficient manner.

About the speaker:

Dimitrios Charalampidis received his Diploma in Electrical Engineering and Computer Technology (5 year studies) from the University of Patras, Greece, in 1996, and the M.S. and Ph.D. degrees in Electrical Engineering from the University of Central Florida, Orlando, Florida, in 1998 and 2001, respectively. In 2001, he joined the University of New Orleans, where he is currently an Associate Professor and the Associate Chair of the department. His research interests include image and signal processing, and applications in medical and remote sensing imaging.

Abstract:

Frequent patterns provide solutions to datasets that do not have well-structured feature vectors. However, frequent pattern mining is non-trivial since the number of unique pat-terns is exponential but many are non-discriminative and correlated. Currently, frequent pattern mining is performed in two sequential steps: enumerating a set of frequent patterns, followed by feature selection. Although many methods have been proposed in the past few years on how to perform each separate step efficiently, there is still limited success in eventually finding highly compact and discriminative patterns. The culprit is due to the inherent nature of this widely adopted two-step approach. This paper discusses these problems and proposes a new and different method. It builds a decision tree that partitions the data onto different nodes. Then at each node, it directly discovers a discriminative pattern to further divide its examples into purer subsets. Since the number of examples towards leaf level is relatively small, the new approach is able to examine patterns with extremely low global support that could not be enumerated on the whole dataset by the two-step method. The discovered feature vectors are more accurate on some of the most difficult graph as well as frequent itemset problems than most recently proposed algorithms but the total size is typically 50% or more smaller. Importantly, the minimum support of some discriminative patterns can be extremely low (e.g. 0.03%). In order to enumerate these low support patterns, state-of-the-art frequent pattern algorithm either cannot finish due to huge memory consumption or have to enumerate 101 to 103 times more patterns before they can even be found. Software and datasets are available by contacting the author.

About the speaker:

Kun Zhang received her PhD in Computer Science from Tulane University in 2006 and has been working as Assistant Professor at Xavier University of Louisiana since then. Her main research interests and experiences are in various areas of data mining and machine learning, such as probability estimation trees, extremely skewed distribution, ensemble methods, feature discovery, frequent pattern classification. She is particularly interested in developing simple, unconventional, but effective methods to solve challenging real-world problems. Her co-authored paper that uses "Randomized Decision Tree" to predict skewed ozone days received "the best application paper award" at IEEE ICDM06. The group she collaborated is the Crown Winner of IEEE ICDM08 Data Mining Contest. Her research is sponsored by NIH, BOR, LBRN, LCRC and Tulane Cancer Center.

Abstract:

In this talk, I will discuss the performance comparison of different algorithms for classification, estimation and filtering problems. Two information theoretic measures, namely, the empirical mutual information and the asymptotic information rate are proposed for simulation based performance evaluation and algorithm comparison. They can be used as a guideline for designing a practical procedure to measure the performance of different algorithms with limited computational resources. Other useful performance measures will be reviewed and their relation to the two new measures discussed. Several practical examples will be used to provide some insights on the inherent difficulty of algorithm ranking and the advantage of using the information theoretic measures for algorithm comparison.

About the speaker:

Huimin Chen received the B.E. and M.E. degrees from Department of Automation, Tsinghua University, Beijing, China, in 1996 and 1998, respectively, and the Ph.D. degree from Department of Electrical and Computer Engineering, University of Connecticut, Storrs, U.S.A., in May 2002, all in electrical engineering. He was a visiting researcher with Electrical and Computer Engineering Department, Carnegie Mellon University and a post doctorate associate with Physics and Astronomy Department, University of California, Los Angeles from June to December 2002. He has been with Electrical Engineering Department, the University of New Orleans since January 2003 where he is currently an associate professor. He held Air Force summer faculty fellowship in 2007. He has published more than 100 papers in peer reviewed journals and conferences. His research interests are in general areas of signal processing, data fusion, estimation theory, target detection and target tracking.

Abstract:

Modern storage systems offer high I/O performance and disk capacity at low costs. However two of the most critical issues that confront researchers and engineers in designing large scale storage systems are disk reliability and energy efficiency. Although numerous energy saving techniques are available for storage systems, most of them are not effective in reliability critical environments due to the lack of clear insight on the interplay between energy efficiency and reliability. In this talk, we present approaches to building energy-efficient and reliable large-scale storage systems. The primary focus of our research is to achieve the twin goals of maximizing reliability and minimizing energy consumption by incorporating these energy-efficient and reliable techniques to large scale storage systems. Experimental results using both synthetic and real world I/O traces have shown that energy dissipation in storage systems can be significantly reduced while maintaining high reliability for disks with a marginal degradation of performance.

About the speaker:

Xiao Qin is an Assistant Professor in the Department of Computer Science and Software Engineering at Auburn University. He received the B.S. and M.S. degrees in Computer Science from Huazhong University of Science and Technology, China, in 1996 and 1999, respectively. He received the Ph.D. in Computer Science from the University of Nebraska-Lincoln in 2004. Prior to joining Auburn University in 2007, he had been an assistant professor with New Mexico Institute of Mining and Technology (New Mexico Tech) for three years. He won an NSF CAREER award in 2009 to conduct research on multicore-based parallel disk systems. His research interests include parallel and distributed systems, real-time computing, storage systems, fault tolerance, and performance evaluation. His research is supported by the U.S. National Science Foundation, Auburn University, and Intel Corporation.

Abstract:

High throughput data acquisition technology has generated an enormous amount of molecular profiling data in which every bio-molecule is associated with a series of numerical measurements. These measurements are often replicated to combat the inherent noise in order to reliably discover patterns hiddenwithin data. Moreover, the measurements are sometimes incomplete which creates problems in the follow-up pattern discovery analyses. This leads to the need of new methods which can discover patterns from incomplete and replicated genome-wide measurements. Existing methods impute the missing values and estimate pair-wise correlation by averaging over the replicates or by computing the median. The ad hoc approaches do not exploit the information completely. The situation is even worse when the data is excessively noisy and incomplete, which is often the case for real-world genomic measurements. The replication mechanism of molecular profiling data falls under one of the two categories - informed case i.e. replication mechanism is known a priori or the blind case, when the replication mechanism is unknown. We develop an unified methodological framework to estimate correlation between a pair of genes that will work for both cases. We also consider the case of incomplete replicated data and impute the missing values using Expectation-Maximization (EM) algorithm. We present a comparative study of various approaches for pattern discovery from incomplete and replicated genome-wide data and show that the proposed method outperforms the existing methods, where we assess the robustness of the proposed method for data sets with up to 30% missing values.

About the speaker:

Lipi R. Acharya is a Ph.D. student in the Department of Computer Science at UNO since Fall 2008 and is working with Dr. Dongxiao Zhu. She holds a master’s degree in Mathematics from Indian Institute of Technology Madras,India and recently received a Ph.D. in Pure Mathematics from Indian Institute of Technology Kanpur, India. She has three publications in peer-reviewed Mathematical journals. She received various scholarships and awards for outstanding performance in studies as well as teaching to the undergraduates. Currenly she is a recipient of the Crescent City doctoral scholarship from UNO. Her current area of research is computational and statistical methods for analyzing genome-scale replicated data.

Abstract:

The Louisiana Optical Network Initiative, or LONI, is a state-of-the-art fiber optics network that runs throughout Louisiana, and connects Louisiana and Mississippi research universities to one another as well as National LambdaRail and Internet2. LONI connects Louisiana's major research universities, allowing greater collaboration on research that produces results faster and with greater accuracy. LONI provides Louisiana researchers with one of the most advanced optical networks in the country and the most powerful distributed supercomputer resources available to any academic community with over 85 teraflops of computational capacity. TeraGrid is an open scientific discovery infrastructure combining leadership class resources at eleven partner sites to create an integrated, persistent computational resource. Using high-performance network connections, the TeraGrid integrates high-performance computers, data resources and tools, and high-end experimental facilities around the country. Currently, TeraGrid resources include more than a petaflop of computing capability and more than 30 petabytes of online and archival data storage, with rapid access and retrieval over high-performance networks. Researchers can also access more than 100 discipline-specific databases. With this combination of resources, the TeraGrid is the world's largest, most comprehensive distributed cyberinfrastructure for open scientific research. This talk gives information about the high-performance computing resources available at LONI and TeraGrid. Topics includes cyberinfrastructure, hardware, software, user accounts / allocations, how to get started using these HPC resources, where to get help, free supercomputing training opportunities, and so on.

About the speaker:

Dr. Zhiyu Zhao is the UNO Computational Scientist for the LONI Institute. She received her B.E. (1997) and M.E. (2000) degrees in Computer Engineering from the Huazhong University of Science and Technology, China. In the year 2004 with a governmental scholarship she initialized her Bioinformatics research in DNA micro-array data analysis at the Politecnico di Milano University, Italy. In the year 2005 she started her Ph.D. study in the Department of Computer Science at the University of New Orleans, where she received a M.S. degree and a Ph.D. degree in Computer Science in 2006 and 2008, respectively. Her research field is Bioinformatics and her current research topics include protein 3-D structure alignment, searching similar structures in the Protein Data Bank, haplotype reconstruction from SNP matrices, and genome rearrangement based on non-breaking similarity.

Abstract:

Molecular profiling data is often replicated to accurately quantify the noise. Correlation-based pattern discovery from molecular profiling is one of most performed analyses in genomics research. Unfortunately, pairwise correlation estimates, which are the core of many unsupervised pattern discovery analyses, do not automatically accommodate replicated measurements. An extra step of averaging over replicates, either weighted, un-weighted or something in-between, is often followed. Averaging over replicates might wipe out patterns of low-magnitude and/or cancel out patterns of similar magnitude. Thus, it is not completely satisfactory. In this talk, we explicitly consider each replicate and their correlation structure. Simulation studies and real-world data analyses demonstrate the superior performance of our correlation-based pattern discovery methods. We develop a Graphic User Interface (GUI) software tools for use of our methods in biomedical research community. Finally we introduce a recent extension and a recent generalization of our methods for noisy and incomplete data. The proposed methodology applies to the replicated measurements generated in other scientific fields as well.

About the speaker:

Dr. Zhu received a M.A. degree in Statistics from the University of Michigan, Ann Arbor (2005) and a Ph.D. degree in Bioinformatics from the University of Michigan, Ann Arbor (2006). Before joining UNO's Computer Science Department, he has been with Stowers Institute for Medical Research since June, 2006, where he holds a primary appointment as Biostatistician. He also holds a secondary appointment with the University of Kansas Medical Center, Department of Biostatistics as an Assistant Professor. Dr. Zhu's methodology research interests fall in many aspects of bioinformatics and biostatisitcs. Dr. Zhu's bioinformatics research interest area includes: computational and statistical methods for analyzing genome-wide data, biological network inference, comparative genomics and biological sequence analysis. Dr. Zhu's applied statistics research interest area includes: applied multivariate analysis, statistical inference and computing, machine learning. In addition, Dr. Zhu actively collaborates with biomedical researchers to apply appropriate computational and statistical techniques to solve real-world biomedical problems.

Abstract:

Movement is crucial to the biological function of many proteins, yet crystallographic structures of proteins can give us only a static snapshot. The protein dynamics that are important to biological function often happen on a time scale that is unattainable through detailed simulation methods such as molecular dynamics as they often involve crossing high-energy barriers. We propose a novel interpolation method to study protein transitions between known crystal structures that does not extrapolate motion linearly and can therefore move around high energy barriers. We propose a novel test set of high-fidelity protein structures with at least three crystallographic on-pathway intermediate structures from the Protein Databank (PDB) to test with an objective criteria the biological significance of existing coarse-grained dynamics methods to study protein motion between two known conformations and of our novel interpolation method.

About the speaker:

Dahlia Weiss is a recent Ph.D. graduate from Stanford University.

Abstract:

We have developed a feedback algorithm for protein structure alignment that uses a series of phases to improve the global alignment between two protein backbones. The method implements a self-improving learning strategy by sending the output of one phase, the global alignment, to the next phase as an input. A web portal implementing this method has been constructed and is freely available for use at http://fpsa.cs.uno.edu/ with a mirror site at http://fpsa.cs.panam.edu/. Based on hundreds of test cases, we compare our algorithm with three other, commonly used methods: CE, Dali, and SSM. Our results show that, in most cases, our algorithm outputs a larger number of aligned positions when the (C_alpha) RMSD is comparable. Also, in many cases where the number of aligned positions is larger or comparable to the other methods, our learning method is able to achieve a smaller (C_alpha) RMSD than the other methods tested.

About the speaker:

Dr.Zhiyu is the UNO Computational Scientist for the Louisiana Optical Network Initiative (LONI). She received her B.E. (1997) and M.E. (2000) degrees in Computer Engineering from the Huazhong University of Science and Technology, China. In the year 2004 with a governmental scholarship she initialized her Bioinformatics research in DNA micro-array data analysis at the Politecnico di Milano University, Italy. In the year 2005 she started her Ph.D. study in the Department of Computer Science at the University of New Orleans. She received a M.S. degree in Computer Science in 2006. Her research field is Bioinformatics and her current research topics include protein 3-D structure alignment, protein structure searching from the Protein Data Bank, haplotype reconstruction from SNP matrices, and genome rearrangement based on non-breaking similarity.

Abstract:

Kat-5 is a 2005 Ford Escape Hybrid modified with the sensors and actuators needed for autonomous operation. It has an INS/GPS (Inertial Navigation System/Global Positioning System) from Oxford Technical Solutions and LIDAR units from SICK and Riegl. Kat-5 finished the 2005 DARPA Grand Challenge and placed fourth with a time of 7 hours, 30 minutes, only 37 minutes behind Stanley, the winner of the competition. Kat-5 is powered by almost 100% pure Java and runs both the Mac OS X and Linux operating systems. Kat-5 uses oscillating LIDARs and information from the INS/GPS unit to create a picture of the surrounding environment. This information is then used to build a path for the vehicle to follow. Kat- 5's primary electrical system, used to run the computers and drive-by-wire system, is powered by the standard electrical system of the vehicle while the 24-volt system, used to power the LIDAR sensors, is powered by six solar panels on the roof platform of the vehicle.

