student defenses

An Approach to Automated Agent Deployment in Service-based Systems
by Luping Zhu

Ph.D. Dissertation Defense

Date: May 16, 2008
Time: 1:30 p.m. - 3:30 p.m.
Location: BYENG 455

Committee Members: Dr. Stephen S. Yau (Chair), Dr. Hasan Davulcu, Dr. Hessam S. Sarjoughian, Dr. Guoliang Xue

ABSTRACT
In service-based systems, various capabilities are provided by different organizations as services, which normally reside on geographically distributed hosts belonging to multiple administrative network domains. These services are often utilized by distributed software agents, which collaboratively invoke proper services following specific workflows to achieve users' goals. These agents need to be deployed on hosts with respect to various deployment requirements which determine the hosts the agents need to be installed, activated, deactivated, updated, and uninstalled under different situations. Three important types of deployment requirements are considered in this dissertation: the access permission of the agents to the domains or hosts, the communication bandwidth between hosts and agents, and the reliability of the overall system. Deploying these agents manually is often time-consuming and error-prone and cannot effectively handle agents generated at runtime. Furthermore, in order to satisfy deployment requirements, an agent may need to be moved from one host to another under certain situation due to changes in environments. Hence, an agent deployment mechanism should support automated agent deployment and migration while satisfying deployment requirements for service-based systems. In this dissertation, an approach is developed to automated agent deployment in service-base systems to satisfy the deployment requirements provided by users, service providers, and domain administrators. By analyzing these requirements, deployment policies are specified and checked for consistency. Deployment constraints are generated from deployment policies with respect to network topology specifications and system configuration. Agent deployment plans are generated by solving the deployment constraints. An agent deployment plan is a mapping function between an agent and a set of hosts for determining on which host the agent needs to be deployed under various situations. Based on the agent deployment plans, controllers are synthesized and compiled to executable code to control the initial deployment and migration of agents at runtime.

Design, Development, and Evaluation of Coaptive Multimodal Interfaces
by Priyamvada Tripathi

Ph.D. Proposal Defense

Date: May 7, 2008
Time: 8:00 a.m. - 10:00 a.m.
Location: BYENG 365

Committee Members: Sethuraman Panchanathan

ABSTRACT
Multimodal interfaces have long held the promise of enhanced and effective human machine interaction. The ultimate goal of multimodal interfaces is to adapt to human activity allowing seamless exchange of information. This goal requires a coordinated development effort that incorporates a thorough understanding of human perceptual system in the design of interfaces. In this way, multimodal interfaces can not only supplement the current activities of humans but also expand their functionality to novel approaches in interaction. In recent years, multimodal interfaces have become synonymous with linguistic interaction. We propose that this category of multimodality does not fully exploit or include the human user’s capability of effectively interacting and communicating with the machine. Both concepts of semantic congruence and syntactic constraints can be dealt with only when we attempt to create interfaces that involve the human perceptual system in its design. In order to do so, we propose a novel framework for design, development and evaluation of perceptual interaction between human and the computer called coaptive interfaces. An empirically driven conceptual approach is used to derive guidelines, artefacts, and models for task analysis.

3D from Images Based on Rectified Views via Sequential Virtual Rotation
by Jin Zhou

Ph.D. Proposal Defense

Date: May 5, 2008
Time: 10:00 a.m. - 12:00 p.m.
Location: BYENG 365

Committee Members: Dr. Baoxin Li

ABSTRACT
Extracting 3-dimensional (3D) information from 2-dimensional images is a fundamental problem in computer vision. It can be used in various applications such as robotics, medical image, modeling, augmented reality and etc. Based on the number of views used, 3D from images can be classified into two categories: 3D from single view and 3D from multiple views. Traditional techniques such as stereo matching and Structure from Motion (SfM) belong to the latter approach, which can be fully automatic. However, these approaches are usually based on points matching and the results are 3D point cloud which is not good for modeling and the structure information is lost. In addition, the fundamental technique of point matching is very shaky due to unpredictable image conditions such as lack of texture, blurring, changing illumination and etc. On the other hand, 3D from single view can achieve accurate and complex modeling without concern of ill image conditions, while the systems usually require extensive interaction. The disadvantage of single view based modeling is that single view can only provide partial information of the scene, for instance, the occlusion problem. This proposal proposed a framework of extracting 3D information based on rectified views, which is able to avoid the problems mentioned above. The core technique is first rectifying images by sequential virtual rotation. After rectification, the imaged entities (points/lines/circles/planes or entire views) lie on the same plane or in the same Euclidean space, from which the 3D information of the entities can be directly extracted as well as matched across different views. This technique has great potential that can be used in lots of applications. This proposal presented three of them. First, stereoscopic rectification from uncalibrated view; Second, rapid modeling from single image and third, monocular view based Simultaneous Localization and Mapping (SLAM) for a mobile robot. Preliminary results demonstrate the feasibility and effectiveness of our approach.

