UHD REU Program > Project Description
Project Description
1. Large-sclae Knowledge
Representation and Validation
Mentor: Dr. Ping Chen
Student:
Stanley
Roberts
In this project we target on
validating and representing dependency knowledge acquired in 2009 summer and
fall. Although frequency of a dependency relation (acquired in 2009 summer)
can reflect the semantic relatedness of head word and dependent word to a certain
degree, this direct measure is inevitably distorted by occurrence frequencies
of head word and dependent word. For example, suppose that both wine ->
red and water ->red occur 5 times in the knowledge base respectively,
which may suggest that these two pairs of words are equally related. However,
wine ->red should be a stronger connection since "water" is a
more common word than "wine" and occurs more frequently. To
overcome this bias, we tried to assess the strength of dependency relations
with multiple statistical hypothesis tests, such as likelihood interval,
likelihood ratio test, u test, and chi-square test. Participants of this
project will work on a complete cycle of research, including problem
analysis, design, implementation, evaluation, and paper publishing.
Requirements: Proficient
programming skill of Visual C++ and understanding of data structures.
2. Word sense
disambiguation
Mentor: Dr. Ping Chen
Student:
Somto Ozoka, Max
Choly
Continued
project from summer 2009. We participated in SemEval 2010, an International
WSD competition.
Requirements: Proficient
programming skill of Visual C++ and understanding of data structures.
3. Semantic space
modeling
Mentor: Dr. Ping Chen
Student:
Benjamin
Walker
2-D
Lexical Semantic Space Building is an attempt to map human knowledge from
natural language into a 2-dimensional matrix. The reasoning behind this
method can be traced to cognitive neuropsychology and the theories of concept
learning. Cognitive neuropsychological studies performed on brain-damaged
patients have demonstrated strong evidence that the human brain stores
knowledge into categories. Patients damaged in certain areas of the brain are
unable to process information about entire categories of objects such as:
fruit/vegetable, living animate things, conspecifics, and non-living. Based
on the experiments on brain-damaged patients, it can be inferred that if
knowledge can be lost in entire categories due to brain damage, the learning
of new knowledge in fully functional brains will classify that knowledge into
categories as well. The prototype theory of concept learning explains that
people categorize knowledge based on one or more central examples that branch
out from strongly related ideas into lesser related ideas based on the
experiences of the person. We can apply these theories when attempting to
represent human knowledge by storing concepts into a 2-D Lexical Semantic
Space. The insight, "You shall know a word by the company it keeps (Firth,
1968)," sheds light on valuable workings of natural language. The sense
of a word can be disambiguated by the context it is used in. When two
different words are often used within very similar contexts, it is highly
likely that those two words have a strong relation. By analyzing large corpa
of natural language, we are able to gather information about individual words
and map these words onto a 2-D matrix.
Requirements: Proficient
programming skill of Visual C++ and understanding of data structures.
4. Comminication path
problem in networks
Mentor: Dr. Shengli Yuan
Student:
Michael
Butterworth, Luis A. Neco Ortiz
With the recent advances in
networking technology, communication paths are now capable of
transporting traffic at
data rate up to 100 Gbits/s or even 160Gbits/s. More and more business,
education and social activates are now online. To prevent service disruption
and data loss, the reliability of the communication paths should be
maximized. Equally important is the detection and prevention of security
breaches. One of the approaches to address these challenges is to
deploy efficient routing schemes so that at the time when the network
connections are established, the possibilities of network survavalbility and
security are maximized.
Requirements:proficient in C++
or JAVA programming. Study in a network introduction course or programming
knowledge of simple routing algorithms will be highly desirable.
5. Designing
3D Virtual Models
Mentor: Dr. Ongard Sirisaengtaksin
Student: Brian
Holtkamp, Stephanie Thames
The objective of the project is to provide student with knowledge of creating
3D virtual models. The concepts of graphics, such as polygons, ray tracing,
texturing, and shading were implemented to develop a University of
Houston-Downtowns main building 3D model. To be able to render the model
quickly, the number of polygons of the model was optimized. The skeleton of
first floor inside the model was completed. We still need to add six
additional floors to the model. A navigation system to move around the model
was also developed but due to the time constraint it has not been completed.
The navigation system uses the idea of first-person-shooter game approach.
The concept positioning camera was used as guide in the model. We also
optimized the number of the polygons to implement collision areas such as
walls and windows for rendering purposes. To optimize rendering for
navigation system, we render only the floor that the user enters. We found
that by lowering the number of polygons for collision areas and present only
the floor upon the user entering help rendering process much faster.
Requirements:
Work knowledge of C and understanding of computer organization and
architecture concepts, general computer hardware and operating systems.
6. Automatic Detection of
Sub-Kilometer Craters in High Resolution Planetary Images.
Mentor:
Dr.
Tomasz
Stepinski
Student:
William
Isaac Miller
 Impact craters, the structures formed by collisions of
meteoroids with planetary surfaces, are among the most studied geomorphic
features in the solar system because they yield information about the past
and present geological processes and provide the only tool for measuring
relative ages of observed geologic formations. Thus, surveying impact craters
is an important task in the planetary research. Planetary probes deliver ever
increasing volume of high resolution images; however, the scientific
utilization of these images is hampered by the lack of tools for their
effective automated analysis. This project seeks to develop a processing
pipeline for fast and accurate surveys of small craters from high resolution
images. Such system will make possible assembling global, million crater
catalogs of sub-kilometer craters on Mars, Mercury, and the Moon.
Requirements: Proficient
programming skills of Mathematica
or Matlab
7. Client-Server
simulation framework for agent-based crowd simulation of evacuation scenarios
Mentor: Dr. Hong Lin, Dr. Ongard
Sirisaengtaksin
Student: Borislav
Ganov
The research in agent-based simulation in evacuation scenarios and evacuation
practices has a significant background. This research was focused on studying
the previous research done, and developing a model based on fuzzy logic in
the same direction as other simulators. The research and development led to
laying the geographical and communication framework of the model and
providing the basis for future development of the project. The developed
framework differs form others in its nature, being a client-server model
allows scalability of simulation and load distribution allowing the simulation
to be run on large number of computers such as clusters of computers. The
floor plan model (EscapeSim) consists of a two-dimensional array containing
objects called tiles (SimTile). Each of the SimTile objects contained in the
array represents a point or and segment of the area, and possesses a list of
characteristics. Moving objects (people) are modeled as agents, which act as
clients in the client-server architecture of the program. A client-server
approach was chosen as basis of the framework. The model of the server side
was designed to accept (or listen) for variable amount of clients
(connections). A list of clients connections (or thread count) is maintained
at all times. The client-server communication is expressed by following a
simple proprietary protocol, which is composed of three major stages, namely
the clients request (initialization), the continual client-server data
exchange to maintain a running simulation (update), and the exit. The client
slide is the driver of an agent, hence the need of an object representing the
agent. Each agent has multiple attributes which will play a different role in
determining agents next move. The agents sight is used to determine how many
surrounding tiles are needed by the agent to make a move. Further the agent
has knowledge of surrounding area, exits and type of strategy of making the
next move.
|