About the speaker:

Dr. Paul Trepagnier earned his B.S. in Computer Science from Clemson University, M.S in Computer Science from the University of New Orleans, and Ph.D. in Computer Science from Tulane University. Prior to Gray Matter, he spent 8 years developing enterprise Java applications for the Gray Insurance Company. His research interest is in the area of vehicle autonomy.

Abstract:

Risks permeate virtually every sector in existence today. An effective way of dealing with these threats is to employ a real time "spatially enabled" risk analysis system as described herein. With the ability to add and change complex risk algorithms, the Integrity Risk system can allow for continual improvement of risk analysis as more data and better modeling scenarios are developed. With the incorporation of existing technologies, real time data can be continually updated to the system giving a real time status necessary for real time risk analysis. Lastly, all of these technologies are available now and have been tested extensively in multiple environments. The systems can be readily modified to take them beyond oil and gas applications, thereby saving a tremendous amount of time and expense as opposed to developing a project of similar capabilities and magnitude from scratch.

About the speaker:

David Odom is currently a Client Manager for Shaw Group and manages over $200 million in programs.

Abstract:

Existing conventional databases provide us the facility to create and maintain indexed Databases for conventional data. The requirement for facilitating Un-conventional data, like Expressions and Digital Forensic Traces (DFT), in Conventional Databases is highly sought after in today’s world. The main contribution of our work is to provide the functionality of maintaining Expression in PostgreSQL Databases and creating multi-facet indexes for them to show that conventional databases can also be used to store un-conventional data and also their performance can be enhanced. The DFT’s provide us the same pattern of data as Expressions, where short comings of the conventional database can hinder us from using them in forensic analysis tools. We are going to discuss the possibility of using the same multi-facet indexes used in Expressions to be used in DFT’s. We are also going to discuss the possibility of creating In-Memory indexes for streamlining the process of querying the DFT Databases at a rapid pace.

About the speaker:

Salman Javaid received his Masters in Information Technology from NUST (National University of Scince and Technology) Pakistan and is currently a PhD student in Computer Science at University of New Orleans. His main area of research is databases and their usage in large-scale digital forensic methods.

Abstract:

Existing generic techniques for data fingerprinting are all based on some form of randomized feature selection, which was pioneered by Rabin in 1981. The main contribution of this work is a new, statistical approach for selecting fingerprinting features. Our method relies on entropy estimates and a sizeable empirical study to pick out the features that are most likely to be unique to a data object and, therefore, least likely to trigger false positives. We also gain precise control over the granularity and coverage of the feature set. The overall computational efficiency of the scheme allows its deployment during drive duplication and live network traffic monitoring. Finally, our empirical data pinpoints some common sources of false positives and has methodological implications in terms of evaluating the accuracy of data fingerprinting schemes in general.

About the speaker:

Dr. Vassil Roussev is an Assistant Professor in Computer Science at UNO. His main research interests are in high-performance computing and large-scale digital forensics methods. He has over 20 peer-reviewed publications in these areas.

Abstract:

Machine Learning Methods have been applied, previously, to nanopore signal analysis. A new nanopore detector platform is described where a modulatory molecule is introduced into the detector and where interactions with this modulator can be tracked with incredibly high precision. The signal processing needs for the nanopore transduction platform are even more demanding on the machine learning methods, and also require distributed speedup implementations of the core algorithms for practical use in the experimental environment. Novel implementations for HMMs and SVMs will be described in this setting.

About the speaker:

Dr. Stephen Winters-Hilt is an Assistant Professor in Computer Science at UNO and a Principal Investigator at the Research Institute for Children. His research interest include bioinformatics, cheminformatics, machine learning, and biophysics and he has over 30 peer-reviewed publications.

Abstract:

Digital forensic tools are being developed at a brisk pace in response to the ever increasing variety of forensic targets. Most tools are created for specific tasks – filesystem analysis, memory analysis, network analysis, etc. – and make little effort to interoperate with one another. This makes it difficult and extremely time-consuming for an investigator to build a wider view of the state of the system under investigation. In this work, we present FACE, a framework for automatic evidence discovery and correlation from a variety of forensic targets. Our prototype implementation demonstrates the integrated analysis and correlation of a disk image, memory image, network capture, and configuration log files. The results of this analysis are presented as a coherent view of the state of a target system, allowing investigators to quickly understand it. We also present an advanced open-source memory analysis tool, ramparser, for the automated analysis of Linux systems.

About the speaker:

Lodovico Marziale is a Ph.D. candidate at UNO’s Computer Science Department. His main research interest is in the area of digital forensics with an emphasis on high-performance tools for automatic correlation. He is also a member of the vaunted UNO Collegiate Cyber Defense Team.

Abstract:

We discuss how to tackle the detection of highly evolved, modern malware. We first show some worrisome anti-virus detection trends and developments. We then impart the results of two years' worth of investigating structural properties of modern malware. We conclude that given the empirical, practical, and fundamental theoretical limitations of traditional 'white-box' AV, it is time to move on: From predominantly byte sequence-matching white-box AV premised on classic Turing Machine models (computation-as-functions) towards high entropic defenses, black-box process modeling, as expressed by interactive computing models (computation-as-interaction). If time permits, we shall scare the audience with malware to come: RoQ, Satan, generalized side channel, IC, and the idea of quantum malware.

About the speaker:

Abstract:

The status of electricity deregulation in the United States will be presented and the use of Locational Marginal Pricing (LMP) in deregulated power markets will be described. The use of LMP in analyzing electricity pricing in a deregulated electric utility environment will also be presented. Using LMP, the proposed methodology for determining electricity prices in deregulated power markets is presented as an optimization problem that aims to minimize the total system production cost subject to physical and operational power system constraints. As building blocks in the modeling and analysis, NERC guidelines for regional generation and transmission planning are considered.

About the speaker:

Dr. Ittiphong Leevongwat earned his B.Eng. in Control Engineering from King Mongkut's Institute of Technology, Thailand in 1996, Master of Manufacturing Management from the Pennsylvania State University in 2002, and Ph.D. in Electrical Engineering and Computer Science from Tulane University in 2007 under Dr. Parviz Rastgoufard's advising. He worked in petrochemical and software industries. His research interests are power systems optimization and economics, energy efficiency, software development, and business intelligence. He is a Visiting Assistant Professor at the University of New Orleans.

Abstract:

Intracellular calcium has been found to be the central regulator of cardiac cell contractions. Modeling calcium dynamics is hence fundamental in understanding the excitation-contraction (E-C) coupling in cardiac myocytes. Both experimental and computational studies have revealed that the geometry of calcium-regulating organelles, such as transverse-tubules (T-tubules) and junctional sarcoplasmic reticulum (jSR), can significantly affect the calcium dynamics, suggesting that 3D structures from imaging data would provide more realistic modeling of the EC coupling mechanism. To bridge the gap between imaging and simulation, I shall present a chain of image analysis and geometric processing approaches to constructing multiscale models of ventricular cells. In particular, two imaging techniques are considered: one is the two-photon laserscanning microscopy imaging at the micro-scale, and the other is the electron tomography imaging at the nano-scale. Accordingly, 3D realistic geometric models of T-tubular systems and individual calcium release units (CRUs) are computed from the imaging data and represented by high-quality surface and volumetric meshes. Both stochastic (Monte-Carlo-based) methods and deterministic (PDE-based) numerical approaches are utilized to simulate calcium release, buffering, and diffusion in ventricular myocytes.

About the speaker:

National Biomedicine Computation Resource, University of California, San Diego

Abstract:

In this talk, I will discuss the recent advances in sparse representation, model selection and compressed sensing. I will start with some data streaming and dimension reduction examples and bring the unified view of variable/model selection and compressed sensing through an optimization framework. The geometric interpretation and implication to some interesting applications in computer science area will be briefly mentioned.

About the speaker:

Dr. Huimin Chen received the B.E. and M.E. degrees from Department of Automation, Tsinghua University, Beijing, China, in 1996 and 1998, respectively, and the Ph.D. degree from the Department of Electrical and Computer Engineering, University of Connecticut, Storrs, in 2002, all in electrical engineering. He was a post doctorate research associate at Physics and Astronomy Department, University of California, Los Angeles, and a visiting researcher with the Department of Electrical and Computer Engineering, Carnegie Mellon University from July 2002 where his research focus was on weak signal detection for single electron spin microscopy. He joined the Department of Electrical Engineering, University of New Orleans in Jan. 2003 as an assistant professor. His research interests are in general areas of signal processing, estimation theory, and information theory with applications to target detection and target tracking.

Abstract:

We are entering an exciting era of genomic research that is being driven by constantly evolving DNA sequencing technologies. In just the last decade, we have witnessed the completion of the human genome sequence (Human Genome Project 2001), nearly 1000 bacterial genomes, and over 165 eukaryotic species. The advent of DNA microarrays has significantly decreased the cost of many resequencing applications and a number of emerging ultra high-throughput sequencing technologies promise to move us ever closer to the much sought-after $1,000 genome. This talk will present recent work using DNA microarray technology that investigates human DNA replication timing. The focus will be on algorithms developed to classify, visualize, and simulate the process of DNA replication and its timing. Due to the sheer size of the full human genome, efficiency of our algorithms are a paramount concern in terms of both timecomplexity and memory requirements. The computational challenges involved with anticipated ultra high-throughput sequencing technologies provide an ideal realm for future research which can build on the same algorithmic techniques we use to analyze and process microarray data. include security and privacy, pervasive computing, computer networks, distributed systems, and statistical system analysis and design.

About the speaker:

Abstract:

A technique for synthesizing natural textures will be presented, with emphasis on quasiperiodic and structural textures. Textures are assumed to be composed of three components, namely illumination, structure, and stochastic. In contrast to previous techniques, a joint approach for handling the texture's global illumination, irregular structure, and stochastic component is used. Furthermore, the proposed technique does not produce verbatim copies in the synthesized texture. More specifically, a top-down approach is used for extraction of texture elements (textons) in which, in contrast to previous texton-based approaches, no assumptions regarding perfect periodicity are made. Results show that the proposed method is successful in synthesizing structural textures visually indistinguishable to the original. Moreover, the method is successful in synthesizing a variety of stochastic textures.

About the speaker:

Dr. Dimitrios Charalampidis received the Diploma degree in Electrical Engineering and Computer Technology from the University of Patras, Patras, Greece, in 1996, and the M.S. and Ph.D. degrees in Electrical Engineering from the University of Central Florida, Orlando, in 1998 and 2001, respectively. In 2001, he joined the Electrical Engineering department at University of New Orleans, where he is currently an Associate Professor and the Associate Chair of the department. His research interests include image and signal processing, pattern recognition, and neural networks.

Abstract:

A mobile ad-hoc network (MANET) contains a cluster of mobile hosts, which communicate with each other through the shared wireless bandwidth. MAC (Medium Access Control) protocols are designed and implemented to reduce transmission collisions and to improve bandwidth sharing efficiency. While many of the existing MAC schemes take the single-channel approach, there are several multi-channel MAC schemes, in which the shared bandwidth is divided into several subchannels. In this talk, we will introduce several of such multi-channel MAC schemes and discuss their advantages and disadvantages. These include Dynamic Channel Assignment (DCA) and Multi-Channel MAC (MMAC). DCA requires two transceiver interfaces on each node but no time synchronization is needed. MMAC, on the other hand, needs time synchronization and only one transceiver interface. We will also briefly introduce and discuss our proposed multi-channel MAC scheme and its potential benefits.

About the speaker:

Mr. Jingqi Wu received his master of M.E. in 2003 at Zhejiang University, China, and his bachelor of E.E. in 2000 at the same university. He had 3 years experience of instructor at Zhejiang University of Technology before he joined the Department of Computer Science at the University of New Orleans as a graduate research assistant. His current research is mainly focused in the areas of Medium Access Control (MAC) and information assurance in wireless communication.

Abstract:

In this talk, I will introduce a novel supernode caching scheme to reduce IP lookup latencies and energy consumption in network processors. In stead of using an expensive TCAM based scheme, we implement a set associative SRAM based cache. We organize the IP routing table as a supernode tree (a tree bitmap structure). We add a small supernode cache in-between the processor and the low level memory containing the IP routing table in a tree structure. The supernode cache stores recently visited supernodes of the longest matched prefixes in the IP routing tree. A supernode hitting in the cache reduces the number of accesses to the low level memory, leading to a fast IP lookup. According to our simulations, up to 72% memory accesses can be avoided by a 128KB supernode cache for the selected three trace files. Average supernode cache miss ratio is as low as 4%. Compared to a TCAM with the same size, 77% of energy consumption can be reduced. Other on-going work in my group will be also introduced in this talk.

About the speaker:

Dr. Lu Peng received his Bachelor and Master degrees in Computer Science and Engineering from Shanghai Jiaotong University, China. He obtained his Ph.D. degree in Computer Engineering from the University of Florida in Gainesville in April 2005. He joined the Electrical and Computer Engineering department at Louisiana State University as an Assistant Professor in August, 2005. His research focus on memory hierarchy system, multi-core interconnection, power efficiency and other issues in CPU design. He also has interests in Network Processor. He received an ORAU Ralph E. Powe Junior Faculty Enhancement Awards in 2007 and a Best Paper Award from IEEE International Conference on Computer Design in 2001.

Abstract:

Most of the attacks and fraudulent activities on the Internet are carried by malware. In particular, botnets have become the primary "platforms" for attacks on the Internet. A botnet is a network of compromised computers (or, bots) that are under the control of an attacker (or, botmaster). A botnet typically has tens to hundreds of thousands of bots, but some have several millions of bots. Botnets are now used for distributed denial-of-service attacks, spam, phishing, information theft, etc. With the magnitude and the potency of attacks afforded by their combined bandwidth and processing power, botnets are now considered as the largest threat to Internet security. In this talk, I focus on addressing the botnet detection problem in an enterprise-like network environment. I present a correlation-based framework for botnet detection that consists detection technologies already demonstrated in several systems (BotHunter, BotSniffer, BotMiner, and BotProbe). The common thread of these systems is correlation analysis (vertical correlation, horizontal correlation, and cause-effect correlation). I will mainly discuss BotHunter, BotSniffer and their corresponding correlation techniques/algorithms in this talk. These systems have been evaluated in live networks and/or real-world network traces, and the results show that they can detect real-world botnets with a very low false positive rate. These systems are starting to make an impact in the real-world. For example, there have been more than 6,000 downloads of BotHunter in the first five months after its public release. In addition, BotHunter is now being transitioned into products by several security vendors.

