The collected papers are available as the Journal of Undergraduate Research in High-Performance Computing, Volume 5, P.D. Coddington, editor, Northeast Parallel Architectures Center at Syracuse University Technical Report SCCS-747, December 1995. Copies of this report are available via request to reu-info@npac.syr.edu
Research proposals written by the students at the beginning of their research program are also available.
As part of an ongoing effort to develop a real-time terrain rendering
application for NPAC's Living Textbook Project, data structures and
algorithms for texture mapping of triangular regions were developed in C.
A graphical demonstration program, written in Tcl/Tk, shows that the
algorithms work well for 2D to 2D mappings. Future work includes
adaptation of the code for integration into an existing 3D rendering engine.
It is expected that the inclusion of texture mapping in the renderer will
improve both the image quality and speed of the renderer.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
A World Wide Web multimedia prototype for a developing
Digital Video Wire Service is introduced. The prototype retrieves
archived news scripts, as well as digitized video, from a
searchable database and presents them in a fashion consistent
with the current needs of communications professionals. The
prototype, designed for future expansion, utilizes a tight
integration of recent WWW technology, specifically centered
around database generated WWW pages.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
An interface for a parallel and optimized version of the sequential
Optimal Interpolation algorithm is developed. At the NASA Goddard
Space Flight Center's Data Assimilation Office, four-dimensional data
assimilation is conducted using a regional optimal interpolation
algorithm to obtain a data set that can be used for scientific studies
and forecasting. This is a very computationally intensive problem;
thus a parallel version of the Optimal Interpolation algorithm is
desired. Fortran 90 along with the Message Passing Interface was used
to develop a parallel library that allows parallelization of the
system. This library includes a parallel I/O and a parallel vector set
of routines with an HPF-like interface. A parallel sorting and a
restoring routine was also incorporated into the library. This library
also increases the portability of the code for MIMD supercomputers.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
The system of distributing multimedia is centered on the client-server
interface.
It has been developed such that the Microsoft Filemanager (client) connects to
a prototype video server which fetches the needed media. The prototype system
includes a working model of a single simulated file system residing at the
bottom of the system.
The simulated file system provides an organized single file system
to represent all files in the media archives. Clients draw on resources
through respective servers.
The system displays a high level of uniformity and portability.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
We investigate the new Internet browser
HotJava for its capabilities in providing dynamic and interactive
educational Web pages. The first task of the project was to monitor
Java-related activities on the Internet and to develop NPAC Web information
pages that summarize the current status of Java evolution by building a
navigable database of Java classes and applications. These are written
using applets, programs written in the Java language to run within
HotJava. The second task was to explore the use of applets in developing
educational modules by modifying and adding new functionalities to the
existing object classes.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
A path integral Monte Carlo method is implemented in a parallel
computing environment to price a family of derivative securities known
as options. In the path integral method, we construct a probability
distribution of asset price histories in a phase space where price is
a function of time. An asset's price history is the path of asset
price movements from the present time to the expiration time of the
option. We sample these histories in their entirety using the
Metropolis algorithm. This procedure allows us to obtain information
about an asset's sensitivities to multiple parameter sets. Once we
integrate the sampled histories, we can average over the terminal
prices to estimate a single value of the asset's true price. Using
standard financial models, we can then approximate the values for
options of various durations.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
This paper describes a project centered around the streamlining of
the Physics Department's World Wide Web (WWW) site that is used in teaching
PHY106: ``Science for the 21st Century'' students
about neural networks. The Mind and Machine Module Web site
offers various demonstrations at
a cost to both CPU time and site security. Target areas for
improvement include the user interface, temporary file usage, and
server load balance. By using additional WWW forms and GD extensions
to C, more advanced demonstrations were added, to a more secure site,
demonstrating practical applications of neural networks in every-day
life.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
An implementation of artificial neural networks was completed as an
addition to the ``Mind and Machine'' Web page for teaching neural
networks on the World Wide Web. A summary of unsupervised learning
was added to the Web site and an unsupervised network that uses
Principal Component Analysis to compress data and develop pattern
recognition capabilities was designed for the simulation. The
algorithm used was Sanger's learning rule for a single-layer network.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
The parallelization of the pre-existing, mathematically intensive
electromagnetic software provided by Rome Laboratories has been
accomplished. As predicted by the studies of Leskiw and Rome Labs on
parallelizing computational electromagnetic software, a major breakthrough
in efficiency now allows for the calculation of larger problem sets than
can be handled in serial computing. Specifically, the calculation of the
paths of reflection, diffraction and creepage in a large electrical
simulation has benefited the most from parallelization. Furthermore,
this project also demonstrates, through the use of advanced software
engineering methods, the ability to convert large, complex, serial
software systems to massively parallel architectures. The knowledge
gained from this research will serve as a guideline for the
parallelization of other large software systems in the near future.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
A method using graph coloring was developed for optimizing solutions
to the timetabling problem, specifically the example of scheduling
courses at a university given fixed time and space constraints.
