From fliggins@umbc.edu Tue Apr 17 14:55:14 2007 Date: Fri, 13 Apr 2007 11:56:40 -0400 From: Keara Fliggins To: csee-colloquium-out@csee.umbc.edu Subject: [csee-faculty-lecturer] [csee-colloquium-out] CSEE Research Review-Poster Abstracts [ The following text is in the "UTF-8" character set. ] [ Your display is set for the "US-ASCII" character set. ] [ Some characters may be displayed incorrectly. ] CSEE Research Review â^À^Ó Poster Abstracts Friday, May 4, 2007 Department of Computer Science and Electrical Engineering University of Maryland, Baltimore County (UMBC) Adam Anthony, MS Student in CS (Advisor: Marie DesJardins), Multi-Agent Planning and Learning Lab (MAPLE) Clustering Data with the Relational Push-Pull Model Relational data clustering is the task of grouping data objects together when both features and relations between objects are present. I present a new generative model for relational data in which relations between objects can have either a binding or separating eï¬^Àect. For example, with a group of students separated into gender clusters, a â^À^Üdatingâ^À^Ý relation would appear most frequently between the clusters, but a â^À^Üroommateâ^À^Ý relation would appear more often within clusters. In visualizing these relations, one can imagine that the "dating" relation eï¬^Àectively pushes clusters apart, while the "roommate" relation pulls clusters into tighter formations. I use simulated annealing to search for optimal values of the unknown model parameters, where the objective function is a Bayesian score derived from the generative model. Experiments on synthetic data demonstrate the robustness of the model. Additional results on a food web data set will also be presented. Jesus J. Caban, PhD Student in CS (Advisor: Penny Rheingans), VANGOGH Lab Texture-Based Feature Tracking for Effective Time-Varying Data Visualization Analyzing, visualizing, and illustrating changes within time-varying volumetric data is challenging due to the dynamic changes occurring between timesteps. The changes and variations in computational fluid dynamic volumes and atmospheric 3D datasets do not follow any particular transformation. Features within the data move at different speeds and directions making the tracking and visualization of these features a difficult task. We introduce a texture-based feature tracking technique to overcome some of the current limitations found in the illustration and visualization of dynamic changes within time-varying volumetric data. Our texture-based technique tracks various features individually and then uses the tracked objects to better visualize structural changes. We show the effectiveness of our texture-based tracking technique with both synthetic and real world time-varying data. Furthermore, we highlight the specific visualization, annotation, registration, and feature isolation benefits of our technique. For instance, we show how our texture-based tracking can lead to insightful visualizations of time-varying data. Such visualizations, more than traditional visualization techniques, can assist domain scientists to explore and understand dynamic changes. Richard T. Carback III, MS Student in CS (Advisor: Alan T. Sherman), Cyber Defense Lab A Hash-Based Secret Sharing Scheme Secret sharing protects secret information by requiring a group of users to be present to decode the secret. This feature is an important component of many systems that try to prevent individual user access to confidential information. Unfortunately, most secret sharing schemes rely on specific mathematical phenomena that would require a complicated implementation that would not be easy to verify for correctness. We present a scheme that relies on generic cryptographic hash and encryption functions to provide a threshold secret sharing scheme with a username and password from each user. Additionally, this scheme does not use a dealer, detects any cheating participants, requires only one storage device, and can be completely regenerated if that storage device is destroyed Akshay Java, PhD Student in CS (Advisor: Tim Finin), eBiquity Modeling Influence, Opinions and Structure in Social Media The Blogosphere provides an interesting opportunity to study social interactions. Blogs are a channel to express opinions, facts and thoughts. Through these pieces of information, also known as memes, bloggers influence each other and engage in conversations that ultimately lead to exchange of ideas and spread of information. We aim to characterize and model the Blogosphere to study the spread of influence, opinion formation and social interaction. Further, we propose a simple generative process that models creation and evolution of blogs. Using available blog graph and the proposed generative model, we hypothesize, compare and validate different approaches to modeling influence and opinion formation in social media. Alark Joshi, PhD Student in CS (Advisor: Penny Rheingans) Exaggerated Shading for Volumetric Data Illustrations convey overall shape as well as surface detail using certain lighting and shading principles. We investigate the use of an illustrative lighting model to accentuate features automatically by dynamically modulating the light position. This technique is useful for emphasizing regions of interest at various levels of detail. We apply this shading technique to accentuate detail in volumetric data. As the method is based on the use of gradients, we discuss and compare gradient computation techniques and their effectiveness. Since the technique takes into account multiple scales, the technique is able to highlight details at various levels. The results demonstrate that surface detail is accentuated regardless of the surface orientation and the size of features. Anubhav Kale, MS Student in CS (Advisor: Tim Finin), eBiquity Modeling Trust and Influence in the Blogosphere Using Link Polarity There is a growing interest in exploring the role of social networks for understanding how communities and individuals spread influence. In a densely connected world where much of our communication happens online, social media and networks have a great potential in influencing our thoughts and actions. The key contribution of our work is generation of a fully-connected polar social network graph from the sparsely connected social network