About the speaker:

Abstract:

In recent years, there have been growing interests in achieving perfect secrecy in physical layer of communication systems without restriction of computational complexity on eavesdroppers. It is a well known result from 70's and 80's that perfect secrecy is possible only when eavesdropper's channel is not as good as the channel between the legitimate transceiver in terms of receiver signal to noise ratios. In this talk, we focus on using feedback to achieve a strictly positive secrecy rate over an eavesdropped communication link when the eavesdropper's channel is less noisy than the legitimate receiver's channel. All channels are assumed binary and symmetric (BSC). The proposed novel scheme exploits the channel randomness inherent in feedback channels. We show that our feedback scheme not only achieves a positive secrecy rate when the eavesdropper's channel is better, but also improves the secrecy rate achievable by the Wyner's method when the eavesdropper's channel is worse. Some pre-/post-processing schemes to intentionally improve and degrade the equivalent wiretapper channels, respectively, are also proposed to further enhance the overall secrecy rate.

About the speaker:

Dr. Shuangqing Wei received his Ph.D. in 2003 at the University of Massachusetts, Amherst. He then assumed his current position as a tenure-tracked Assistant Professor in the Electrical and Computer Engineering Department of Louisiana State University, Baton Rouge. His current research interests are in the areas of information theory and communication theory, in particular their applications to wireless communication systems and networks.

Abstract:

Test Driven Development is a central tenet of Agile software development methodologies and is a powerful design technique for any software developer. The reasons for this are many: it encourages more modular, simpler code. It helps you avoid YAGNI (You Aren't Gonna Need It). It can be used to document a software system, by providing a specification and examples of its use. A test suite is also a necessary safety net for any serious refactoring work. In this session we'll talk about these aspects of TDD, how it fits with other Agile development practices, what makes up a good test, testing myths and introduce tools you can use to TDD.

About the speaker:

Muness has 11 years of experience developing software in a variety of domains using various tools for the job at hand. Muness has built resilient systems that can adapt to changing business needs. To that end he has been a proponent of agile practices, Ruby on Rails, Domain-Specific Languages (DSLs) and tagging. Muness received his B.S and M.S in Computer Science from Franklin University, Columbus, OH, and was also an adjunct professor at Franklin. He has spoken at both the No Fluff Just Stuff (NFJS) Software Symposium Series and The O'Reilly Open Source Convention (OSCON).

Abstract:

Traditionally, digital forensic analysis is performed "post mortem" on a disk, after it has been seized and the power cord to the computer has been unceremoniously yanked out of the wall. In recent years, two forces have been changing this practice. First, as the disk space on a typical system grows, and court orders for shutting down all machines become harder to obtain, it is getting more difficult to perform a complete "old school" analysis. And second, a new generation of tools has been created that look at volatile information that helps provide context to the static analysis. In this talk, we will first describe "old school" forensics, the important priciples behind the techniques, and the information they yield. Then we will present live forensics, the type of information that is available, and how it can be used, not as a replacement for, but in concert with static analysis to help investigators understand what happened and what is happening now to a system. We will conclude with some predictions on how the field will change based on current trends.

About the speaker:

Dr. Frank Adelstein is the technical director of computer security at ATC-NY in Ithaca, NY. He is the principal designer of a live forensic investigation product (marketed as Online Digital Forensic Suite and LiveWire Investigator) and has worked in the area of live investigation for the last 5 years. He has also been the principal investigator on numerous research and development projects including security, wireless networking, intrusion detection, and training. Adelstein is the vice-chair of the Digital Forensic Research Workshop, the premier workshop on research advances in the area of digital forensics and a co-authored of the the book Fundamentals of Mobile and Pervasive Computing (McGraw-Hill).

Abstract:

In the past, utilizing graphics processors for general purpose computing required re-engineering algorithms in terms applicable to graphics processors - vertices, fragments, shaders, etc... The added programming complexity was not worth the potential performance gains for many types of problems. Modern GPUs, however, have made significant advances on two fronts: their processing power has increased tremendously, and general purpose programming APIs for them have begun to appear. This has made them much more attractive for use as general purpose co-processors. In this talk, I will present the results of utilizing the power of modern GPUs (specifically, the NVIDIA G80 series) for sizable performance gains in computer forensics applications.

About the speaker:

Lodovico Marziale received a M.S. in Computer Science and a B.S. in Finance from the University of New Orleans and is currently pursuing a Ph.D. in Engineering and Applied Sciences. He is a research assistant in the department of Computer Science working on Next Generation File Carving. His research interests include Computer Security, Digital Forensics, and Parallel and Distributed Computing.

Abstract:

Creation of accurate and coherent surface models is vital to the effective planning and construction of flood control and hurricane protection projects. Typically, such topographic surface models are synthesized from Delaunay triangulations or interpolated raster grids. Although these techniques are adequate in most general situations, they do not effectively address the specific case where topographic data is available only as cross-section and profile centerline data, such as the elevation sampling produced by traditional hydrographic surveys. The hydraulic spline algorithm was developed to generate irregular two-dimensional channel grids from hydrographic cross-sections at any desired resolution. Hydraulic spline output grids can be easily merged with datasets of higher resolution, such as LIDAR data, to build a complete model of channel geometry and overbank topography. In testing, the hydraulic spline algorithm faithfully reproduces elevations of known input cross-section points where they exist, while generating a smooth transition between known cross-sections. The algorithm performs particularly well compared to traditional techniques with respect to aesthetics and accuracy when input data is sparse. These qualities make the hydraulic spline an ideal choice for practical applications where available data may be limited due to historic or budgetary reasons.

About the speaker:

Maik Flanagin's Ph.D Dissertation Defense.

Abstract:

Embedded systems are driving an information revolution with their pervasion in our everyday lives. For heterogeneous parallel embedded systems, I exploit the time and power optimization in various aspects. In high-level architecture synthesis, I address high-level architecture synthesis for real-time Digital Signal Processing (DSP) using heterogeneous functional units (FUs). With more and more different types of FUs available, same type of operations can be processed by heterogeneous FUs with different costs, where the cost may relate to power, reliability, etc. Furthermore, some tasks may not have fixed execution time. Such tasks usually contain conditional instructions and/or operations that could have different execution times for different inputs. Therefore, for such special purpose architecture synthesis, an important problem is how to assign a proper function unit type to each operation of a DSP application and generate a schedule in such a way that I can minimize the total costs while satisfying timing constraints with guaranteed confidence probabilities. We proposed several efficient algorithms to solve it. The experiments show that my algorithms can effectively reduce the total cost compared with the previous work.

About the speaker:

Professor Meikang Qiu received BE and ME in Engineering from Shanghai Jiao Tong University (SJTU). Before he came to U.S., Dr. Qiu worked at IBM HSPC and Chinese Helicopter R & D Institute for several years. Dr. Qiu obtained his MS and Ph.D. in Computer Science from University of Texas at Dallas (UTD) in 2003 and 2007, respectively. His research areas of expertise are Embedded Systems, Heterogeneous Sensor Networks, Parallel Computing, Data Mining and Fusion, Information Security, and Mobile Interface Browsing. He is an IEEE Senior member and ACM member.

Abstract:

We prove our identities daily by showing the access tokens we possess. Using a key to open a lock may be the most common form, which has about 4000 years of history since ancient Egypt. As one may access many locks, traditional master keys were designed to enable accessing multiple locks with a single key. Nevertheless, master keys are not widely used because of their security design limitations. Instead, people carry multiple access tokens for entity authentications, for example, keys, magnetic stripe cards, smart cards, RFID tags, and other tokens. In pervasive computing environments, entity authentications might be ubiquitously necessary. The management of access tokens and memorizing the token-lock relationships become overwhelming as the number of tokens increases. An intuitive question is how to achieve both the advantages of traditional master keys and multiple access tokens while avoiding their disadvantages. In this talk, I will present the Master Key, a novel entity authentication approach for pervasive computing environments.

About the speaker:

Feng Zhu received the Ph.D. in Computer Science and Engineering, the M.S. in Statistics, and the M.S. in Computer Science and Engineering from Michigan State University and the B.S. in Computer Science from East China Normal University. He is a program manager at Microsoft. His research interests include security and privacy, pervasive computing, computer networks, distributed systems, and statistical system analysis and design.

Abstract:

Background: The Baum-Welch training procedure for Hidden Markov Models (HMMs), an Expectation Maximization (EM) method, provides powerful tool for tailoring HMM topologies to data for possible use in knowledge discovery and clustering. A procedure recently proposed by Miklos, I. and Meyer, I.M., for implementing a memory sparse version of Baum-Welch training, has opened the possibility for an efficient distributed implementation of this algorithm. The original description of the technique has certain omissions that we amend, and experimental results are given. Results: The Baum-Welch algorithm has been implemented in memory proportional to the number of states in the HMM, and then thoroughly tested on number of data sets. We heavily modified the originally proposed algorithm to meet our objectives in data series analysis. Particularly, we reversed the originally proposed forward sweep to estimate the prior HMM state probabilities. We have corrected and rewritten the recurrence relation for the emission parameter estimations and extended it to parameter estimates of the Normal distribution. We describe our scaling procedure, necessary in all real implementations of the algorithm to prevent underflow. We also discuss the parallel implementation of the method and carefully described interpretation to the somewhat counter-intuitive recurrent relations. In this paper we also describe our approach to a linear memory implementation of the Viterbi decoding algorithm and demonstrated its linear memory use in an extended Duration Hidden Markov Model (DHMM) and spike detection topologies. Conclusions: Our speed-optimized Java implementation of linear memory Baum-Welch algorithm is available at http://logos.cs.uno.edu/~achurban. The proposed methods and implementation will aid sequence alignment, gene structure prediction, HMM profile training, nanopore ionic flow blockades analysis, and many other domains that require efficient HMM training with EM.

About the speaker:

Dr. Tchourbanov received his Engineer degree in automation and remote control from the Southern - Urals State University, M.S. in Computer Science from the University of Nebraska - Omaha and his Ph.D. in Computer Science (bioinformatics track) from the University of Nebraska - Lincoln. Dr. Tchourbanov was a one-year postdoctoral research associate in department of molecular biology at University of Wyoming before taking postdoctoral researcher position at Children's hospital of New Orleans. Despite substantial recent progress, gene structural prediction remains a challenging problem in bioinformatics. The importance of detailed understanding of gene splicing can be underlined by noting that ~10-15% of human genetic disorders come from mutations disrupting splice junctions. Dr. Tchourbanov's doctoral research focused on understanding of the constitutive splicing. Number of tools have been created to predict the human splice sites better, among them GIGOgene, Bayesian splice cite sensor and SpliceScan. As postdoctoral researcher at University of Wyoming he was evolved with study of correlations between Ka/Ks ratios and fidelity of phylogenetic reconstruction. Currently Dr. Tchourbanov actively collaborates with Dr. Eugeny Koonoin's evolutionary genomics research group at NIH/NCBI to bring study of splice sites signals to the evolutionary context. At children's hospital Dr. Tchourbanov conducts research on nanopore ionic current blockade signal processing, which offers a promising new way to analyze single molecule properties, with potential implications for DNA sequencing. He developed and evaluated number of HMM topologies for duration modeling and spike detection in ionic flow. Linear memory HMM learning was implemented by him recently which opened venue to efficient distributed implementation of an HMM.

Abstract:

In order to improve the throughput performance of Medium Access Control (MAC) schemes in wireless communication networks, some researchers proposed to divide a single shared channel into several sub-channels: one as control sub-channel and the others as data sub-channels. In this talk, we analyze and evaluate the maximum achievable throughput of a class of generic multi-channel MAC schemes that are based on the RTS/CTS (Ready-To-Send/Clear-To-Send) dialogue and on ALOHA contention resolution. Our analysis suggests some surprising conclusions.

About the speaker:

Dr. Jing Deng is an assistant professor in the Department of Computer Science (CS) at the University of New Orleans (UNO). He visited the Department of Electrical Engineering & Computer Science at Syracuse University from 2002 to 2004. He received his Ph.D. degree from School of Electrical and Computer Engineering at Cornell University, Ithaca, NY in January, 2002. Dr. Deng's research interests include Multiple Access Control (MAC), energy efficiency, Mobile Ad Hoc Networks (MANETs) and Wireless Sensor Networks (WSNs), key pre-distribution in WSNs, and information assurance.

Abstract:

As Web services grow in maturity and use, so do the methods which are being used to test and maintain them. Regression Testing is a major component of most major testing systems but has only begun to be applied to Web Services. The majority of the tools and techniques applying regression test to Web services are focused on test-case generation, thus ignoring the potential savings of regression test selection. Regression test selection optimizes the regression testing process by selecting a subset of all tests, while still maintaining some level of confidence about the system performing no worse than the unmodified system. A safe regression test selection technique implies that after selection, the level of confidence is as high as it would be if no tests were removed. Since safe regression test selection techniques generally involve code-based (white-box) testing, they cannot be directly applied to Web services due to their loosely-coupled, standards-based, and distributed nature. A framework which automates both the regression test selection and regression testing processes for Web services in a distributed, end-to-end manner is proposed. As part of this approach, special consideration is given to the concurrency issues which may occur in an autonomous and distributed system. The resulting synchronization method will be presented along with a set of algorithms which manage the regression testing and regression test selection processes throughout the system. A set of empirical results demonstrate the feasibility and benefit of the approach.

About the speaker:

Michael E. Ruth received his Bachelor of Science in Computer Science at the University of New Orleans in 2002. A year later, was awarded the Crescent City Doctoral Scholarship. In 2005, he received a Master's of Science in Computer Science at the University of New Orleans. Currently, he is a Research Assistant working towards a Doctorate in Engineering and Applied Science at the University of New Orleans. His research interests include Web services, distributed systems, and software engineering.