System Support for Exploration and Expert Feedback in Resolving Conflicts during Integration of Metadata
by Yan Qi

Ph.D. Proposal Defense

Date: May 2, 2008
Time: 10:00 a.m. - 12:00 p.m.
Location: BYENG 455

Committee Members: Kasim Candan

ABSTRACT
A critical reality in data integration is that knowledge obtained from different sources may often be conflicting. Conflict-resolution, whether performed during the design phase or during run-time, can be costly and, if done without a proper understanding of the usage context, can be ineffective. FICSR (Feedback-based InConSistency Resolution) proposes a novel exploration and feedback-based approach to conflict resolution when integrating metadata from different sources. Rather than relying on purely automated conflict resolution mechanisms, FICSR brings the domain expert in the conflict resolution process and informs the integration based on the expert’s feedback. My dissertation work will mainly focus on (but not limit to) the following aspects: 1.(Query Execution Engine with Query Optimizer) Instead of relying on traditional model based definition of consistency (which, whenever there are conflicts, picks a possible world among many), FICSR introduces a ranked interpretation of the meta-data and statements about the metadata. This strategy not only enables FICSR to avoid committing to an interpretation too early, but also helps in achieving a more direct correspondence between the experts’ interpretation of the data and the system’s treatment of the available alternatives. In order to support FICSR’s ranking mechanism, we need implement highly efficient on-line query optimizer and query execution engine. FICSR starts with a matching/alignment phase, and integrates metadata from different sources. A challenge of processing the twig query q over the integrated metadata and obtaining ranked results is computationally expensive. Furthermore, while the query evaluation task is viewed in terms of ranked structural-joins along query axes, the monotonicity property, necessary for ranked join algorithms, is violated. Thus, an alternative, sum-max monotonicity property is explored and leveraged to develop a self-punctuating, horizon-based ranked join (HR-Join) operator to solve this problem. The query optimizer is able to generate a query evaluation plan to maximize the performance of collaborative HR-Join operators. 2.(User Feedback) In FICSR, the ranked interpretation leads to new opportunities for exploratory feedback for conflict resolution: within the context of a given statement of interest (i.e., a query): - a preliminary ranking of candidate matches, representing different resolutions of the conflicts, informs the user about the alternative interpretations of the metadata, while - user feedback regarding the preferences among alternatives is exploited to inform the system about the expert’s relevant domain knowledge. When modifying system status is allowed, the expert’s feedback is used for resolving not only the conflicts among different sources, but also possible mis-alignments due to the initial matching phase. Otherwise, an interactive process of query refinement is provided to reformulate the user query to help the user quickly locate the information of interest.

Exploiting Structured Knowledge for Segmentation, Enrichment, and Summarization of Data
by Jong Wook Kim

Ph.D. Proposal Defense

Date: May 1, 2008
Time: 10:00 a.m. - 12:00 p.m.
Location: BYENG 455

Committee Members: Professor K. Selçuk Candan

ABSTRACT
In our previous work, we explored structural mining of hierarchical content and identified how the hierarchical content grows through generalizations and specializations, which can be used in segmenting the message hierarchy into coherent units to facilitate indexing, retrieval, and ranking. We, then, built on existing works by developing keyword propagation technique in various application domains: structured web contents, an IS-A hierarchy and blogosphere. In this proposal, we extend the structured knowledge extraction and propagation techniques to find the relationship among attributes in relation database, which is a common problem on data mining. In relation database, the attributes of data objects are linked together via multiple entity-relationship links. The conditional probability of the attribute values, which are connected by links in a relation database, can be useful when identifying relationship across certain data attributes given context defined by other attributes. Consider a naive approach where multiple relationships are converted into a single relation, such as a universal relation. However, considering huge size of data in real applications, constructing and maintaining such a single table can be very costly. Recently, there have been growing research on relational data synopses (i.e., histograms, wavelets, join synopses and PRM) to approximately estimate the result size of complex query in a relation database. However, those existing approach cannot be directly applied to our problem since they do not consider all possible join combinations. Thus, we propose a technique for approximately discovering the conditional probability of the attribute values in a relation database containing multiple relationships. We propose a technique for effectively summarizing a relation graph. We then propose the attribute count propagation scheme to accurately estimate the conditional probability of the attribute values.