graph in the context of blogs, where the vertex represents a blogger and the weight of an edge in the polar network represents the bias/trust/distrust between its connecting vertices (the source and destination bloggers). Our approach uses the link structure of blog graph to associate sentiments with the links connecting two blogs. (By link we mean the url that blogger a uses in his blog post to refer to post from blogger b). We term this sentiment as link polarity and the sign and magnitude of this value is based on the sentiment of text surrounding the link. We then use trust propagation models to spread this sentiment from a subset of connected blogs to other blogs to generate the fully connected polar blog graph. Our simple heuristics for analysis of text surrounding links and generation of missing polar links (links with positive or negative sentiment) using trust propagation is highly applicable for domains having weak link structure. This work has numerous applications such as finding â^À^Ülike mindedâ^À^Ý blogs, detecting influential bloggers, locating bloggers with specific biases about a predefined set of topics etc. Our experimental validation on determining â^À^Ülike mindedâ^À^Ý blogs on the political blogosphere demonstrates the potential of using polar links for more generic problems such as detecting trustworthy nodes in web graphs. Amit Karandikar, MS Student in CS (Advisor: Anupam Joshi), eBiquity Stateful Preferential Attachment Model for Constructing Blog Graphs WWW graphs have been very useful in the structural and statistical analysis of the web. Various models have been proposed to simulate web graphs that generate degree distributions similar to the web. Real world blog networks resemble many properties of web graphs. But the dynamic nature of the blogosphere and the link structure evolving due to blog readership and social interactions is not well expressed by the existing models. In this research work we propose a stateful model for a blogger to simulate blog graphs. We combine the existing preferential attachment and random surfer model for evolution blog graphs which is considered to be a class of scale-free networks. The blogger is modeled using read, write, idle states and finite read memory. The combination of these techniques helps in evolution of time stamped blog-blog and post-post network (through citations within the blog-blog network). Other parameters like the links per post and policy by which a new blog connects to the network help is the formation of different link structures. We empirically show that these simulated blog graph exhibits properties similar to the real world blog networks in their degree distributions, average path length, clustering coefficient and correlation. We speculate that the model will to be useful to evaluate and analyze the blogosphere properties. Pranam Kolari, PhD Student in CS (Advisor: Tim Finin), eBiquity Group Detecting Spam Blogs at Ping Servers Spam blogs, or splogs, are blogs featuring plagiarized or auto-generated content. They create link farms to promote affiliates, and are motivated by the profitability of hosting ads. Splogs infiltrate the blogosphere at ping servers, systems that aggregate blog update pings. Over the past year, our work has focused on detecting and eliminating splogs. As techniques used by spammers have evolved, we have learned how splog signatures are tied to tools that create them, that they are beginning to be a problem across languages, and that they require a much quicker assessment. Based on this background we present our continuing work on eliminating splogs in this poster. We discuss our larger goal of developing a scalable meta-ping filter that detects and eliminates update pings from splogs. This will considerably reduce computational requirements and manual efforts at downstream services (search engines) and involve the community in detecting spam blogs. Justin Martineau, PhD Student in CS (Advisor: Tim Finin), eBiquity BlogVox: Learning Sentiment Classifiers While sentiment detection, identification, and classification are popular research areas, researchers frequently work in only one domain at a time. Typical domains include movie reviews (Pang et al. 2002) and product reviews (Dave et al. 2003). Performing sentiment detection upon keywords chosen at run time is more difficult. The techniques applied to de- termine the sentiment of keywords in movie and product reviews are less effective when used on blogs due to a va- riety of reasons. Unlike reviews blogs tend to talk about many different subjects at a time making many NLP and ma- chine learning approaches more difficult. Finally, many of the techniques used in the different review domains incorpo- rate domain specific knowledge. The 2006 NIST TREC Blog track (Ounis et al. 2006) on â^À^Üopinion retrievalâ^À^Ý from blog posts, presents an oppor- tunity to tackle this problem. The task was defined as follows: build a system that will take a query string describing a topic, e.g., â^À^ÜMarch of the Penguinsâ^À^Ý, and return a ranked list of blog posts that express an opinion, positive or nega- tive, about the topic. NIST provided researchers with a data set of over three million blogs, and judged entries upon re- trieval results for a set of fifty test queries. Joseph C. (J.C.) Montminy III, MS Student in CS (Advisor: Krishna Sivalingam) On-The-Record: A Non-Repudiable, Authenticated, and Confidential Chat Client We propose an instant messaging protocol which provides a secure communication channel, with all participantsauthenticated and the conversation secure and non-repudiable. Our inspiration comes from the work of Borisov, Goldberg, and Brewer, who developed Off-The-Record Instant Messaging. Their goal was to create a system where two users could communicate confidentially, assured of each other's identity, but where the conversation was completely deniable and repudiable after taking place. Our aim is to create the exact opposite. In addition, our system attempts to mask the amount of network traffic, so that the information sent through our protocol is not only encrypted in content, but also encrypted in distribution. We implement our system in Java, as a client/server application with a GUI. Since the underlying encryption is proven to be secure, we analyze, as