Abstract:

Searchable symmetric encryption (SSE) allows a party to outsource the storage of its data to another party (a server) in a private manner, while maintaining the ability to selectively search over it. This problem has been the focus of active research in recent years. In this talk, we present two solutions to SSE that simultaneously enjoy the following properties: 1. Both solutions are more efficient than all previous constant-round schemes. In particular, the work performed by the server per returned document is constant as opposed to linear in the size of the data. 2. Both solutions enjoy stronger security guarantees than previous constant-round schemes. In fact, we point out subtle but serious problems with previous notions of security for SSE, and show how to design constructions which avoid these pitfalls. Further, our second solution also achieves what we call adaptive SSE security, where queries to the server can be chosen adaptively (by the adversary) during the execution of the search; this notion is both important in practice and has not been previously considered. In addition, we also consider multi-user SSE. All prior work on SSE studied the setting where only the owner of the data is capable of submitting search queries. We consider the natural extension where an arbitrary group of parties other than the owner can submit search queries.

About the speaker:

Reza Curtmola is from the computer science department at John Hopkins University.

Abstract:

The RNA folding problem is analogous to the protein-folding problem in that it amounts to predicting the three-dimensional conformation of a large bio-polymer. RNA is only comprised of four nucleotides, compared to 21 amino-acids in proteins. However, the promiscuous hydrogen-bonding chemistry of RNA bases allows for a surprisingly large number of different types of intermolecular interactions. As a result, the RNA folding problem remains hard. Fortunately, significant advances in our ability to crystallize RNA has greatly increased the availability of known structures enabling the development of knowledge-based modeling approaches. We will present an automated approach that incorporates a statistical analysis of the geometry for all known RNA structures that allows for ab initio prediction of structure. Using commodity distributed computing grids, we generate coarse-grained decoy structures, which can then be filtered using low-resolution RNA-specific experimental data. This hybrid method allows us to make predictions of RNA structures with as low as a 6 A RMSD.

About the speaker:

Dr. Alain Laederach is a post-doctoral fellow in the Department of Genetics at Stanford University working in Russ Altman’s lab (also known as The Helix Group). His primary research interests are in using computational methodology to understand and predict biomolecular structure and dynamics. He is currently focusing on modeling RNA structure and the process of RNA folding. More specifically, he is interested in using limited experimental information to generate models of RNA folding intermediates and in determining the information content of these experimental results. He holds a B.S. from the Swiss Federal Institute of Technology, Lausanne, and a Ph.D. from Iowa State University. He is the recipient of a Damon Runyan Cancer Research foundation fellowship.

Abstract:

Modeling motion has a wide range of applications including robotics, computer animation, computer-aided design/virtual prototyping, computational biology, and drug design. Motion is particularly important for understanding many biochemical processes as it is often essential for functions. For example, some devastating diseases such as Alzheimer's and bovine spongiform encephalopathy (Mad Cow) are associated with the misfolding of proteins. Additionally, many biochemical processes such as gene regulation and catalysis are regulated by RNA folding. Despite the explosion in our knowledge of structural and functional data, our understanding of molecular movement is still very limited because it is difficult to measure experimentally and computationally expensive to simulate. In this talk, we describe a novel approach developed in our group for modeling biological molecular motions that is based on probabilistic road map methods (PRMs) originally proposed for robotics motion planning. We have successfully applied this strategy to protein and RNA folding, and ligand binding. After providing an overview of the general approach and it applications, we focus on our work in RNA folding. Our PRM-based framework uses a roadmap to approximate the folding energy landscape. By running folding simulations on the approximated energy landscape, we can study both global folding properties (such as folding rates) and microscopic features (e.g. folding of particular subsequences). We present comparisons with two experimental cases to show how we can use our method to predict kinetics-based functional rates of ColE1 RNAII and MS2 phage RNA and their mutants.

About the speaker:

Xinyu Tang is a Ph.D. candidate at Texas A&M University.

Abstract:

Expressions are used in a range of applications like publish/subscribe, website personalization, etc. Integrating support for expressions in a database management system (DBMS) provides an efficient and scalable platform for applications that use expressions. Additionally, in many cases these applications can benefit from support for uncertain data and expressions. Current DBMS lack such support. In this talk, I will discuss how expressions with uncertainty can be integrated in a DBMS and processed like other data stored in the DBMS. I will describe the underlying theory and implementation of UNXS (UNcertain eXpression System), a system that we have developed to handle uncertainty in expressions and data. We first develop a theoretical model to compare previous efforts in supporting uncertainty in DBMS and publish/subscribe systems. We propose new techniques for matching uncertain expressions to uncertain data. We then describe an implementation that integrates thissupport in the Postgresql DBMS.

About the speaker:

Prof. Nauman Chaudhry received his B.Sc. degree in Electrical Engineering from University of Engineering and Technology (UET) in Lahore, Pakistan in 1991. He received his Ph.D. in Computer Science & Engineering from the University of Michigan, Ann Arbor in 1998. He then worked at Oracle Corporation for 5 years. He is currently an Assistant Professor in the Dept. of Computer Science at the University of New Orleans. His research interests are in extending database management systems for advanced applications. Currently he is researching extensions to database systems to support uncertain data, and to support data streams.

Abstract:

As Web services grow in maturity and use, so do the methods which are being used to test and maintain them. Regression Testing (RT) is a major component of most major testing systems but has only begun to be applied to Web Services. The majority of the tools and techniques applying RT to Web services are focused on test-case generation, thus ignoring the potential savings of regression test selection (RTS). RTS implies that a subset of tests can be selected, rather than all tests, while maintaining some level of confidence about the system performing at least as well as the unmodified version post-modification. A safe RTS technique implies that after selection, the level of confidence is as high as it would be if we removed no tests. Since safe RTS techniques generally involve code-based (white-box) testing, they cannot be directly applied to Web services due to their loosely-coupled, standards-based, and distributed nature. An approach to automate safe RTS in an end-to-end manner will be proposed. As part of this system, some issues regarding autonomous, distributed, and more importantly concurrent modifications will be presented along with their solutions in a set of algorithms which manage the RT and RTS processes throughout the system. Lastly, an empirical analysis showing how the described mechanism performs in terms of selectivity, which will show a measure of cost savings.

About the speaker:

Dr. Michael E. Ruth received his Bachelor of Science in Computer Science at the University of New Orleans in 2002. A year later, was awarded the Crescent City Doctoral Scholarship. In 2005, he received a Master's of Science in Computer Science at the University of New Orleans. Currently, he is a Research Assistant working towards a Doctorate in Engineering and Applied Science at the University of New Orleans. His research interests include Web services, distributed systems, and software engineering.

Abstract:

The concept of a book is about to change. Although many will undoubtedly prefer to acquire, read, keep, and distribute literature in the way we have since Gutenberg, we will soon be in the era where most literature will be accessible electronically, and this may lead to many new approaches to the analysis of literature. Our current research explores the question of authentication of authorship using letter frequencies and classical pattern recognition techniques.

About the speaker:

Dr. Wayne Patterson is the Program Manager for the Office of International Science and Engineering at the National Science Foundation.

Abstract:

With the increase in the complexity of computer systems, security prevention measures are not enough to prevent all attacks. Intrusion detection systems (IDS) have become an integral part of computer security to detect attempted intrusions. Intrusion detection systems need to be robust against the attacks that are disguised to evade them. To analyze the robustness of network anomaly detection systems, we introduce a new class of polymorphic attacks, called polymorphic blending attacks (PBA). PBA can effectively evade a payload-based network anomaly IDS by carefully matching the statistics of the mutated attack instances to the normal profile. We present a formal framework for the analysis of PBAs. We show that in general, generating a PBA that optimally matches the normal traffic profile is a hard problem (NP-complete). However, the problem of finding a PBA can be reduced to the SAT or ILP problems so that solvers available in those domains can be used to find a nearoptimal solution. We also present a heuristic (hill-climbing) to find an approximate solution. We have experimented with our framework using the PAYL 1-gram and 2-gram anomaly detection system, and demonstrate that these attacks are indeed feasible. We provide some insight into possible countermeasures that can be used as defense against PBAs.

About the speaker:

Prahlad Fogla is a Ph.D. candidate in Computer Science specializing in Information Assurance at Georgia Institute of Technology.

Abstract:

Computational methods have become powerful enabling tools in modern structural biology, allowing the manipulation and prediction of protein structures with fairly high level of detail and control. The fields of de novo protein design and protein structure prediction are united by their common critical reliance on potential functions for modeling protein energetics. Computational aspects of protein design and structure prediction will be discussed, with highlights of some recent work on the generation of forcefields for automated refinement of close-to-native protein structure models (such as homology models).

About the speaker:

Dr. Summa is currently a NSF Fellow in Biological Informatics in the Department of Structural Biology at Stanford University Medical School.

Abstract:

We study routing misbehaviors in MANETs (Mobile Ad Hoc Networks) in this talk. In general, routing protocols for MANETs are designed based on the assumption that all participating nodes are fully cooperative. However, due to the open structure and scarcely available resources, node misbehaviors may exist. One such routing misbehavior is that some selfish nodes will participate in the route discovery and maintenance processes but refuse to forward data packets. In this talk, we analyze the adverse effects of such misbehaving nodes on data delivery performance and some prior solutions to this problem. We then propose a new technique termed 2ACK. The 2ACK scheme serves as an add-on technique for routing schemes to detect routing misbehaviors and to mitigate their adverse effect. The main idea of the 2ACK scheme is to send two-hop acknowledgment packets in the opposite direction of the routing path. We present the salient features and performance analysis of the 2ACK scheme in this talk.

About the speaker:

Dr. Jing Deng is an assistant professor in the Department of Computer Science at the University of New Orleans (UNO). He visited the Department of Electrical Engineering & Computer Science at Syracuse University from 2002 to 2004. He received his Ph.D. degree from School of Electrical and Computer Engineering at Cornell University, Ithaca, NY in January, 2002. His research website is at http://www.cs.uno.edu/~jing

Abstract:

In a peer-to-peer file distribution network, a large file is split into blocks residing in multiple storage locations. A peer node tries to retrieve the original file by downloading blocks from randomly chosen peers. We compare the performance of four storage strategies: uncoded, erasure coding, random linear coding, and random linear coding over coded blocks. We show that, in principle, random linear coding makes a better tradeoff between the storage requirement and decoding complexity. However, the sparsity of the file blocks is not fully exploited by random linear combinations of all original blocks. Motivated by the recent results from compressed sensing, we study the design tradeoff in random linear coding over coded blocks and propose an efficient decoding algorithm based on basis pursuit. We show that the minimum number of storage locations that a peer note has to connect to reconstruct the entire file with high probability can be significantly smaller than the total number of blocks that the file is broken into.

About the speaker:

Dr. Huimin Chen received the B.E. and M.E. degrees from Department of Automation, Tsinghua University, Beijing, China, in 1996 and 1998, respectively, and the Ph.D. degree from the Department of Electrical and Computer Engineering, University of Connecticut, Storrs, in 2002, all in electrical engineering. He was a post doctorate research associate at Physics and Astronomy Department, University of California, Los Angeles, and a visiting researcher with the Department of Electrical and Computer Engineering, Carnegie Mellon University from July 2002 where his research focus was on weak signal detection for single electron spin microscopy. He joined the Department of Electrical Engineering, University of New Orleans in Jan. 2003 as an assistant professor. His research interests are in general areas of signal processing, estimation theory, and information theory with applications to target detection and target tracking.

Abstract:

About the speaker:

Abstract:

Digital forensics is the science and discovering and preserving digital evidence, which exists on a wide variety of devices, from traditional computers to PDAs, voice recorders, copy machines, and cell phones. The traditional digital forensics investigative process involves stabilizing evidence, typically by making bit-perfect copies of evidence from a powered-down target, and then investigating the evidence using a variety of digital forensics tools. This investigation typically takes place in a laboratory environment, with the investigator using a single, powerful workstation. This simplistic approach will soon reach end-of-life, as investigators are pressed to provide faster case turnaround, to perform live investigations, to handle much larger targets, and to integrate evidence from many networked targets. The talk begins with a brief introduction to digital forensics investigation and then discusses "next generation" digital forensics research in the Department of Computer Science at the University of New Orleans. The aim of "next generation" digital forensics is to significantly improve the processes and tools available to an investigator, in all phases of the investigation. These improvements are partially motivated by overwhelming growth in the size of typical forensic targets; with 500GB hard drives available for under $200 and a correspondingly large appetite for storage of digital media, targets in excess of 1TB are common. An investigator trying to use traditional tools, powered by a single workstation, will be completely overwhelmed. The improvements in process and tools are also motivated by the need to investigate live, mission-critical targets, which cannot be taken down for investigation without disrupting business operations, and the need to effectively investigate networked targets. In order to modernize digital forensics investigation, the digital forensics research group at the University of New Orleans is creating new approaches and new software tools that allow investigative triage, where investigators may preview evidence before performing copies for further investigation, "live" forensics investigation of mission-critical computers, better file carving strategies, and the use of high-performance, cluster-based computing to speed investigations. This talk, rather than focusing on a single, specific research problem, quickly covers a number of ongoing projects, as well as some difficult, open problems, with the goal of (perhaps) revealing opportunities for collaboration. The speaker will provide homemade cookies in exchange for your rapt attention, polite criticism, and/or thoughtful questions.