Semantics Inference and Efficiency Improvement for XML Keyword Search
by Ziyang Liu

Master's Thesis Defense

Date: April 25, 2008
Time: 9:00 a.m. - 11:00 a.m.
Location: BYENG 420

Committee Members: Yi Chen

ABSTRACT
Keyword search provides a simple and user-friendly query interface to access XML data in web and scientific applications, where users may not know XPath/XQuery, or the data schema is unavailable, complex, or fast-evolving. As a result, keyword search has recently attracted more and more research interests. In this thesis we study the problems with respect to keyword search on XML data. We study XML keyword search from two different but related perspectives: semantics and efficiency. Due to the inherent ambiguity of keyword search, it is often challenging to infer the semantics of a keyword query, identify relevant XML data fragments and compose meaningful search results. Consequently, two semantics-related problems are studied in this thesis: identifying relevant keyword matches, and identifying meaningful information to return to the user. For the first problem, since it is hard to directly evaluate the relevance of keyword matches due to the inherent ambiguity of search semantics, we investigate an axiomatic framework that includes two intuitive and non-redundant properties that an XML keyword search technique should ideally satisfy: monotonicity and consistency with respect to data and query. Then we propose a novel semantics for identifying relevant matches that satisfies both properties. For the second problem, we infer the semantics of the search and identify return nodes effectively by analyzing XML data structures and keyword search patterns. Besides, efficiency is also a very important aspect of processing searches. We study the problem of how to improve the efficiency of processing keyword search on XML. Inspired by the performance benefits of exploiting materialized views when processing structured queries, we investigate the feasibility and present a general framework for answering XML keyword search using materialized views. Then we develop an XML keyword search engine that leverages materialized views for query evaluation and maintains materialized views incrementally upon XML data update. At last, we present our ongoing project, which extends our search engine to handle graph structured XML data with IDs and IDREFs.

A Model-Driven Framework for the Specification, Grounding, and Execution of Semantic Web Services
by John Thomas Edward Timm

Ph.D. Dissertation Defense

Date: April 18, 2008
Time: 2:30 p.m. - 4:30 p.m.
Location: BYENG 420

Committee Members: Dr. Gerald C. Gannod, Dr. James Collofello, Dr. Susan Urban, Dr. Hasan Davulcu, Dr. Timothy E. Lindquist

ABSTRACT
The semantic web promises to bring automation to the areas of web service discovery, composition, and invocation. In order to realize these benefits, rich semantic descriptions of web services must be created by the software developer. Therefore, a steep learning curve and lack of tool support for developing such descriptions thus far have created significant adoption barriers for semantic web service technologies. This research makes four primary contributions: (1) the development of an approach that enables the specification of atomic and composite services, (2) the creation of model transformations that convert UML class and activity diagrams into OWL-S and other associated representations, (3) the development of software tools to facilitate the conversion, grounding, and execution activities, and (4) a demonstration of semantic web service technology and an evaluation of the approach. These contributions together form a model-driven framework for semantic web service development.

A Parallel Approach for K Nearest Neighbor Search in Metric Space
by Jamieson Arthur Wayne French

Master's Thesis Defense

Date: April 18, 2008
Time: 1:30 p.m. - 3:30 p.m.
Location: BYENG 365

Committee Members: Charles J. Colbourn

ABSTRACT
Similarity searches are very common in spatial applications and databases. In particular the K nearest neighbor search in metric space has been researched extensively over the years, resulting in many different algorithms. It is useful in many applications such as (medical and satellite) image processing, database mining, shape similarities, and graph theory. Most applications require an exact determination of the K nearest neighbors of all elements, and need to perform the search within a specific amount of time. However, as the number of elements increases, the time can grow quadratically, creating bottlenecks in the applications. With the recent introduction of parallel architectures (such as the NVIDIA 8 series GPU), new avenues can be explored to perform such a search. This thesis proposes a new algorithm that exploits the GPU architecture to find the k nearest neighbors of all elements within a set, or all elements of one set within another. The proposed GPU algorithm presented here is based on a well known sequential algorithm that uses a search structure called Vantage Point Forests. The algorithm, once implemented on the GPU, is able to find the exact K nearest neighbors in times that are substantially lower than other methods tested. Experimental results also show that the proposed algorithm is able to handle a wide variety of data sets and metric functions.