did Goldberg, whether the system can operate efficiently in a real-world scenario, so as to not hamper creativity and natural conversation. We also analyze whether the system can actually mask the true nature of traffic traversing our system. Sourav Mukherjee, MS Student in CS (Advisor: Hillol Kargupta) Distributed Probabilistic Inferencing in Sensor Networks using Variational Approximation This paper considers the problem of distributed inferencing in a sensor network. It particularly explores the probabilistic inferencing problem in the context of a distributed Boltzmann machine-based framework for monitoring the network. The paper offers a variational mean-field approach to develop communication-efficient local algorithm for Variational Inferencing in Distributed Environments (VIDE). It compares the performance of the proposed approximate variational technique with respect to the exact and centralized techniques. It shows that the VIDE offers a much more communication-efficient solution at very little cost in terms of the accuracy. It also offers experimental results in order to substantiate the scalability of the proposed algorithm. Patti Ordonez, PhD Student in CS (Advisor: Anupam Joshi), eBiquity Traumapod Traumapod is a DARPA funded project to build an unmanned vessel that will provide lifesaving medical care by a medic at remote location to soldiers wounded on the battlefield as they are being transported to hospital in the pod. We are working in conjunction with SRI Inc. and several other organizations and universities to build a prototype of this vessel. The pod consists of several subsystems. We are responsible for the creation and maintenance of the RMS. Resource and Monitoring System (RMS) The RMS is responsible for inventory management and tracking the supplies and location of tools around the pod by monitoring the messages passed between the subsystems. We use the Jess rule engine to infer medically relevant events by snooping on the low level messages such as tools changed, supplies dispensed etc which are exchanged in the network. These events are then recorded in the Medical Encounter Record (MER), which provides an entire summary of the surgery such as tools used, medicines administered and medically significant events which are time stamped. Andriy Parafiynyk, MS Stduent in CS (Advisor: Tim Finin), eBiquity SPIRE: Semantic Prototypes in Research Ecoinformatics Today, the information on the World Wide Web is growing at an astonishing rate providing a rapidly expanding source of valuable data. The abundance of distributed information on the Web increases the importance of efficient organization, sharing and retrieval of available data. This is particularly important in scientific research where efficient collaboration, exchange of results of experiments and observations, fast discovery of relevant information and data integration from different sources are the key elements for success. This spire project investigates how the Semantic Web can be used to facilitate research in ecoinformatics. We develop a set of ontologies, tools and applications within the Spire (Semantic Prototypes In Research Ecoinformatics) project which help us to demonstrate how the aforementioned goals can be achieved. In particular, we develop a number of ontologies (SpireEcoConcepts, EthanKeywords, EthanAnimals, EthanPlants) which help us to describe biological data obtained from multiple sources (as well as relationships among different parts of that data) in a machine-understandable way and apply OWL reasoners to that data to find answers for various questions which are of great interest for environmentalists and ecologists at the moment. As an effort to exploit popular on-line resources and knowledge generated by hundreds of thousands of people, we designed Splickr application to produce semantic web content and show how Semantic Web technologies can be used to generate new knowledge from existing resources. The goal of our research is to investigate the advantages and successful use cases of the Semantic Web as well as identify improvements that can be made to boost the expressivity of Semantic Web languages and usefulness of Semantic Web technologies. Soumi Ray, MS Student in CS (Advisor: Tim Oates), CORAL Transfer in the Context of Reinforcement Learning by Mapping Q-Tables Transfer in machine learning is the process of using knowledge learned in a source domain to speed learning in one or more related target domains. In human learning, transfer is a ubiquitous phenomenon. In machine learning, transfer is far less common. In this paper we present a transfer method for reinforcement learning when the learner does not have access to a model of either the source or the target domains (i.e., transition and reward probabilities are unknown) and there is no prior knowledge about how to map states or actions in the source domain to corresponding or similar states or actions in the target domain. Empirical results in a variety of grid worlds and a multi-agent block loading domain that is exceptionally difficult to solve using standard reinforcement learning algorithms show significant speedups in learning in the target domain. Shiva Sevarajan, MS Student in CE (Advisor: James Plusquellic), VLSI Research Lab Silicon-Based Timing Validation Using Race Condition and Glitch Detectors The role of first-silicon validation is to confirm the functional and timing correctness of a design. As levels of systematic and random sources of process variations increase, the role of validation is increasingly challenged to determine the source and location of timing related problems. Methods are needed to facilitate timing measurements in first silicon to better characterize and diagnose circuit marginalities. Such information is also valuable for calibrating models with hardware during process bring-up and beyond. In this work, we propose a novel test infrastructure designed to characterize path delays in a circuit to aid in the understanding of a designâ^À^Ùs marginalities and to provide information for the improvement of model-to-hardware correlation. - For more info see , call 410-455-3500, or email dept@csee.umbc.edu. To join/drop the colloquium list, email majordomo@csee.umbc.edu with "subscribe/unsubscribe csee-colloquium-out" in the message body.