About the speaker:

Prof. Golden G. Richard III, an experimental computer scientist and the third in a line of Golden's now numbering four, was born in 1964 in Jennings, LA. Golden's primary research areas are in digital forensics, specifically, next-generation digital forensics tools, network security, and operating systems internals. He is a GIAC-certified digital forensics investigator, co-founder of Digital Forensics Solutions, LLC, a private digital investigation firm, and technical advisor to the Gulf Coast Computer Forensics Laboratory. Golden is currently an Associate Professor in the Department of Computer Science at the University of New Orleans and director of the Networking, Security, and Systems Administration Laboratory (NSSAL). After Hurricane Katrina, he evacuated to Austin, TX, where he was a Visiting Associate Professor of Computer Science during 2005-2006. Golden completed his undergraduate degree in computer science at the University of New Orleans and holds the honor of being the first student ever to graduate with honors in computer science at UNO. His undergraduate advisor was the late Howard Evans. After UNO, he ventured north to Ohio State and braved the midwestern cold for almost 6 years, earning M.S. and Ph.D. degrees in computer science in 1991 and 1995, respectively. His Ph.D. advisor was Mukesh Singhal, who is now the Gartner Group Chair in Networking at the University of Kentucky. His Ph.D. dissertation was on process recovery mechanisms for message-passing and distributed shared memory systems. Golden set a record still unbroken among his peers by sending forth a single job application after graduation, to the University of New Orleans, where he was hired as an Assistant Professor in 1994. Golden is a notorious food snob, regularly shunning foods whose recipes contain the words "packet" or "can". When he's not engaged in benevolent hacking or dumping his brain contents into Powerpoint slides, he can be found cooking, gardening, consuming music (generally, jazz, reggae, trip hop, blues, or punk), or skateboarding.

Abstract:

As Web services grow in maturity and use, so do the methods which are being used to test and maintain them. Regression Testing (RT) is a major component of most major testing systems but has only begun to be applied to Web Services. The majority of the tools and techniques applying RT to Web services are focused on test-case generation, thus ignoring the potential cost savings of regression test selection (RTS). RTS implies that we only test the system under test (SUT) using a minimum amount of test cases to prove that the system performs at least as well as the unmodified version with some level of confidence. Safe RTS (SRTS) implies that we remove tests safely, thus not removing any modification-revealing tests. Since SRTS involves white-box testing, SRTS cannot be directly applied to Web Services due to their loosely-coupled, standards-based, and distributed nature. We will propose a framework in which we will automate SRTS to Web Services in an End-to-End manner.

About the speaker:

Michael E. Ruth received his Bachelor of Science in Computer Science at the University of New Orleans in 2002. A year later, he was awarded the Crescent City Doctoral Scholarship. In 2005, he received a Master's of Science in Computer Science at the University of New Orleans. Currently he is a Research Assistant working towards a Doctorate in Engineering and Applied Science at the University of New Orleans. His current research interest include Web services, distributed systems, and software engineering.

Abstract:

Efforts by others to sequence ssDNA from its translocation blockade signal with a nanopore detector have not succeeded. The key reason being the paradox of the translocation-based detection mechanism: as you make your channel fit tighter around the translocating molecule -- to get a better interaction to reveal the molecules identity -- the surrounding current signal is reduced, thereby reducing the overall detector sensitivity. What retains detector sensitivity along with strong interaction, while allowing for an even greater spectrum of molecular sizes and linkages to be examined, is blockade analysis for molecules interacting with the channel opening but NOT translocating. The non-translocational signals are rich with information only some of the time, and noisy and prone to drift all of the time, creating a computationally intensive adaptive learning problem. For non-translocational signal analysis there is now a clearer role for bifunctional molecules: one function being to enter and blockade the channel in an information-rich self-modulating manner, the other function, for binding, located on a non-channel-captured portion of the molecule that is free to bind or rigidly link to a larger molecule of interest. Recent results indicate that it is possible to directly track the bound versus unbound state of a molecule by this means. Sophisticated machine learning software has been brought to bear on this type of signal analysis in what is referred to as "channel current cheminformatics". Studies of antibody binding are being pursued both via the antibody itself being the bifunctional molecule, or via (unique) linkage to a dsDNA gauge at its carboxy terminus. Studies of antibody- and aptamer-based biosensing and immunological screening protocols are being developed. The prospects for directed molecular design and rapid immunological screening by this means could have a dramatic impact on medicine and drug discovery.

About the speaker:

Dr. Stephen Winters-Hilt; BS Physics and Electrical Engineering, Caltech, 1987; MS Applied Physics, Caltech, 1990; Visiting Researcher, Mathematics Institute, Oxford, 1994; PhD Theoretical Physics, Univ. of Wisconsin, 1997; PhD Computer Science, UCSC, 2003; Asst Prof., UNO, 2003-present

Abstract:

The cellular concept was introduced for wireless communication to address the problem of having scarce frequency resource. It is based on the sub-division of geographical area to be covered by the network into a number of smaller areas called cells. Frequency reuse in the cells far away from each other increases system's capacity. But at the same time, the cell boundaries prevent the channel resource of a system to be fully available for users. No access to the channel resources in other cell by the mobile host (or MH) limits the channel efficiency and consequently the system capacity. In this presentation, I will introduce a novel wireless system architecture based on the integration of cellular and modern ad hoc relaying technologies, called iCAR. It can efficiently balance traffic loads and share channel resource between cells by using ad hoc relaying stations (ARS) to relay traffic from one cell to another dynamically. This not only increases the system's capacity cost-effectively, but also reduces transmission power for MHs and extends system coverage. I will discuss an analytical model based on multi-dimensional Markov chains for performance evaluation. Our results show that with a limited number of ARSs and some increase in the signaling overhead (as well as hardware complexity), the call blocking/dropping probability in a congested cell as well as the overall system can be reduced. In addition, I will talk about several important design issues in iCAR, such as the ARS placement strategies involving a new performance metric called quality of (ARS) coverage (QoC), the signalling and routing protocols for establishing QoS guaranteed connections for IP traffic, as well as the ARS mobility management for adapting to the dynamic traffic load in the iCAR systems.

About the speaker:

Hongyi Wu received his Ph.D. degree in computer science and M.S. degree in electrical engineering from State University of New York (SUNY) at Buffalo in 2002 and 2000, respectively. He received his B.S. degree in scientific instruments from Zhejiang University in 1996.

Abstract:

About the speaker:

Abstract:

HMM feature extraction methods are sought for tracking individual molecular interactions via the changes in channel current blockade "signal" that results from the interaction with (and occlusion of) a single nanometer-scale channel established in a lipid bilayer. This involves developing noise tolerant feature identification (knowledge discovery) and feature extraction methods for single molecule studies, and intra-molecular interaction studies (reaction kinetics). To this end, HMMs are used for level identification, HMM/EM with boosted variance emissions are used for level projection pre-processing, and time-domain FSAs are used to parse the level-projected waveform. This provides a robust kinetic feature extraction formalism with a minimal amount of tuning. Classification, and/or clustering, of the HMM-based feature vector associated with a given blockade is then done by a variety of SVM implementations. HMM feature extraction results are described for five DNA molecules, as is the SVM classification performance based on those feature vectors. Results for two SVM approaches to multiclass discrimination are also described: (1) internal multiclass (with a single optimization), and (2) external multiclass (using and optimized decision tree). Each SVM approach encapsulates a significant amount of model-fitting information in its choice of kernel. In work thus far, novel, information-theoretic, kernels were successfully employed for notably better performance over standard kernels. Sometimes the data isn't clearly separable, making for poor discrimination. For such problems signal clustering may still provide useful information – to this end, novel, SVM-based clustering methods are also described.

About the speaker:

BS Physics and Electrical Engineering, Caltech, 1987; MS Applied Physics, Caltech, 1990; Visiting Researcher, Mathematics Institute, Oxford, 1994; PhD Theoretical Physics, Univ. of Wisconsin, 1997; PhD Computer Science, UCSC, 2003; Asst Prof., UNO, 2003-present

Abstract:

In this talk, I will give a comprehensive overview of statistical challenges with high dimensionality in diverse disciplines such as data mining and statistical inference. Then I will discuss the problem of variable selection and feature extraction using a unified framework: penalized likelihood methods. Issues relevant to the choice of penalty functions will be addressed. The desired properties in penalized likelihood framework and its connection to the structural risk minimization will also be discussed. The applicability of such a theory and method to diverse statistical problems will be illustrated with practical applications to a taxonomic problem and a stock selection problem.

About the speaker:

Huimin Chen received the B.E. and M.E. degrees from Department of Automation, Tsinghua University, Beijing, China, in 1996 and 1998, respectively, and the Ph.D. degree from the Department of Electrical and Computer Engineering, University of Connecticut, Storrs, in 2002, all in electrical engineering. He was a post doctorate research associate at Physics and Astronomy Department, University of California, Los Angeles, and a visiting researcher with the Department of Electrical and Computer Engineering, Carnegie Mellon University from July 2002 where his research focus was on weak signal detection for single electron spin microscopy. He joined the Department of Electrical Engineering, University of New Orleans in Jan. 2003 as an assistant professor. His research interests are in general areas of signal processing, estimation theory, and information theory with applications to target detection and target tracking.

Abstract:

Recent work on distributed RAM sharing has largely focused on leveraging low-latency networking technolo-gies to optimize remote memory access. In contrast, we revisit the idea of RAM sharing on a commodity cluster with an emphasis on the prevalent Gigabit Ethernet tech-nology. The main point of the paper is to present a practi-cal solution-a distributed RAM disk (dRamDisk) with an adaptive read-ahead scheme-which demonstrates that spare RAM capacity can greatly benefit I/O-constrained applications. Specifically, our experiments show that se-quential read/write operations can be sped up approxi-mately 3.5 times relative to a commodity hard drive and that, for more random access patterns, such as the ones experienced on a server, the speedup can be much higher. Our experiments demonstrate that this speedup is ap-proximately 90% of what is practically achievable for the tested system.

About the speaker:

BS, MS in Computer Science, Sofia University, Bulgaria MS, PhD in Computer Science, University of North Carolina, Chapel Hill Research interests: distributed and collaborative systems, digital forensics and security, human-computer interaction, software engineering. Recent publications (w/ Golden Richard): two book chapters on digital forensics and an article in Communications of ACM (02/06).

Abstract:

Ensemble methods provide a principled framework in which to build high performance classifiers and represent many types of data. As a result, these methods can be useful for making inferences about biometric and biological events. We introduce an ensemble method for combining multiple representations (or views). The method is a multiple view generalization of AdaBoost. Similar to AdaBoost, weak classifiers are independently built from each represetation. However, all data types share the same sampling distribution computed from the type of data having the smallest error rate. As such, the most consistent data type dominates over time, thereby significantly reducing sensitivity to noise. The method is applied to the problem of facial and gender prediction based on biometric traits. The new method outperforms several competing techniques including kernel-based data fusion, and is provably better than AdaBoost trained on any single type of data.

About the speaker:

Jing Peng is on the faculty of Electrical Engineering and Computer Science at Tulane University. He received the M.A. and Ph.D. degrees in computer science from Brandeis University and Northeastern University, in 1987 and 1994, respectively. From 1994 to 1995, he was a poc-doc fellow in the Computer Vision Laboratory at the University of California, Riverside. From 1996 to 1997, he served as a Senior Scientist at in the Machine Vision Department at Amherst Systems. From 1999 to 2001, he was on the faculty of Computer Science at Oklahoma State University. Dr. Peng's research interests include machine learning, bioinformatics, biometrics and image databases. He has served as the Guest Co-Editor of the Special Issue on Learning in Computer Vision and Pattern Recognition, IEEE Transactions on Systems, Man, and Cybernetics. He has served on the program committees for various international conferences. He has authored or co-authored over 80 technical publications in the areas of his interests.

Abstract:

In this talk, we present an efficient key pre-distribution scheme for Wireless Sensor Networks. Compared to previous schemes, our scheme substantially improves the resilience of the network. Our scheme exhibits a savvy threshold property: when the number of compromised nodes is less than the threshold, the probability that communications between any additional nodes are compromised is negligible. This desirable property lowers the initial payoff of smaller-scale network breaches to an adversary, and makes it necessary for the adversary to attack a large fraction of the network before it can achieve any significant gain. The background of the security problem and our approach will be presented in detail. We will also provide an in-depth analysis of our scheme in terms of network resilience and associated overhead.

About the speaker:

Dr. Jing Deng is an assistant professor in the Department of Computer Science at UNO. He received his Ph.D. degree in Electrical and Computer Engineering at Cornell University in 2002. Dr. Deng visited Syracuse University as a research assistant professor from 2002 to 2004. His research interests include Mobile Ad Hoc Networks, Wireless Sensor Networks, and Wireless Security.