Distributed Simulation Framework for Multi-Formalism Semiconductor Manufacturing Supply-Chain Systems
by Dongping Huang



Date: April 18, 2008
Time: 12:30 p.m. - 2:30 p.m.
Location: BYENG 210

Committee Members:

ABSTRACT
Supply chain networks such as a semiconductor manufacturing system exhibit high degree of structural and behavioral complexity. Simulation modeling concepts, approaches, and tools are the primary means for analysis and design of intricate relationships within various parts of a supply chain network. A fundamental barrier in developing rigorous simulation models of supply chain systems is the necessity of using inherently different kinds of models and simulators. This is because no single modeling and simulation framework can adequately represent, at a realistic level of detail, a semiconductor supply chain system with tactical (short-term) control and strategic (long-term) planning policies. Composition of disparate model types affords rigorous synthesis of complementary classes of simulation, control, and optimization models. We developed a framework using an approach called Knowledge Interchange Broker (KIB) for composing the distinct classes of Discrete Event System Specification (DEVS), Model Predictive Control (MPC), and Linear Optimization (LP) models. First, we employed the KIB model composability concept to support model composition specification for DEVS and MPC modeling approaches. We created a prototype realization of the KIB and used it to integrate the DEVSJAVA and MATLAB environments that support simulating discrete-event and control-theoretic models, respectively. We demonstrated the proposed approach to a scaled, but generalized and realistic, semiconductor supply chain system. The integrated prototype environment was used to validate and examine the benefits of the proposed approach for prototypical semiconductor supply chain systems. Then we introduced a Knowledge Interchange Broker for composing DEVS, MPC, and LP models. This novel KIB accounts for (i) message transformation, (ii) causal parallel execution, (iii) synchronization, and (iv) timing. We developed a distributed simulation environment using DEVSJAVA, MATLAB, and OPLStudio, a linear optimization environment. The details of the KIB structure and behavior specifications are realized in this prototype simulation environment. The resulting hybrid composition of DEVS, MPC, and LP model types offers a robust basis for modeling and simulating complex discrete-part systems and specifically semiconductor manufacturing supply chain systems.

Impact of Agile Methods on Risk Factors in Mission Critical Systems
by Sindhuja Sadayandi

Master's Thesis Defense

Date: April 16, 2008
Time: 2:00 p.m. - 4:00 p.m.
Location: BYENG 598

Committee Members: Dr.James Collofello (Chair)

ABSTRACT
Agile techniques are becoming very popular in several non-critical applications due to their various benefits including early return on investment, inherent ability to accommodate rapidly changing requirements, lower defect rates, and enhanced customer satisfaction. However, organizations developing mission critical systems are still hesitant in adopting agile practices. This is mainly because of the very high safety requirements of mission critical systems as well as the risk factors related to agile practices. Although in general, agile practices can introduce new risks, proper choice of specific agile techniques can significantly reduce the severity or in some cases even prevent the inherent risks of mission critical systems and general software development. Therefore, selecting the suitable agile practices that are beneficial to the particular mission critical system based on risk factors is a critical requirement. A hybrid approach would be more beneficial for mission critical systems, in which mission critical organizations can tailor the agile practices based on their unique environment and project needs, and successfully integrate them with the existing practices to increase product quality. This thesis describes the development of an agile risk analysis framework. The agile risk analysis framework can be easily used by project leaders performing risk analysis to select the appropriate agile techniques that are most suitable for their particular project. The framework is based on a set of rules to assess the risks associated with agile techniques. This framework is an extension of existing Expert COCOMO model to include the risk analysis of the best practices in agile software development.

Graceful Degradation in Server Overload Using Adaptive Content
by Henri Q. Naccache

Ph.D. Dissertation Defense

Date: April 15, 2008
Time: 9:00 a.m. - 11:00 a.m.
Location: BYENG 420

Committee Members: Gannod, Collofello, Lindquist, Urban & Gary

ABSTRACT
Current web application-level Quality of Service (QoS) frameworks resort to refusing requests when met with overload. The self-healing adaptive content framework (SHAC) we are proposing will allow dynamic web-based applications to adapt their output in order to minimize the effects of flash crowds and server overload without refusing requests. We present a methodology to create a SHAC system for existing web applications. An analytical Queuing Network model is created to predict the performance impact of the framework on existing systems. A PID controller-based autonomic manager is implemented to maintain response times below that of a preset service level agreement. Evaluation of the framework is shown through implementations in web services, a portal, an AJAX-driven systems and a blog server along with significant performance testing and QoS analysis.