Previous work on this problem preprocessed the data with
an expert system to arrive at a partial solution, and then performed
simulated annealing to fill out the solution. The heuristic using
graph coloring was designed to replace this expert system. The expert
system/simulated annealing combination often did not schedule all the
courses, whereas the graph coloring method was designed
to produce a complete solution on its own. This complete solution
will then be refined through the use of simulated annealing.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
An implementation of 32 and 64 bit parallel random number generators is
presented.
In the near future, 64 bit processors will be commonplace and parallel
architectures will dominate the computing scene.
These developments will lead to an increase in
computing power, which will enable simulations of larger proportions,
in particular those using random number generators.
Longer periods, better randomness properties, and portability to
different machines and architectures are needed for these generators to
keep up with simulation demands. The combination of two linear
congruential random number generators, used as a single generator, efficiently
implemented in a language such as High Performance Fortran, is expected to meet
these requirements.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
As part of the Living Textbook Project, a three-dimensional terrain
rendering program has been developed. Unfortunately,
three-dimensional rendering requires intense computation and
is thus rather slow. As a partial solution to this problem, a
two-dimensional terrain renderer has been produced. This new renderer
uses C and a high-level package called Tcl/Tk, and betters the 3D
renderer's speed.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
The use of 3dviewnix, a three-dimensional medical imaging program, is
explored for the display of data slices of a human male provided by the
National Library of Medicine. The data underwent a series of conversions,
including changing the data from RGB color to greyscale, in
order to be compatible with 3dviewnix. Once the data is converted,
3dviewnix can be used to create three-dimensional reconstructions of the data
slices, as well as extracting the skeletal structure from the reconstructions.
The data slices, which are in one millimeter increments across the body, do
not always convey much information. A program was designed to alleviate this
by making vertical slices instead of horizontal. This program required a
small amount of pre-processing because the slices were surrounded by a
background of blue frozen gelatin that had to be removed.
The goal of the project is to make the slices, reconstructions, and the
extracted skeleton available via the World Wide Web for educational purposes.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
File Navigator is an authoring system that allows the user to browse through a
selection of parallel algorithms. It consists basically of a configuration
file, a form, and a Perl script which work in conjunction with each other to
search for the file and then display it for the user. It creates a
disciplined method of navigation through the algorithms by allowing the
user to choose along which view and path that they would like to see the
algorithms on.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
A parallel database is developed to improve the DR-LINK information retrieval system. DR-LINK is a system which finds the documents most relevant to a user's natural language queries. Relevant documents are found by analyzing index files.
Index files contain weighted lists of key words and phrases and other characteristics of the document. The distinguishing information of the queries are matched with the index files in order to find the most relevant documents to the user.
The work requires transforming the current vector-implementation of index files into a parallel database, an innovation which should have several positive effects on the system. Results include a test on performance and a test to find whether this new implementation is a better way to add new documents to the system.
Future work will include testing on whether the database provides a way to detect new information about linguistic patterns.
A
Postscript
version of the paper is available for immediate retrieval and viewing.
Texture Mapping in a Real-Time Terrain Rendering Environment
Marc S. Abramowitz
& Paul Coddington
News-On-Demand: Prototype of a Multimedia Digital-Video Wire Service
Philip J. Anzalone, Joan A. Deppa & Kim Mills
Four-Dimensional Data Assimilation: Parallelization of an Optimal Interpolation Algorithm
Reza Behforooz, Gregor von Laszewski & Miloje Makivic
Distributed Multimedia
Adam Foxman & Chris M. Sedore
HotJava - Capabilities and Its Use in Education
Meryem Ispirli, Wojtek Furmanski, Geoffrey C. Fox & Nancy J. McCracken
Pricing of Options Using the Path Integral Monte Carlo Approach
Eugene H. Kim & Miloje S. Makivic
Teaching Neural Networks using the World Wide Web
Jason Krug, Simon Catterall, Marco Falcioni & Hon W. Yau
Interactive Simulation of Unsupervised Neural Networks
Colleen McCarthy & Hon W. Yau
Parallelization of Computational Electromagnetics Software
Grant Ingersoll & Donald Leskiw
Optimizing Timetabling Solutions Using Graph Coloring
Sara Miner, Saleh Elmohamed & Hon W. Yau
Implementation of a Random Number Generator for High Performance Fortran
Olukayode O. Odeyemi & Paul Coddington
Real-Time Interactive Terrain Rendering
Christine Patton & Paul Coddington
The Visible Human
Kelley Pellini & Paul Coddington
A World Wide Web Authoring Tool for Hyperlinked Information with Multiple Views
Vareck Self & Nancy J. McCracken
Parallel Database Design for a Natural Language Information Retrieval System
Anne Thomas & Nancy J. McCracken
1995 REU Program,
Northeast Parallel Architectures Center,
Syracuse University,
reu-staff@npac.syr.edu