| No | Demo Name | Presenters | Demo Description |
| 1 | | | |
| 2 | A Math and Equations Search System | Abdou Youssef | To process and disseminate technical knowledge more effectively, efforts are underway worldwide to create digital libraries of mathematical contents. A notable example is the Digital Library of Mathematical Functions (DLMF) at the National Institute of Standards and Technology (NIST), which will be an online and hardcopy replacement of the well-known mathematical Handbook by Abramowitz and Stegun. To benefit from such digital libraries, users should be able to search not only for text, but also for equations. The mature information retrieval (IR) technology is primarily for text contents. When applied to math search, text IR systems run into problems that stem from their inability to understand mathematical symbols and structures. In this research project, we developed a new search and retrieval system for mathematical contents. The system accepts and understands mathematical equation queries, syntax and structures. It also has the ability to present all matching equations in a distilled document created dynamically on the fly. In addition, it highlights matching equations in their native HTML documents. This demo will present the system, highlighting its novel features and capabilities. |
| 3 | AIMS: An Immersidata Management Sytem | Cyrus Shahabi | The goal of this demonstration is to describe a system that enables the acquisition, storage, querying and analysis of immersive sensor data streams, termed immersidata, in the wavelet-transformed domain. The system is called AIMS for An Immersidata Management System. Immersive sensors are the user interfaces of the future and it is absolutely critical to facilitate the manipulation and analysis of their generated datasets. Management of immersidata is challenging because they are: 1) multi-dimensional, 2) spatio-temporal, 3) continuous data streams, and 4) noisy. In addition, they are produced with a high rate and hence their real-time query processing is a challenging task. The use of wavelets alleviates these challenges collectively. Furthermore, the resulting system enables many other applications dealing with large multidimensional data streams that
Require approximate and/or progressive query support. This project is benefiting from the collaboration of a multidisciplinary team of scientists in the areas of computer vision, graphics, signal processing and human factor studies at USC. The main benefits of this project to the society would be through specific application developments, which would enable more natural interaction methods with software systems especially for disabled and elderly users.
In the video demonstration, we motivate the need for a general-purpose immersidata management system such as AIMS and then explain different components of AIMS in details. We also illustrate the use of AIMS in collecting data within a couple of immersive applications as well as its utilization to query and analyze the collected data towards the application objectives. For more information, please see http://infolab.usc.edu
|
| 4 | An Interactive GUI for Authoring Queries on Human Motion | Jezekiel Ben-Arie | We plan to demonstrate how a user can compose a query that defines a set of human activities or motions that could used to find similar activities in a video database. The user in the demo will operate an interactive Graphical User Interface (GUI) that allows her/him to define the desired motions/activites and their trajectories and locations. Standard activities such as walking, running etc. will be available via a menu in the GUI. Whereas specialized motions can also be defined interactively by graphical tools as well. The trajectories of activities such as walking and the locations of other activities or specialized motions also can be defined interactively by the GUI. When the user finishes his composition of the query, he can view it in any direction, orientation or scale via the GUI. The part of activity recognition and query search is already demonstrated in our web site and could be incorporated with the demo of the GUI. I plan to prepare a live demo and also a 5 minute video of the same. |
| 5 | Animation databases for authoring | B. Prabhakaran | We demonstrate an animation authoring toolkit that reuses 3D models and motion sequences. The reusability is supported using an XML-database that can handle different animation formats such as VRML and MPEG-4. |
| 6 | Collaborative, Content-Based, and Hybrid Recommendation of Research Papers | Joseph A. Konstan | A follow-up to our 2002 demonstration that shows a new set of algorithms for research paper recommendation. These algorithms--which include variants of collaborative filtering, content-based filtering, and hybrid approaches--produce qualitatively different sets of recommendations. The demonstration will be an enhanced replication of a Summer 2003 Internet-based user study where we invited paper authors to sample recommendations and provide feedback on their usefulness. We invite attendees to try the system with their own papers, and to explore the nature of different recommendation algorithms. The experiment is built upon the CiteSeer database of research papers, and is being used to help design and build our TechLens research paper recommender system. |
| 7 | Computing Consistent Query Answers Using Conflict Hypergraphs | Jan Chomicki | We will demonstrate HIPPO, a system for computing consistent query
answers for inconsistent databases (databases that violate integrity
constraints). HIPPO is founded on the definitions of repair and
consistent query answer from [Arenas, Bertossi and Chomicki; PODS 99]
but uses a new computational mechanism based on conflict hypergraphs.