Dual Pointer Computational Support for Developing Covariational Reasoning Skills
by Pooja P. Ambekar

Master's Thesis Defense

Date: April 14, 2008
Time: 1:30 p.m. - 3:00 p.m.
Location: BYENG 420

Committee Members: Dr. Winslow Burleson (Chair)

ABSTRACT
This thesis develops and evaluates an interactive educational tool: “The Mouse-based Coordinate Quantities Tool” that will help students to develop their ability for covariational reasoning. This tool is developed with a novel human computer interface that allows a user to use both hands to control two mice simultaneously on a single display. Covariational reasoning is defined as the cognitive activities involved in coordinating two varying quantities while attending to the ways in which they each change in relation to the other. Research has shown that covariational reasoning is essential to success in calculus and that most current curricula in this area have not been sufficiently effective in developing this ability. Currently, to our knowledge, no computational tools exist for this important part of mathematics learning. The tool developed in this thesis is based on a novel method designed in the Center for Research on Education in Science, Mathematics, Engineering and Technology of Arizona State University and implemented in their curricula, called “the finger tool”. This tool has been developed to enable students to envision three things: (1) a graph as being a collection of points; (2) the collection of points as being generated by keeping track, simultaneously, of two quantities whose values vary; and (3) that every point in a graph represents, at the same time, values of two quantities. This thesis develops a technologically enhanced tool, based on the “finger tool” that leads the user through a step-by-step interactive process to improve student’s experience of, familiarity, and understanding of covariation. To ensure a broad impact, the tool uses two standard mice instead of other input devices because they are widely available, deployable and affordable. The tool provides visualization and automated feedback to the learners in order to enable them to independently use this tool. This supports students in developing an understanding of the concept of covariation without the need for constant teacher support. User studies and interviews conducted with a group of 8 college algebra students, their instructors and a group of 8 freshmen or sophomores from various departments have shown that the mouse-based tool is effective, engaging and easy to use.

Content Based Mining of Query Replacements
by Hung V. Nguyen

Ph.D. Dissertation Defense

Date: April 14, 2008
Time: 10:00 a.m. - 12:00 p.m.
Location: BYENG 455

Committee Members: Dr. Hasan Davulcu (Chair)

ABSTRACT
Keyword searching is the most common form of product search on the Web. Most search engines do their text query and retrieval using keywords. The average keyword query length is under three words (2.2 words). Recent research found that 40 percent of e-commerce companies rate their search tools as ”not very useful” or ”only somewhat useful.” Further, a review of 89 sites found that 75 percent have keyword search engines fail to retrieve important information and put results in order of relevance; 92 percent fail to provide guided search interfaces to help offset keyword deficiencies, and seven out of 10 web shoppers were unable to find products using the search engine, even when the items were stocked and available. Online advertisers are eager to find all relevant popular search phrases to promote their products. However, features in natural language such as synonymy, relations of terms and perceptions of terms pose difficulties for advertisers. Vendors cannot anticipate all possible ways in which shoppers search for their products. The ultimate aim of search engines is to increase the number of relevant results and to reduce irrelevant results as much as possible. Many times, user’s search query may not be a perfect description of their information needs. Thus, for certain phrases, there is a semantic gap between the search phrase used and the way the corresponding matching products are described. A serious consequence of this gap is that the users are not satisfied. Thus there is a critical need to boost product findability by bridging the semantic gap that exists between search phrases and product descriptions. Closing this gap has the strong potential to translate web traffic into higher conversion rates and more satisfied user experience. Because of the nature of the problem and several potential applications e.g. Web advertising, vertical search, recommendation system, we want a solution that is domain independent which requires minimal users’ feedbacks. Therefore, we develop content-based solution to the problem. We propose a robust technique based on Markov Chain Monte Carlo method to model the dependence among terms selected from the keyword matched item descriptions. Based on this term dependence, we develop an algorithm for generating phrase definitions which are short text snippets. These definitions of the original query can then be used as alternative queries to improve the retrieval/matching accuracy. We provide experimental evaluations with two large datasets, i.e., real vendor catalogs and the TREC-8 dataset to demonstrate the efficacy of our approach. The results show that our approach can capture insights on the statistical relationship among terms and it produces definition rules with high accuracy without suffering from the \"query drift\" phenomenon.