Abstract:

About the speaker:

Abstract:

By "experimental mathematics"is meant an approach to gaining insight about mathematical phenomena that is realistically only possible when one is able to create and view a large enough body of data to be able to formulate hypotheses. This presentation will focus on four such examples: 1. Computation of determinants 2. An interesting sequence 3. The Goldbach conjecture 4. The Collatz conjecture

About the speaker:

Wayne Patterson Biographical Notes Wayne Patterson was born in Moncton, New Brunswick, Canada. He received the B. Sc. (Honours) in Applied Mathematics at the University of Toronto in 1966; M. Sc. in Mathematics also from Toronto in 1967; and the Ph. D. in Mathematics from the University of Michigan in 1971, in the field of differential topology. He also later received the M. Sc. in Computer Science from the University of New Brunswick in 1982. In 1968, he started the Michigan component of Project SEED, a national program of advanced mathematics for inner-city poverty and minority children. He taught the first Michigan classes in Project SEED, first in Ypsilanti and later in Detroit. He worked with then Senator Coleman Young to develop legislation, which eventually funded Project SEED statewide in Michigan. He also formed Project SEED, Incorporated, in 1970, and served as one of the founding directors of the corporation. He has continued to serve as a member of the Board of Directors since that time and has been the Chair of this Board since 1984. Dr. Patterson continued as Associate National Director of Project SEED, teaching in New Jersey, California, and Washington, DC, where he worked with congressional sponsors Senators Kennedy, Taft, Mondale and Magnuson to develop national funding for Project SEED. In his Project SEED teaching career, he taught classes in Lansing, Michigan, East Orange, New Jersey, Oakland, California, as well as those cities mentioned above. During this period, he also joined the mathematics faculty at both Princeton University and the University of California at Berkeley as a Post-Doctoral Fellow. In 1975, he returned to Canada to join the Government of Canada, serving as Special Assistant and Economic Advisor to the Secretary of State and later the Deputy Prime Minister of Canada. He returned to higher education as a professor of mathematics and later computer science at the Universit¨¦ de Moncton, the only French-language university in Canada outside of Qu¨¦bec. While teaching at Moncton, he also was a candidate for the House of Commons of Canada, and the Legislature of the Province of New Brunswick; and was twice elected as the national Vice-President of the Liberal Party of Canada. In 1984, Dr. Patterson was appointed Chair of the Department of Computer Science at the University of New Orleans, and in 1988 Associate Vice Chancellor for Research at that university. In 1993, he was appointed Vice President for Research and Professional and Community Services, and Dean of the Graduate School at the College of Charleston and the University of Charleston, South Carolina. In 1998, he was selected by the Council of Graduate Schools, the national organization of graduate deans and graduate schools, as the Dean in Residence at the national office in Washington, DC. His other service to the graduate community in the United States has included being elected to the Presidency of the Conference of Southern Graduate Schools, and also to the Board of Directors of the Council of Graduate Schools.In 1999, he was appointed Senior Fellow at the Council of Graduate Schools. Since the year 2000, he has been the Senior Fellow for International Programs and Academic Program Review in the Graduate School at Howard University, and Professor of Computer Science at Howard as well. In December 2003, he also became Associate Vice Provost for Research at Howard, in which office he continues to serve today. In his own research, Dr. Patterson has published more than 40 scholarly articles, and a leading textbook, Mathematical Cryptology (Rowman and Littlefield, 1986). He has been the principal investigator on over 20 external grants valued at over $6,000,000. As an administrator, he increased the external grants at New Orleans by over 400% and at Charleston by 250%. He also increased the number of degree-seeking graduate students at Charleston by over 70%. Among his other activities, Dr. Patterson also co-founded (with Savanah Williams) the North America-South Africa University Linkages Project, which has sponsored more than 40 exchanges with the Historically Black Universities of South Africa. He has also been a visiting faculty member or fellow at American University, Howard University, and the Oak Ridge National Laboratories. He has served on the boards of many community organizations, including Big Brothers, Chamber of Commerce, United Way, Cannon Street YMCA in Charleston (the oldest African-American YMCA in the US), Charleston Farmers¡¯ Market, and the College of Charleston Cougar Club. He has also competed for five years in the US Open Volleyball Championships, and has been official scorer for many college and professional baseball teams, including the Oakland As and New York Mets in the Grapefruit League.

Abstract:

In this talk, CDR Langley will give us a capabilities brief about SSC NOLA. He will also help us learn how the Navy is handling various parts of IT management, and information on several distributed network and application efforts that SSC NOLA are currently working with. He will also discuss the current and future initiatives at SSC NO that may lend to case studies, thesis research or collaborative work efforts.

About the speaker:

Scott Langley is a Navy Commander and an executive officer of SSC NOLA

Abstract:

Histology is the science of understanding the structure of animals and plants, and studying the functional implications of biological structures. In this talk, we present a statistical modeling approach to human histological image categorization. Texture features of the images are characterized by localized Gabor filters. The probabilistic distribution of the texture patterns from each category is approximated by a finite Gaussian mixture model. Expectation maximization (EM) procedure and minimum message length (MML) principle are used to perform density estimation and model selection, respectively. Componentwise EM and weak component annihilation are applied to avoid The drawbacks of the standard EM. Experimental validation is provided based on images from different organs and parts of the body.

About the speaker:

Dehua Zhao received the B.S degree from National University of Defense Technology, China, the M.S degree from Shanghai University, China, both in Electrical Engineering. He received the M.S degree in Computer Science from University of Wyoming. Dehua Zhao is a graduate student in the Computer Science Department at the University of New Orleans, advised by Prof. Yixin Chen. His current research interests include statistical classification and modeling, and machine learning for medical image analysis and retrieval.

Abstract:

The presence of physical obstacles and radio interference results in the so called ¡°shadow regions¡± in wireless networks. When a mobile station roams into a shadow region, it loses its network connectivity. In cellular networks, in order to minimize the connection unreliability, careful cell planning is required to prevent the occurrence of the shadow regions in the first place. In 802.11b/g wireless LANs, however, due to the limited frequency spectrum, it is not always possible to prevent a shadow region by adding another cell at a different frequency. Our contribution in this paper is to propose the alternate approach of tolerating the existence of ¡°shadow regions¡± as opposed to prevention in order to enhance the connection dependability. A redundant access point (AP) is placed in the shadow region to serve the mobile stations which roam into that region. Since the redundant AP operates on the same frequency as the primary AP, it does not constitute a separate cell. In fact, the primary and the secondary AP communicate to grant medium access to stations within the shadow region. We consider two configurations, which differ in how the two APs communicate with each other. In the first, the secondary AP is connected to the same distribution system as the primary AP. In the second, the secondary AP acts as a wireless forwarding bridge for traffic to/from the mobile stations in the shadow region to the primary AP. The paper outlines the details of how redundancy may be implemented by making enhancements to the basic 802.11 channel access protocol. To evaluate the dependability of the network under study, we present the reliability, availability and survivability analysis of the two configurations and compare them with the scheme with no redundancy. With numerical examples, we show that the redundancy schemes demonstrate significant improvement in connection dependability over the scheme with no redundancy.

About the speaker:

Dongyan Chen received the B.S degree from Southeast University, China, the M.Eng degree from Nanyang Technological University, Singapore, and the Ph.D degree from Duke University in Electrical and Computer Engineering. He is now with Computer Sciences and Computer Engineering Department, Xavier University of Louisiana. Dongyan Chen's research interests include discrete and fluid modeling techniques and their applications to communication system design and reliability/performance evaluation.

Abstract:

Relevance feedback is an attractive approach to developing flexible metrics for content-based retrieval in image and video databases. Large image databases require an index structure in order to reduce nearest neighbor computation. However, flexible metrics can alter an input space in a highly nonlinear fashion, thereby rendering the index structure useless. Few systems have been developed that address the apparent flexible metric/indexing dilemma. In this talk we present a kernel indexing method to try to address this conflict. The key observation is that kernel distances may be non-linear and highly dynamic in the input space but remain Euclidean in a kernel induced space. It is this linear invariance in the induced space that enables us to learn customized distance functions without changing the index. As a result, kernel indexing supports efficient relevance feedback retrieval in large image databases. We demonstrate the efficacy of the kernel indexing method using a large set of image data..

About the speaker:

Jing Peng is an Assistant Professor of Electrical Engineering and Computer Science at Tulane University. He completed his Ph.D. in the area of reinforcement learning from Northeastern University, 1994. Previously, he was on the faculty of Computer Science at Oklahoma State University. His research interests include machine learning and content-based image retrieval. He has served on the program committees of various international conferences. Currently, he is serving as the Guest Co-Editor of the Special Issue on Learning in Computer Vision and Pattern Recognition, IEEE Transactions on Systems, Man, and Cybernetics. He has authored a number of technical publications in the areas of his interests.

Abstract:

There exist quite a few criteria for penalty based model selection. Although they have various justifications for large sample problems, their performance under small or moderate sample size is unclear which hinders the development of model combination methods using the appropriate penalty term. In this paper, we assess the performance of seven model selection criteria based on linear regression models with unknown noise variance. We set the true data generation mechanism to be within the model set and outside the model set. In the latter case, soft model selection through multiple model fusion is proposed and its difference from Bayesian model averaging is highlighted. The penalty term used in each model selection criterion provides a natural link to estimate the model probability without assuming any prior knowledge of the unknown parameter. An important question is whether the estimated model probabilities are consistent when multiple models are fused for prediction or interpolation. We argue that strong consistency only holds under large sample regime while soft model selection can still be better than choosing a single model with small sample size. Our numerical results using different model selection criteria for polynomial fitting indicate that the conditional model estimator (CME) has the best performance in selection the correct model order and fusing multiple models for prediction and interpolation. The minimum description length (MDL) based criteria are next to CME and outperform Bayesian information criterion (BIC) and Akaike information criterion (AIC) significantly.

About the speaker:

Huimin Chen received the B.E. and M.E. degrees from Department of Automation, Tsinghua University, Beijing, China, in 1996 and 1998, respectively, and the Ph.D. degree from the Department of Electrical and Computer Engineering, University of Connecticut, Storrs, in 2002, all in electrical engineering. He was a post doctorate research associate at Physics and Astronomy Department, University of California, Los Angeles, and a visiting researcher with the Department of Electrical and Computer Engineering, Carnegie Mellon University from July 2002 where his research focus was on weak signal detection for single electron spin microscopy. He joined the Department of Electrical Engineering, University of New Orleans in Jan. 2003 as an assistant professor. His research interests are in general areas of signal processing, estimation theory, and information theory with applications to target detection and target tracking.

Abstract:

We introduce the notion of width bounded geometric separator, develop the techniques for its existence as well as algorithm. Combining it with our new local binding method, we improve the exact algorithms for a large class of NP-complete problems, and also obtain the first sub-exponential time algorithm for protein folding problem in the HP-model. We obtain a 2^{O(\sqrt{n})} time exact algorithm for the disk covering problem, which seeks to determine the minimal number of fixed size disks to cover n points on plane. Applying our separator to a class of NP-hard problems on disk graphs, we also greatly improve the exact algorithm for maximum independent set problem on disk graph to 2^{O(\sqrt{n})}-time from the previous n^{O(\sqrt{n})}. For a constant w>0 and a set of points Q on the plane, an w-wide separator is the region between two parallel lines of distance w that partitions Q into Q_1 (in the left side of the region), S (inside the region), and Q_2 (in the right side of the region). If the distance is at least one between every two points in the set Q with n points, called 1-separated set, there is an w-wide separator that partitions Q into Q_1,S and Q_2 such that |Q_1|,|Q_2|\le (2/3)n, and |S|\le 1.2126w\sqrt{n}.

About the speaker:

Bin Fu is an assistant professor in the Computer Science Department at the University of New Orleans. His research interests include Bioinformatics algorithm and computational complexity and security. He received a B.S. and M.S. degree in computer science from Wuhan University, P.R.China, and a PhD in computer science from Yale University at New Haven. He worked as both software and hardware engineers in California before he joined UNO computer science department.

Abstract:

Spatialization of text documents has received considerable attention in recent years, as it promises to make high-dimensional structures more readily accessible to the human cognitive system. One particularly promising approach lies in the fusion of geographic metaphors and cartographic principles with techniques developed in information science and computer science, so that decidedly map-like visualizations of non-geographic information can be derived. This presentation highlights some recent results of this endeavor, with particular focus on how text processing, neural computing, and cartographic design can be combined toward knowledge domain visualization.

About the speaker:

André Skupin is an associate professor in the Department of Geography at the University of New Orleans. His research interests include text document visualization, geographic visualization, cartographic animation and hypermedia, and cartographic generalization. He received a Dipl.-Ing. degree in cartography from the Technical University Dresden, Germany, and a PhD in geography from the State University of New York at Buffalo. He did graduate research with the National Center for Geographic Information and Analysis (NCGIA) and has worked in the GIS industry in the US, Germany, and South Africa.

Abstract:

The Virtual Synergy interface combines a three dimensional graphical interface with physical robots to allow for collaboration among multiple people, simulated software agents and physical robots. I employed Virtual Synergy in a variety of areas, for the AiBee robot art project, for the UNO urban search and rescue robot team, and for robot construction workers on Mars at NASA/JPL. To make human-robot collaboration effective and safe both the user interface and the autonomous robots must be designed to specifically handle the issues that arise when multiple human operators must interact with teams of robots to perform tasks, such as urban search and rescue. Human-robot interaction raises issues such as: * Appropriate inputs from humans * Adjusting the level of autonomy of the robots * Changing the distribution of roles and responsibilities between autonomous robots and humans * Modeling humans and their tasks * Facilitating human understanding of the goals, tasks and contexts of robots Implementing adjustable autonomy via roles can allow for better team collaboration during urban search and rescue (USAR). In this talk I present results from preliminary experiments with our mixed-autonomy USAR team. Also, I describe in detail Virtual Synergy, the human-robot interface that we developed, as well as the projects where it was employed.

About the speaker:

Dr. Sheila Tejada is currently an Assistant Professor in the Computer Science Department at the University of New Orleans. In 1993 she received her Bachelor of Science degree in Computer Science from the University of California, Los Angeles. She was awarded her Masters and Doctoral degrees in Computer Science from the University of Southern California in 1998 and 2002, respectively. Dr. Tejada has developed awarding-winning robots, such as, the robot YODA that took the silver medal at the AAAI office navigation robot competition, held in Portland, Oregon, and the robot soccer team DreamTeam that were the first world champions at the RoboCup International Robot Soccer Competition in Nagoya, Japan. Most recently, the UNO Robotics Team won a technical award for research on human-agent-robot interfaces at the AAAI/IJCAI Urban Search and Rescue Competition in Acapulco, Mexico and an Open Interaction award at AAAI 2004 as the Audience's favorite for the AiBee interactive robotic art project.

Abstract:

Efficient resource management techniques are of critical importance to the growing cellular/PCS industry. Unlike FDMA/TDMA system where capacity is fixed due to the frequency/time allocation, the capacity of a CDMA system is determined by the overall quality of signal-to-interference ratio (SIR). Compared to FDMA/TDMA, CDMA offers a relatively higher capacity with the same quality of services. With the growing demand of multimedia services on the internet, the wireless network must also expand their services to users from voice service to the multimedia services using advanced multiple access techniques. Due to its attractive features such as high system capacity, soft handoff, multi-path mitigation, interference suppression and low power transmission, DS-CDMA systems have become dominant in the 3rd generation multi-media wireless communication. In CDMA, power control is significant not only for prolonging battery life, but also for minimizing the problem caused by near-far effect. If all mobiles were to transmit at the same power level, the mobile closest to the base station will overpower all others. Therefore, the goal of power control is to make the signal strength received at base station (uplink power control) to be equalized for all mobile stations at a minimum possible level. Power control and handoff have been two significant problems for cellular networks. While both problems have received considerable attention of late, the problems are not often treated in a joint manner. In this talk, some preliminary simulation results on Call Admission Control of a multimedia CDMA system with consideration of both handoff prioritization and power control will be presented.