HIPPO computes consistent query answers for denial constraints and
quantifier-free queries. It runs on top of PostgreSQL and thus is not
restricted to databases that fit in main memory. |
| 8 | Creating Customized Metasearch Engines on Demand | Clement Yu | A metasearch engine is a system that provides unfied access to
existing search engines. In this demo, we demonstrate SE-LEGO,
a tool that allows customized metasearch engines to be created on-demand
instantly. A user provides the URLs of the search engines he/she
wants to use as components of a metasearch engine. SE-LEGO takes
these URLs as input, automatically creates a program to connect to the server
of the server of each search engine, automatically create another
program to extract the URLs of documents retrieved by each search engine
when a query is sent to each search engine. |
| 9 | Crime Data Analysis | Lin Liu | This demo aims to illustrate how GIS can play an important role in crime data analysis. Real databases from the Cincinnati Police Department, including crime database, police vehicle stop database and others, will be used in the demo. This demo shows an example how scientific research can directly contribute to solving and understanding real problems in the society. |
| 10 | Edward Chang | Multimodal Perception-based Image Retrieval | We demonstrate our MM-PBIR retrieval system, which was recognized by the Multimedia community as a major breakthrough in image retrieval. MM-PBIR integrates traditional IR algorithms with novel machine-learning techniques to support three query paradigms: query-by-nothing, query-by-example, and query-by-keyword. In addition to image retrieval, the system supports semi-automatic image annotation. MM-PBIR is being deployed at several stock-photo companies to conduct searches in large-scale image repositories. |
| 11 | Flamenco: Incorporating Faceted Metadata into Search Interfaces | Marti Hearst | In this demonstration I will present an interface paradigm called
Flamenco which allows users to navigate explicitly along conceptual dimensions that describe a collections items. The interface makes use of hierarchical faceted metadata and dynamically generated query previews to seamlessly integrate category browsing with keyword searching. Usability studies have found strong preference results for this kind of faceted category interface over that of a strong baseline, suggesting this is a promising approach for image search interfaces specifically, and potentially for text collections as well. I will demonstrate the system on a collection of fine arts images and a collection of PubMed bioscience abstracts.
|
| 12 | Information Fusion Across Multiple Text Sources | Dragomir Radev | We will be demonstrating NewsInEssence - the oldest live Web-based multidocument news summarizer as well as its recent analogue for summarization of Web queries - WebInEssence. Both systems are based on Cross-document Structure Theory which predicts semantic relationships such as information subsumption, contradiction, and elaboration across multiple documents in a set.
|
| 13 | Knocking the Doors to the Deep Web: Understanding Web Query | Kevin Chen-Chuan Chang | Toward building a "MetaQuerier" for integrating databases on the
"deep Web," we face a fundamental challenge of understanding their
query forms, to automatically extract source query capabilities.
Such "form understanding" is non-trivial, as evident from the
infamous "butterfly ballots" in Election 2000. We demonstrate our
solution, which develops the key hypothesis of "hidden syntax" and
thus transforms the problem into best-effort parsing with inherent
ambiguities. |
| 14 | MINDS: Detecting Novel Network Intrusions using Data Mining | Vipin Kumar and Jaideep Srivastava | This demo will provide an overview of the MINDS (Minnesota Intrusion
Detection System) project at the University of Minnesota
that is developing a suite of
data mining techniques to automatically detect novel and emerging
attacks against computer networks and systems.
This demo will show how the behavior-based anomaly detection approach of MINDS
is suitable for detecting new and previously unknown types of intrusions,
which often indicate emerging threats.
Second, we show how association pattern analysis can
be used to summarize and characterize anomalous network connections.
Given the very high volume of connections observed per unit time,
such characterization of novel attacks is essential in enabling a security
analyst to understand emerging threats.
MINDS is being used at the Army Research laboratory and the
University of Minnesota in a production mode successfully to detect novel
intrusions that could not be identified using commonly used
signature-based tools such as SNORT.
Summarization of anomalous connections using association pattern analysis
has been very helpful in understanding nature of cyber attacks as well as in
creating new signature rules for intrusion detection systems. |
| 15 | Semantic Video Classification and Indexing | Jianping Fan | In this demo, we will show our achievement on semantic-sensitive medical education video analysis, indexing and classification. The salient objects, that are semantic-sensitive to the relevant semantic medical concepts, are detected automatically. The principal video shots are then classified into some pre-defined semantic medical concepts. |
| 16 | Skipping Streams with XHints | Sudarshan S. Chawathe | When a system processes streaming XML data, a significant portion of
time is spent on simply parsing the input. In optimized systems
(e.g., the XSQ system for streaming evaluation of XPath queries), it
is not uncommon for parsing to account for 90% of the running time.
Thus, further improvements in throughput require methods reduce
parsing times. We have developed the XHints system for augmenting XML
streams with hints (or annotations) that aid downstream processors.
We will demonstrate its utility using a modified version of the XSQ
system, which uses these hints to skip parsing portions of the input
that can be safely ignored.
|
| 17 | Taxontree: A Visualization Tool for Animal | Ben Bederson | TaxonTree allows full-text searching and browsing of a large, widely used
multimedia database of almost 200,000 animal names organized taxonomically.
Binary branching illustrates relationships, synapomorphies supporting those
relationship are available, and links to external websites are integrated.