Cross-layer Opportunistic Adaptation for Voice Communications over Wireless Ad Hoc Networks
by Suhaib Obeidat

Ph.D. Dissertation Defense

Date: April 9, 2008
Time: 10:00 a.m. - 12:00 p.m.
Location: BY455

Committee Members: Dr. Violet R. Syrotiuk, ChairDr. Charles J. ColbournDr. Martin ReissleinDr. Guoliang Xue

ABSTRACT
Recent years have witnessed fast growth in Voice over IP (VoIP) applications. With the rate at which wireless access points are spreading, VoIP over wireless is becoming increasingly important. The all-IP architecture refers to inter-operability of different networks within an IP framework. Supporting VoIP over ad hoc networks is part of realizing this all-IP goal. Voice has many applications over ad hoc networks whether these networks serve as extensions to infrastructure-based wireless networks or in stand-alone configurations. The wireless channel introduces many challenges related to its broadcast nature and temporal response variability. Ad hoc networks additionally suffer from resource scarcity, making supporting real-time applications over ad hoc networks a formidable task. Adaptation to varying channel conditions is essential to supporting applications in ad hoc networks. In addition, cross-layer design allows interactions among protocol parameters to be exploited to improve performance. This research tackles the problem of supporting VoIP over wireless ad hoc networks by proposing opportunistic adaptation within a cross-layer framework. Starting with the requirements of voice, parameters whose adaptation appears important to providing an acceptable quality are identified. The significance of the selected parameters is verified in two ways: through extensive simulations of adaptation over wireless links, and through statistical analysis. The insight gained about the parameters is integrated into an opportunistic cross-layer protocol with the goal of maximizing the multiplexing gain without compromising the quality achieved. To limit the overhead and complexity of the protocol, hop-to-hop-only adaptation is considered first. Then, end-to-end adaptation is incorporated in the protocol design, which works at a different time scale and has a broader view of network conditions. The performance of the refined protocol transporting is studied using actual audio traces in the simulation. The resulting output stream is compared with the original stream using objective audio metrics which are mapped into subjective measure scores when possible. Results indicate the critical role compression, packet-size, and modulation play in supporting voice over ad hoc networks. Header compression contributes to achieving acceptable to good voice quality even over long routes, under reasonably high load conditions and in the presence of mobility.

Free Viewpoint Video with Image-based Rendering
by Wenfeng Li

Ph.D. Proposal Defense

Date: April 9, 2008
Time: 1:30 p.m. - 3:00 p.m.
Location: BYENG 420

Committee Members: Committee Chair: Li, Baoxin

ABSTRACT
Free viewpoint video (FVV) is an innovation to conventional video technology by allowing user to choose his/her viewpoint across a scene. It has a variety of applications in 3D TV, filming and gaming industry. In this document, we propose a practical FVV system which uses image-based rendering technology to render a new view based on only a few input images/videos. All key components for this system are addressed and presented with latest research results. We first propose a geometry framework for virtual view specification based on uncalibrated cameras. A complete system is implemented in the work of our first stage to show the feasibility and effectively of proposed approaches. Preliminary results show that both the research topic and methodology is state of the art while challenge remains. Based on the discussion of this system, problems are addressed and left to be solved in the next stage. To synthesize a view, deterministic methods, such as view dependent texture-mapping are efficient but critically dependent on pre-requisite 3D data, obtaining which is still a hard problem. We propose that the view synthesis problem can be considered as Bayesian inference which treats the new view as an outcome evidenced by multiple input images. A new image can be assembled by exploiting the statistical property of natural images and the strong correlation among the input images as they are photographed from the same scene. We further study the application of Gaussian mixture model and conditional random field in the second part of our work. An introduction of a complementary component of the FVV system, multiple video coding (MVC), is in the last section and possibility of study on this direction is discussed.

SAS: Security, Anonymity, and Survivability in Wireless Sensor and Ad Hoc Networks
by Satyajayant Misra

Ph.D. Proposal Defense

Date: March 28, 2008
Time: 3:00 p.m. - 5:00 p.m.
Location: BYENG 455

Committee Members: Prof. Guoliang Xue (Chair)

ABSTRACT
This research work focuses on four areas in wireless sensor networks namely, anonymous communication among the wireless nodes, secure localization, robust and secure target tracking in the presence of malicious tracking anchors, and placement of relay nodes in a wireless sensor network to meet the connectivity, survivability, and coverage constraints. For each of the problems, we define the problem and describe the system model along with the underlying assumptions. Then we present our techniques to solve the problems. We make a note of our contributions and outline scope for future work in each of the area, which would result in the culmination of the corresponding doctoral thesis.