About the speaker:

Dr. Ming-Hsing Chiu has been a faculty member of Computer Science Department at the University of New Orleans since 1999. He received his MS and Ph.D. degree in Computer Science from the University of Central Florida., Orlando. His research interests include Wireless Networks, Internet Measurements and workload Characterization, and Interactive Distributed Simulation.

Abstract:

Current generation of digital forensic tools has reached a level of maturity that allows forensic investigators to find evidence without in-depth technical knowledge of the investigative target. At the same time, current tools exhibit a number of limitations with respect to performance, scalability, flexibility, and interoperability that are quickly becoming a bottleneck for digital forensics labs. In this talk, we will examine the specific technical issues that present interesting research problems and will present the current work in the digital forensics research group in the department. Our main focus is the development of an open framework that addresses current limitations and provides an extensible platform for including new forensic methods. To address performance and scalability concerns, we are developing the first (public) infrastructure for distributed processing of digital evidence. The goal is to turn the investigation of large targets into an interactive process by employing the resources of a (commodity) compute cluster. Furthermore, we are integrating this approach with our previous work for on-the-spot forensic investigation so that investigators can have a unified and flexible platform that can easily be deployed in different configurations. Another unique feature of our work the integration of multi-user capabilities that would allow a team of investigators to collaborate on solving a case and will eventually allow for much better use of scarce human expertise. Finally, we will discuss a number of advanced processing features that are still in the planning and design phases. The exciting (for us) part is that the computational needs of these new analytical tools dramatically exceed the capabilities of any single machine and are only become possible because of our cluster-based processing. Examples include content-based image analysis, speech/voice recognition, and various forms of cryptographic analysis.

About the speaker:

Vassil Roussev is an Assistant Professor of Computer Science at the University of New Orleans, where he has been a faculty since 2002. Dr. Roussev has M.S. and Ph.D. degrees in computer science from the University of North Carolina, Chapel Hill as well as B.S. and M.S. degrees in computer science from Sofia University (Bulgaria). His research interests include digital forensics, distributed multi-user systems, mobile devices, and software engineering pattern-based techniques, component- and service-based models, agile development methods.

Abstract:

The use of Field Programmable Gate Arrays (FPGAs) for computing in hybrid computer architectures and on applications not well suited to traditional processor architectures is now several years old, but fundamental problems have continued to plague those who would use such machines for computation. At the heart of the problem is and always has been that implementing an application has been a "hardware design" process and not a "programming" process. It now appears, however, that programming of a commercial reconfigurable computer is possible. We will describe the implementation of the DARPA High Productivity Computing Systems Discrete Mathematics Benchmarks on the SRC Computers SRC-6. In most instances, these benchmarks can be programmed in C, debugged, and optimized using techniques that are familiar to programmers of high-end machines. In the few instances in which the compiler tools have proven insufficient, hardware design methods and tools are used to implement libraries and functions called by the C program. The effectiveness of the SRC-6 is still limited due to hardware constraints, but because we can now readily program applications, we are able to explore fully the parameters of memory bandwidth, space/time tradeoff in the use of silicon, and the hardware/software codesign issues of balancing computation on the host processor versus computation on the FPGAs.

About the speaker:

Abstract:

Current research indicates that, to maximize safety and reliability while containing costs, process monitoring and control systems will increasingly rely on distributed networks of intelligent, highly autonomous sensors. Such systems will be capable of reporting quantitative/qualitative sensor readings, assessing system status, explaining current conditions, predicting likely system behaviors, and proposing corrective actions. The resulting information complexity threatens to exceed the capacity of human understanding, especially in high-risk situations requiring rapid human response. Ways for such a system to convey a wealth of information quickly and effectively have not yet been investigated. We describe the development of a prototype visualization of a rocket engine testing subsystem as a first step toward the development of a framework for /visual intelligent integrated health management systems./ Although the system we hope to develop will immediately apply to the propulsion testing being done at NASA's Stennis Space Center, the resulting framework is likely to be applicable to many similar activities involving complex, intelligent systems, including monitoring and control during all phases of space flight.

About the speaker:

Eduardo Kortright holds a Ph.D. in Computer Science from the University of Alabama. His doctoral dissertation work investigated applications of graph algorithms. Dr. Kortright's main research interests are currently in the field of medical imaging and scientific visualization. He was introduced to medical imaging applications during his work as a post-doctoral fellow and research assistant at the University of Alabama at Birmingham. Dr. Kortright is investigating computational methods for quantitative measurement of blood flow using MRI phase contrast angiography in conjunction with the Allegheny Singer Research Institute in Pittsburgh and the University of Alabama at Birmingham. Most recently, Dr. Kortright has begun a research project in the area of visualization of intelligent integrated health management systems in support of rocket engine testing.

Abstract:

Recent advances of electronic and computer technologies have paved the way for the proliferation of ubiquitous wireless networks. Wireless Ad Hoc Networks can be formed temporarily and quickly, without the requirement of any infrastructures. Due to resource constraints and the open nature of these networks, the efficiency and security problems become very important. In this talk, we will firstly discuss the hidden/exposed terminal problems at Multiple Access Control (MAC) layer in wireless ad hoc networks. We then present an efficient solution, the Dual Busy Tone Multiple Access (DBTMA) scheme, which uses the Request-To-Send (RTS) packet and two out-of-band narrow-bandwidth busy tones. Secondly, we will analyze the performance of MAC schemes that employ split-channel technique. Several other interesting studies related to efficiency and security of wireless ad hoc networks will be briefly discussed.

About the speaker:

Dr. Jing Deng is a Research Assistant Professor at Syracuse University.

Abstract:

With today's and tomorrow's wireless technologies, such as IEEE 802.11, BlueTooth, RF-Lite, and G3, mobile devices will frequently be in close, interactive communication. Many environments, including offices, meeting rooms, automobiles, and classrooms, already contain many computers and computerized appliances, and the smart homes of the future will have ubiquitous embedded computation. When the user enters one of these environments carrying a mobile device, how will that device interact with the immediate environment? We are exploring, as part of the Pebbles research project, the many ways that mobile devices such as Palm Personal Organizers or Pocket PC / Windows CE devices, can serve as a useful adjunct to the "fixed" computers in the user's vicinity. This brings up many interesting research questions, such as how to provide a user interface that spans multiple devices which might be in use at the same time? How will users and the system decide which functions should be presented in what manner on what device? Can the user's mobile device be effectively used as a "Personal Universal Controller" to provide an easy-to-use and familiar interface to all of the complex appliances available to the user? Can communicating mobile devices enhance the effectiveness of meetings and classroom lectures? This talk will provide our preliminary observations on these issues, and will include demonstrations of some of our systems that we are using to investigate them.

About the speaker:

Abstract:

Sophisticated and massively parallel experimental techniques have led to a substantial increase in the rate at which new biomedical information is discovered and reported. The predominant reporting mechanism of newly discovered biomedical information is journals and conference proceedings, both of which represent the information as unstructured natural language documents. Data mining and knowledge discovery from such a corpus is extremely challenging if possible at all. I will present a system called BioAnnotator, for identifying biological terms in scientific literature.

About the speaker:

Abstract:

With the rapid growth of the Internet and the falling price of storage devices, it has become increasingly popular to store texts, images, graphics, video, and audio in digital formats. This raises the challenging problem of designing techniques that support effective searching and navigating through the rich contents of large digital archives. As a part of this general problem, content-based image retrieval (CBIR) has been an active research area for more than a decade. It aims at efficient retrieval of relevant images from large image databases based on automatically derived imagery features. However, images with high feature similarities to the query image may be very different from the query in terms of semantics. This discrepancy between low-level content features (such as color, texture, and shape) and high-level semantic concepts (such as sunset, flowers, outdoor scene, etc.) is known as "semantic gap", which is an open challenging problem in current CBIR systems. In this talk, I will present my recent research on tackling this problem: (1) The first approach is motivated by an observation of human visual systems. Although color and texture are fundamental aspects of visual perceptions, human discernment of certain visual contents could potentially be associated with interesting classes of objects or semantic meanings of objects in the image. Therefore, we propose a fully automated algorithm that attempts to associate a semantic concept of images with objects contained in the images. (2) The second approach is a novel image retrieval scheme, CLUster-based rEtrieval of images by unsupervised learning (CLUE), which attempts to tackle the semantic gap problem based on a hypothesis that images of the same semantics are similar in a way, images of different semantics are different in their own ways. CLUE attempts to capture high-level semantic concepts by learning the way that images of the same semantics are similar and retrieving image clusters instead of a set of ordered images. Clustering in CLUE is dynamic. In particular, clusters formed depend on which images are retrieved in response to the query. Therefore, the clusters give the algorithm as well as the users semantic relevant clues as to where to navigate.

About the speaker:

Abstract:

Traditionally the main focus in database management systems (DBMS) research and development has been on achieving high performance. Consequently a large number of tuning knobs are provided in current systems to allow hand-tuning of these systems. However, the level of expertise required to use these tuning opportunities presents a barrier to end-customers' ability to effectively use these features. Recognition of this fact has led to significant efforts in making DBMS easier to tune - by making sub-systems self-tuning and by providing intelligent tuning tools. In my presentation, I will first present an overview of the complexity of the tuning task and approaches to simplify it. I will then focus on the specific problem of tuning physical database structures. I will present algorithms for physical structure tuning and describe systems that provide this functionality. Towards the end, efforts in tuning DBMS will be placed in context of the larger problem of building self-tuning and self-managing computing systems and avenues for future research would be identified.

About the speaker:

Abstract:

XML provides several favorable and powerful data modeling capabilities not present in other data models. For example, if we compare XML and relational models, XML allows the database designer to represent relationships using paths, which enables "easier" querying, XML can represent recursive relationships "better", and XML can represent union types and ordered relationships. Even conceptual models we use at present do not capture these features. In order to make effective use of the powerful data modeling features of XML, we need to extend existing conceptual models. In our talk, we will extend the Entity-Relationship (ER) model, and call it ERex. We examine how ERex can be used to come up with good XML models. This work also finds applications in translation between XML and relational models.

About the speaker:

Abstract:

We discuss the quality assurance issue in video streaming across the best-effort Internet and the optimal resource management issue in both wide-area multimedia overlay networks and cable broadband networks. To address the issue of random quality degradation in streaming across the best-effort Internet, we design a practical technique, named "staggered two-flow video streaming." In this framework, we first develop a novel application-aware transport protocol --- "controlled TCP (cTCP)", and further develop efficient application-aware flowcontrol and adaptation approaches to manage bandwidth sharing and interactions of streaming flows, by exploiting the inherent priority structure in videos, the storage space on proxy servers and the coarse-grain bandwidth assurance of VPN. Our prototype on FreeBSD systems demonstrated the efficacy of the technique in effectively protecting essential data and significantly reducing packet losses and storage requirements on proxy servers. To address the bandwidth contention issue in providing IP-based Video-On-Demand (VOD) service on Cable broadband Network (CBNs), we design an efficient video session scheduling technique, called "optimal full-sharing". This technique fully exploits the unique characteristics of CBNs to reduce the bandwidth consumption of video sessions sharing a cable channel of fixed capacity, thereby maximizing the number of simultaneous video sessions on the single channel. Furthermore, we develop adaptation algorithms which not only minimize the bandwidth consumption of video sessions but also significantly reduce service delays. In addition, we analyze the expected bandwidth and the session blocking probability of the system, and further design an efficient video assignment mechanism to maximize the system profit. At the end of this talk, I will briefly discuss my future research in network security, overlay networks, and multimedia networking.

About the speaker:

Abstract:

Given a collection of boolean spatial features, the co-location pattern discovery process finds the subsets of features frequently located together. For example, the analysis of an ecology dataset may reveal the frequent co-location of a fire ignition source feature with a needle vegetation type feature and a drought feature. The spatial co-location rule problem is different from the association rule problem. Even though boolean spatial feature types (also called spatial events) may correspond to items in association rules over market-basket datasets, there is no natural notion of transactions. This creates difficulty in using traditional measures (e.g., support, confidence) and applying association rule mining algorithms that use support-based pruning. We propose a notion of user-specified neighborhoods in place of transactions to specify groups of items. New interest measures for spatial co-location patterns are developed which are robust in the face of potentially infinite overlapping neighborhoods. We also design a family of algorithms to mine frequent spatial co-location patterns. Experimental results are provided to show the strength of each algorithm and design decisions related to performance tuning and benefits of enabling agents access to these knowledge bases. The target application is a real-time medical diagnosis system for responding to biological agents.

About the speaker:

Abstract:

The first part of my talk will present my research on security schemes for video transmission. That is, techniques that help ensure the privacy of the video data being transmitted. As a baseline comparison, I will show how passive attacks on streaming video are possible, backed up by experimental results from real video streams. I will then describe a prevention scheme that I proposed for MPEG that reveals no information to hackers eavesdropping on the stream. Extending this research, I will then present several efficient encryption and authentication mechanisms for proxy-based streaming aimed at minimizing security overhead at a proxy using end-to-end semantics. Finally, I will describe a lossless video coding framework based on layered coding and an encryption mechanism that reduces the total amount of data encrypted and allows adaptation to network resources. In the second part of the talk, I will present my work on similarity search databases. A similarity search query requests the item closest to the query point. Recently, a secure framework was proposed based on returning the score of a search aimed at protecting contents of database. We show that, in fact, the contents are not protected and can be discovered using repeated queries. For queries that return a score, we developed a framework that rejects queries that can potentially reveal information to users. A query history is stored and the amount of information is reduced by representing a query by a single number. Hashing is used to determine whether to answer the query or reject the query.