TaxonTree combines dynamic query interfaces and zoomable graphics to visually
accommodate highly interconnected data. Because the structure of the data is
visualized, the interface should be suitable both for biologists and for the
general public. Names of organisms are essential to any biological database,
including genomic databases, so a tool that allows effective searching and
browsing of these names has wide application. |
| 18 | Taxontree: A Visualization Tool for Animal Names | Ben Bederson | TaxonTree allows full-text searching and browsing of a large, widely used
multimedia database of almost 200,000 animal names organized taxonomically.
Binary branching illustrates relationships, synapomorphies supporting those
relationship are available, and links to external websites are integrated.
TaxonTree combines dynamic query interfaces and zoomable graphics to visually
accommodate highly interconnected data. Because the structure of the data is
visualized, the interface should be suitable both for biologists and for the
general public. Names of organisms are essential to any biological database,
including genomic databases, so a tool that allows effective searching and
browsing of these names has wide application. |
| 19 | The Telemakus System | Sherrilynne Fuller | Current scientific research takes place in highly specialized contexts that can hamper communication across disciplines. Knowledge from one discipline may be useful for another without scientists knowing it, researchers may be unaware of relevant work by others that could be integrated into theirs, and important findings just outside a scientists focus may go undiscovered. In addition, it is practically impossible to keep current with information that continues to grow at an exponential rate. There is a great need for information retrieval systems that provide more than a list of citations pointing to a group of documents that still must be read through to find answers to questions. Scientists need systems that actually extract desired facts and answer questions and help them synthesize information extracted from multiple documents. Also, scientists need tools to rapidly review the research methods and findings within multiple documents, as well as view the relationships among the documents in a retrieval set.
A comprehensive approach to these information retrieval and management challenges is the goal of the Telemakus research project. Dr. Sherrilynne Fuller will provide a demonstration of the Telemakus system which is designed to enhance the knowledge discovery process through graphical presentation and interaction tools to mine and map author-reported research findings and their associated methods. The Telemakus system provides two visual representations of a research report: (1) a schematic representation of the report that serves as a document surrogate, and (2) interactive concept maps of linked relationships across groups of research reports. These visualizations provide the user with both a macro- and micro-view of domain knowledge. The micro-view presents consistent and detailed document attributes such as research methods and findings for each research report in the database. The macro-view facilitates identification of patterns as well as expected and unexpected occurrences of relationships among research concepts, and permits visualization and dynamic navigation of scientific domains. While other research has demonstrated that aggregating and analyzing research findings across domains augments knowledge discovery, the Telemakus system is unique in combining document surrogates with interactive concept maps of linked relationships across groups of research reports and may re-conceptualize the way we currently represent, retrieve, and assimilate research findings from the published literature.
|
| 20 | Ubiquitous Distributed Data Mining | Hillol Kargupta | Most modern vehicles generate continuous stream of data with large number of features. Mining and monitoring this data stream in real-time can offer many interesting possibilities such as vehicle health monitoring, drunk driving detection, and security threat management. This demonstration will exhibit a distributed data mining system for mobile vehicle fleet monitoring application and the hardware developed at UMBC to support the software. |
| 21 | Video Surveillance with Distributed Camera Networks | Yuan-Fang Wang and Edward Chang | We will demo our multi-camera video surveillance system that is currently under development. Our surveillance framework addresses the detection,
representation, and recognition of motion events. The detection phase
handles spatio-temporal data fusion for efficiently and reliably
extracting motion trajectories from video. The representation phase
summarizes raw trajectory data to construct hierarchical, invariant,
and content-rich descriptions of the motion events. Finally, the
recognition phase deals with learning using imbalanced training
datasets and infinite-dimensional data that also exhibit temporal
ordering. |
| 22 | Visual Text Mining: Examples in Homeland Security and Medical Informatics | Hsinchun Chen | Our research has been focusing on "visual text mining," with techniques ranging from linguistic processing, association networks, social network analysis, and content and spatial-temporal visualization. We illustrate these techniques in two domains: homeland security for crime data mining and visualization (COPLINK) and medical informatics for genomic pathway identification and visualization (GeneScene). |
| 23 | VocalSearch | Gregory H. Wakefield | To learn more about an article you have read, you can search the web
using a variety of engines. However, to learn more about a song you have
heard, few aids exist unless you happen to know such textual
information as the artists name, the songs title, or the lyrics.
VocalSearch has provided the opportunity to address several of the
issues surrounding the design and development of search engines for
such non-textual materials, such as music. Similar to typing on a
keyboard, VocalSearch uses a natural means of musical query - singing
or humming - as input to search over a database of music. Results of
the search are ranked according to their best score against the members
of the database. Features of the search engine will be illustrated as
part of the demo, including the results of some of our research on
sources of error in the search process. Finally, examples of our
preliminary work on such extensions of the search to non-melodic
aspects of music, such as singer identity, will be provided.
|