An Approach to Automated Agent Deployment in Service-based Systems
by Luping Zhu

Ph.D. Dissertation Defense

Date: March 27, 2008
Time: 10:00 a.m. - 12:00 p.m.
Location: BYENG 455

Committee Members: Dr. Stephen S. Yau, Chair

ABSTRACT
In service-based systems, various capabilities are provided by different organizations as services, which normally reside on geographically distributed hosts belonging to multiple administrative network domains. These services are often utilized by distributed software agents, which collaboratively invoke proper services following specific workflows to achieve users\' goals. These agents need to be deployed on hosts with respect to various deployment requirements which determine the hosts the agents need to be installed, activated, deactivated, updated, and uninstalled under different situations. Three important types of deployment requirements are considered in this dissertation: the access permission of the agents to the domains or hosts, the communication bandwidth between hosts and agents, and the reliability of the overall system. Deploying these agents manually is often time-consuming and error-prone and cannot effectively handle agents generated at runtime. Furthermore, in order to satisfy deployment requirements, an agent may need to be moved from one host to another under certain situation due to changes in environments. Hence, an agent deployment mechanism should support automated agent deployment and migration while satisfying deployment requirements for service-based systems. In this dissertation, an approach is developed to automated agent deployment in service-base systems to satisfy the deployment requirements provided by users, service providers, and domain administrators. By analyzing these requirements, deployment policies are specified and checked for consistency. Deployment constraints are generated from deployment policies with respect to network topology specifications and system configuration. Agent deployment plans are generated by solving the deployment constraints. An agent deployment plan is a mapping function between an agent and a set of hosts for determining on which host the agent needs to be deployed under various situations. Based on the agent deployment plans, controllers are synthesized and compiled to executable code to control the initial deployment and migration of agents at runtime.

DEVS/UML - A Framework for Executable UML Models
by Joe Mooney

Master's Thesis Defense

Date: March 26, 2008
Time: 10:30 a.m. - 12:00 p.m.
Location: BYENG 455

Committee Members: Dr. Hessam Sarjoughian, Chair

ABSTRACT
The Discrete Event System Specification (DEVS) formalism excels at modeling complex discrete event systems. A framework capable of simulating a DEVS model is presented via Unified Modeling Language (UML) state machines. A set of rules is enumerated for the creation of UML state machines. Adherence to these rules will result in a model that is both DEVS and UML compliant. The resultant UML state machine models are executable within a prototype framework developed in support of this paper and more generally within a DEVS simulation framework such as DEVSJAVA. This thesis proposes that such an approach to modeling in UML represents a significant improvement over alternative approaches since it enables earlier simulation and verification of a design. The thesis rejects elaborate schemes either in UML or through use of its profiles to render a suitable environment through which DEVS compliant models may be expressed. The thesis asserts that simple abstractions and conventions are sufficient and by themselves yield a modeling approach wherein the specifics of simulation are not burdensome and easily omitted from models as they are evolved towards design and implementation, and easily re-introduced again as additional simulation needs arise.

Modeling and Rendering Realistic Trees
by Deepali Bhagvat

Master's Thesis Defense

Date: March 24, 2008
Time: 1:30 p.m. - 2:30 p.m.
Location: BYENG 210

Committee Members: Dr. Peter Wonka

ABSTRACT
This thesis presents a shape grammar for procedural modeling of trees, that faithfully reproduces the botanical characteristics visible in nature. The shape grammar allows modeling of trunk and higher-level branches, as well as details of compound leaves without resorting to image-based approximation strategies. The prioritized grammar rules illustrate the development of hierarchical and detailed tree models. This thesis introduces a novel idea of using implicit equations to ray trace tree surface as an alternative to triangle meshes. Each atomic shape in the tree model is an individually ray-traced conical frustum. This method yields a tree surface with correct behavior at the joins between adjacent shape geometries. The thesis also presents a new algorithm for relief mapping the ray traced tree surface. Relief mapping provides additional realism by closely imitating the irregular bark structure found on natural trees. The algorithm uses concentric conical frustums, represented by individual implicit equations, to perform relief mapping. This thesis combines these methods with the existing rendering methods such as shadow mapping to produce detailed and high quality realistic trees suitable for close-up views in games and movies at real-time frame rates.

Decentralized Resource Location and Multimedia Service Execution in Peer-to-Peer Network
by Gisik Kwon

Ph.D. Dissertation Defense

Date: March 21, 2008
Time: 10:00 a.m. - 1:00 p.m.
Location: BYENG 455

Committee Members: Ph.D. Dissertation DefenseProf. Kasim Selcuk CandanDr. Kyung Dong RyuProf. Partha DasguptaProf. Karamvir Chatha

ABSTRACT
Peer-to-Peer (P2P) computing has been gaining popularity due to the decentralized organization, high scalability, and the manipulation of abundant resources at the edge machines. Of many P2P applications, file sharing is the most ubiquitous with multiple implementations where users of P2P network can find the files of interest from other computers on the network and download them locally. Most of P2P file sharing systems search the files using either simple flooding or indexed routing with the complex structures, such as Distributed Hash Table (DHT). However, the file locating performance in these approaches turns out to be poor without a careful consideration of the machine heterogeneity and the network hierarchy. From this observation the preliminary works introduce two file placement and retrieval systems, PASS and BYPASS, for large-scale P2P networks. Both systems provide not only the efficient and deterministic file location, but also the self-organization and the load balancing by exploiting underlying network topology. This thesis extends the researches to the dynamic service composition in which multimedia workflow processing system like ARchitecture for Interactive Arts (ARIA) organizes workflow operators in purely decentralized manner. The traditional approaches involve a centralized data warehouse, permitting the extensive resource provisioning and replication to manage server load. Recent works propose to leverage routing paths in DHT to obtain a set of candidate nodes for service placement. However DHT is not designed to optimize workflow processing in overlay networks, yielding small candidate set sizes for service placement. This work proposes a noble Decentralized, Autonomous and Network-wide service workflow processing System (DANS), which explores DHT-based P2P substrate to lookup and publish the underlying peer information. This system achieves network adaptivity by using sender-initiated operator mapping, opposed to the receiver-initiation used in most current systems, eliminating pre-establishment of workflow networks. In addition DANS supports multi-constrained workflow execution. To this end, this thesis solves the nested chain workflow problem based on probabilistic models which select qualified service paths enforcing QoS metrics, while requiring low bandwidth consumption and sharing processing workloads.

Overlay Network Construction in Highly Decentralized Networks
by Melih Onus

Ph.D. Proposal Defense

Date: March 20, 2008
Time: 12:00 p.m. - 3:00 p.m.
Location: BYENG 455

Committee Members: Dr. Andrea Richa (Chair)

ABSTRACT
In recent years, highly decentralized networks such as wireless ad hoc networks and peer-to-peer(P2P) networks have emerged as the most interesting, challenging and innovation rich areas in computer networking. The nature of these networks require efficient, local, scalable and self-stabilizing algorithmic solutions. An important problem in highly decentralized networks is constructing overlay network on top of which one can efficiently provide services such as routing, broadcasting and information gathering in wireless ad hoc networks and routing, data search and redundant storage in P2P networks. In this proposal, we will study the problem of overlay network construction in highly decentralized networks. First, we will present our work on overlay network construction for wireless ad-hoc networks and then our work on construction of line-based P2P overlay networks will be presented.

TwigMF: Extending the XPATH Streaming Processor to Support the Following and Following-Sibling axes
by Karan Popli

Master's Thesis Defense

Date: March 19, 2008
Time: 9:00 a.m. - 10:00 a.m.
Location: Thesis Defense

Committee Members: Dr. Yi Chen (Chair), Dr. K. Candan, Dr. H. Davulcu

ABSTRACT
The semi-structured tree-based layout of XML provides a flexible mechanism for storing and retrieving data. As a result it has become the de-facto standard for portable data exchange over the Web. XPath is one of the dominant standards that is used to access data within an XML document much like SQL is used for querying and retrieval of data from a traditional relational database. XPath allows one to address different search regions within the XML document using the concept of axis. Broadly, the set of all the thirteen axes can be divided into the forward and reverse subsets, where the former set can be evaluated in polynomial time in a single scan of the document. Such an evaluation of an XPath query is said to come under the purview of stream processing. The set of forward axes comprises of four main axes: ‘Child’, ‘Descendant’, ‘Following’ and ‘Following-sibling’. A lot of research focus has gone towards providing efficient algorithms to support the first two axes while not much attention has been paid toward the evaluation of the latter two axes. The goal of this thesis is to provide a novel polynomial time and space algorithm to support the ‘Following’ and ‘Following-sibling’ axes over streaming XML data.