About the speaker:

Abstract:

About the speaker:

Abstract:

About the speaker:

Abstract:

The completion of a draft human genome sequence, along with several other complete genomes, is a great success for modern biology. To take advantage of these successes, new computational methods must be developed to store, integrate, annotate, disseminate and mine the data these projects have created. The required datasets are often spread over many resources, stored in disparate formats, and determined with different experimental protocols. In order to understand the underlying biology, these data must be integrated and presented in a way that can be useful to other researchers and automatic data mining methods. My previous research focuses on several computationally interesting hurdles encountered by researchers in computational biology. Examples based on data within the resource MutDB illustrate how to develop molecular models of disease. First, I present my results in the annotation of human genomic variation with data important for understanding its underlying biological function. Second, I summarize my integration of several genomic resources and the construction of an intuitive interface to the data. Third, I present algorithms that I developed for predicting which human mutations are most likely to participate in causing specific diseases and their application to the annotated genomic variation data. Finally, I discuss the presentation portal, MutDB, as a novel interface for other researchers to understand the underlying function behind human genetic variations.

About the speaker:

Abstract:

Machine Learning (ML) methods are central to recent advances in many ields, including Cheminformatics, Bioinformatics, Voice Recognition, Image Recognition, and Financial Analysis. ML methods provide powerful tools for structure identification, feature extraction and classification. Hidden Markov Models (HMM), for example have been an indispensable tool in bioinformatics, and Support Vector Machines (SVMs) provide a new, powerful, tool for classification (and clustering). This presentation briefly describes application of these tools, along with traditional power signal analysis and statistical methods, to channel current cheminformatics, microbial informatics, and mammalian informatics. In the first half of the talk I describe recent work in cheminformatics: classification of DNA molecules using an alpha-hemolysin channel current detector. Classification with better that 99% accuracy is obtained for DNA molecules that only differ in their terminal base-pairs. Tools from bioinformatics are used here: HMMs with EM are used for feature extraction, and a multi-class SVM hierarchy is used to classify. In the second half of the talk I describe work in bioinformatics: codon-void annotation and a generalized interpolating Markov model provide the basis for good gene prediction in prokaryotes and good identification of regulatory motifs in general. Gene prediction in eukaryotes involves more complicated structure (introns) and regulation (alternate splicing), and is being analyzed with a generalized interpolating HMM. SVMs provide a general use discriminator to help boost gene and regulatory motif identification, and are described here in application to expression analysis (vibrio cholerae) and transcription factor binding site identification (mammalian GPCRs).

About the speaker:

Abstract:

DNA computing uses the biological manipulations of DNA sequences to do the computation. Since we can let each DNA strand simulate Boolean circuit, a test tube containing a large number of DNA strands can run as a highly parallel computer. The maximum number of strands used is very important measure of DNA computing algorithm's complexity. This measure is called the volume used by the algorithm. From this study, we demonstrate an important connection between DNA computing and classical computing, thus enabling us to transform a large class of recursive algorithms into DNA computing algorithms. The research results in much improved the DNA computing algorithms for important problems such as 2-SAT, independent set and 3-colorability.

About the speaker:

Abstract:

The goal of image registration as a problem is the alignment of two or more images of the same scene or object. It is one of the most widely encountered problems in a variety of fields including, but not limited to, medical image analysis, remote sensing, satellite imaging, optical imaging, etc. Currently, mutual information (MI) based methods are widely accepted as the most effective way of handling rigid multi-modal image registration problems. But MI methods have their drawbacks, including being computationally intensive, lacking in robustness, and lack of ease in handling non-rigid motion. My dissertation, research is an attempt to provide an alternative in place of MI methods. In this talk, I will give a description of the three major components of our proposed algoritthm: (1) a local frequency image representation used to capture common information; (2) an ESD measure together with a modified Newton method to achieve a very fast implementation; and (3) L2E measure to handle the different fields of view problem.

About the speaker:

Abstract:

Distributed multi-user applications allow a group of geographically dispersed users to cooperate on a common task. Most commonly, this is achieved by the user with the abstraction of a shared artifact, such as a shared document. The main challenges in building infrastructures that support the development of such multi-user applications in object-oriented systems stem from the apparent conflict between data encapsulation and the need to share state among distributed object replicas. This conflict has led infrastructure designers to provide solutions that either emphasize automation over flexibility or vice versa. In this talk, I will present a novel approach to object sharing, along with its Java implementation, that reconciles the needs of encapsulation and sharing, thereby allowing automation to be achieved without compromising flexibility. The key idea behind the approach is to reuse the naming conventions already present in the object’s design to derive its logical structure from its public appearance. For that purpose, I will introduce a formal XML language that allows informal naming conventions to be concisely expressed as programming patterns that in turn are used to automate the development process. Since pattern analysis can be performed at run time, this approach significantly improves the reuse of existing code, such as single-user versions of the target applications. I will also discuss the general use of patterns and will provide examples of services, such as XML object serialization, UPnP user/system interface generation, and object testing that have been implemented in a generic fashion through patterns. I will conclude my presentation by outlining future directions in which my work can be extended both in and outside the domain of distributed collaboration.

About the speaker:

Abstract:

When integrating information from multiple websites, the same data objects can exist in inconsistent text formats across sites, making it difficult to identify matching objects using exact text match. We have developed an object identification system called Active Atlas, which compares the objects' shared attributes in order to identify matching objects. Certain attributes are more important for deciding if a mapping should exist between two objects. Previous method of object identification have required manual construction of object identification rules or mapping rules for determining the mappings between objects, as well as domain-dependent transformations for recognizing format inconsistencies. This manual process is time consuming and error-prone. In this approach, Active Atlas learns to simultaneously tailor both mapping rules and a set of general transformations to a specific application domain, through limited user input. The experimental results demonstrate that Active Atlas achieves higher accuracy and requires less user involvement than previous methods across various application domains.

About the speaker:

Abstract:

Grouping related data items has many uses in storage systems, and is a more effective use of access prediction than most proposed pre-fetching schemes to date. Our work attempts to provide a mechanism for constructing groupings of highly related files based on observed access patterns. Such groupings can be used for layout decisions to reduce access latencies of disk drives, or reduce the number of media changes in a tertiary storage library. Grouping information can also be used to drive the group-based management of caches, reducing demand fetches and allowing implicit pre-fetching. In this talk, we discuss file successor tracking, its use in file successor prediction, and its utility for the construction of highly related groups of files. We demonstrate how limiting the prediction problem to immediate successors can allow us to build very simple predictors that achieve very high performance with limited resource requirements. We then discuss how using these predictions to form file groups can result in better client and server caches.

About the speaker:

Abstract:

An important aspect of knowledge discovery regards efficient methods to find out how events repeat themselves as the time goes by. Research in this area has been fruitful, resulting in algorithms to discover periodic patterns and cyclic association rules, among other interesting phenomena. Unfortunately, missing from most research reported in the literature is the use of multiple time granularities (years, months, days and etc.) in such repeating patterns. Indeed, most human activities are closely related to time granularities and calendars. To capture such human activities, we propose to use calendar-based patterns that follow some user-given calendar schemas. An example of calendar schema is (year, month, day), which yields a set of calendar-based patterns of the form , where each di is either an integer or the symbol * that means "all." For example, <2000,*,16> is such a pattern, which corresponds to the time periods consisting of all the 16th days of all months in year 2000. Such calendar-based patterns represent all daily, monthly, and yearly patterns. In this talk I will discuss three situations: (i) Mining calendar-based patterns in a single-event sequence; (ii) mining calendar-based patterns in a multiple-event sequence; and (iii) mining calendar-based temporal association rules in a sequence of transaction sets. Concerning the last problem, we extend the well-known Apriori algorithm, and develop two optimization techniques to take advantage of the special properties of the calendar-based patterns. I will also present an implementation of mining calendar-based temporal association rules in anomaly detection of intrusions.

About the speaker:

Abstract:

This presentation describes a framework of employing blind source separation techniques in processing multispectral / hyperspectral images. The talk will cover two main topics. In the first part, I will present theoretical aspects of blind source separation, with emphasis on linear mixtures of sources. The method used is Independent Component Analysis (ICA), that based on assumptions of statistic independence of the sources and nongaussian nature of the data, successfully retrieves the original sources. In particular, I will show how mutual information and joint entropy are used in devising a gradient based iterative technique that produces the unmixing matrix. In the second part, I will show how ICA is employed in feature extraction and target detection for hyperspectral images. I will describe the devised theoretical model and algorithms and discuss them in the context of several issues specifically related to this type of data (size of the data, high redundancy, accuracy of sensors, etc.). In the end, I will show experimental results obtained by applying the obtained method on HYDICE (Hyperspectral Digital Imagery Collection Experiment) data sets.

About the speaker:

Abstract:

This talk describes a unified framework for constructing photo-realistic 3D models of cities from aerial stereo views. The talk will cover two main areas. In the first part of the talk, I will describe a new approach for rectification of stereo pairs. The method presented in this talk is a projective rectification technique, and hence does not require camera calibration or an explicit knowledge of the epipolar geometry, although the fundamental matrix can be recovered as a byproduct. The method combines some nice features from Loop-Zhang (1999) and Hartley (1999) and exploits planar projective transformation (collineation) to streamline the problem. In particular, I will show that the use of collination reduces the problem of estimating a pair of matching homographies to that of estimating a single optimal homography. In the second part of this talk, I will propose a scheme for constructing 3D models of cities based on fusing multiple cues in the framework of Bayesian decision theory. I will show that Bayesian decision theory allows for fusing different local cues represented in the form of binary decision maps, while imposing a global consistency constraint in the form of a prior probability distribution. These decision maps effectively reflect the prior knowledge on the local search window. In this sense the method shares some common ground with adaptive-window stereo techniques. However, I formulate the problem in a general and unified framework of fusion of mutiple cues using a Bayesian approach. Time permitting, I shall describe some of the issues related to generation of optimal meshes and ortho-textures for photo-realistic representation.

About the speaker:

Abstract:

The problem of mining association rules originally came from the need for analyzing large amounts of supermarket basket data in the early 1990s. Since then, mining association rules has become an important topic in the field of data minig and discovery. In this talk, we propose novel algorithms that are fundamentally distinct from the most well known algorithm, Apriori, in two aspects: First, in our algorithms, the database is scanned only once and will not be used any more. Second, there is no phase of candidate frequent item-set generation. We also consider the case where the size of the database exceeds the physical limit of the computers' main memory. We have implemented our algorithms, and our experiments have demonstrated significant improvements in performance over the Apriori algorithm.

About the speaker:

Abstract:

Micro-electro-mechanical systems (MEMS) are rapidly becoming part of everyday´s life through applications in telecommunications, in the automotive industry and in the life sciences, among others. The first part of the presentation will be dedicated to the description of our approach to the development of MEMS, that is, systems based on standard IC processes enhanced by selected additional technological steps. These enhancements include IC compatible etching and deposition. Examples to be discussed include gas flow, pressure, and fluid density microsensor systems. Determining material properties is an important part of our approach, since IC materials with electrical functionality are here being mis-used for unconventional purposes in thermal, mechanical, and other microdevices. Pushed by the need to know such properties for the reliable optimization of IC microsystems, we have developed techniques to determine the relevant electro-thermo-mechanical properties of IC thin films. This line of research will be illustrated with (i) the thermal van-der-Pauw method to measure the thermal conductivity of thin films, and (ii) the mechanical instability transitions of postbuckled micromachined membranes enabling to extract mechanical thin film properties. Finally, I will show in which direction we are pushing the use of standard IC materials beyond current thermal limits. This may open a field of new applications for IC-based MEMS.

About the speaker:

Abstract:

Broadcasting protocols reduce the cost of video-on-demand services by distributing more efficiently vidoes that are likely to be simultaneously watched by many viewers. Rather than answering individual customer requests, they broadcast the contents of each video according to a fixed schedule. We present a fixed-delay pagoda broadcasting protocol that requires all users to wait for a small fixed delay before watching the video they have selected. The protocol uses this delay to reduce the bandwidth required to transmit the first minutes of each video. As a result, it provides the lowest waiting times of all protocols using segments of equal duration and channels of equal bandwidth. In addition, its performance is not very far from the theoretical minimum. We also have show how to modify our protocol to restrict the set-top box receiving bandwidth to two or three times the video consumption rate.

About the speaker:

Abstract:

Data mining from relational databases has recently become a popular way of discovering hidden knowledge. Methods such as association rules, chi square rules, ratio rules, implication rules, etc. that have been proposed in several contexts offer complimentary choices in rule induction in this model. Other than inductive and abductive logic programming, research into data mining from knowledgebases has been almost nonexistent, because contemporary methods involve inherent procedurality which is difficult to cast into the declarativity of knowledgebase systems. In this paper, we propose an efficient logic-based technique for association rule mining from deductive databases, or the so called declarative knowledge, which does not rely on procedural concepts such as candidate generation. This development is significant as this empowers the users with the capability to explore knowledgebases by mining association rules in a declarative and ad hoc fashion.

About the speaker:

Abstract:

We introduce the "clique-separator graph" representation of a chordal graph, which provides significantly more information about the graph's structure than the well-known clique tree representation. We present fundamental properties of the clique-separator graph and additional properties when the input graph is interval. We then introduce the "train tree" representation of interval graphs and use it to decide whether there is a certain linear ordering of the graph's maximal cliques. This yields a fully dynamic algorithm to recognize interval graphs in O(n log n) time per edge insertion or deletion. The clique-separator graph may lead to dynamic algorithms for every proper subclass of chordal graphs, and the train tree may lead to fast dynamic algorithms for problems on interval graphs.

About the speaker:

Abstract:

A coalition is a collection of members who collaborate for some major purpose or goals by sharing and pooling resources. Resources may include information, assets, computing infrastructures, military capabilities and so on, and such resources are shared in a secure and controlled fashion. A Dynamic Coalition is a kind of coalition, in which the membership of the coalition and trust relationship between members could dynamically change, and the non-symmetric trust relationships may exist. This talk will presend the designs of an infrastructure for managing Dynamic Coalitions (from a computing point of view). We define the architecture and the semantics of "directives". Directives are the set of the resource access control rules based on the trust levels between each pair of the partners for members in DC to follow. We will show how to control access to resources, based on directives and how to compute transitive access control lists. We also introduce the notion of quasi-partnerships. Quasi Partnerships are potential relationships between members that do not have direct explicit partnerships between each other. To control the information sharing in a Dynamic Coalition, we have designed and implemented a Secure Information Management System -an information handling system that separates the file storage and file access authorization functions.

About the speaker: