Honors Projects

Analyzing and Extending an Infeasibility Analysis Algorithm

Ammar Malik — Wed, 01 May 2013 03:35:23 PDT

Constraint satisfaction problems (CSPs) involve finding assignments to a set of variables that satisfy some mathematical constraints. Unsatisfiable constraint problems are CSPs with no solution. However, useful characteristic subsets of these problems may be extracted with algorithms such as the MARCO algorithm, which outperforms the best known algorithms in the literature. A heuristic choice in the algorithm affects how it traverses the search space to output these subsets. This work analyzes the effect of this choice and introduces three improvements to the algorithm. The first of these improvements sacrifices completeness in terms of one type of subset in order to improve the output rate of another; the second and third are variations of a local search in between iterations of the algorithm which result in improved guidance in the search space. The performance of these improvements is analyzed both individually and in combinations across a variety of benchmarks and they are shown to improve the output rate of MARCO.

Native Cardinality Constraints: More Expressive, More Efficient Constraints

Jordyn C. Maglalang — Fri, 20 Apr 2012 13:28:44 PDT

Boolean cardinality constraints are commonly translated (encoded) into Boolean CNF, a standard form for Boolean satisﬁability problems, which can be solved using a standard SAT solving program. However, cardinality constraints are a simple generalization of clauses, and the complexity entailed by encoding them into CNF can be avoided by reasoning about cardinality constraints natively within a SAT solver. In this work, we compare the performance of two forms of native cardinality constraints against some of the best performing encodings from the literature. We designed a number of experiments, modeling the general use of cardinality constraints including crafted, random and application problems, to be run in parallel on a cluster of computers. Results show that native implementations substantially outperform CNF encodings on instances composed entirely of cardinality constraints, and instances that are mostly clauses with few cardinality constraints exhibit mixed results warranting further study.

QuickAdvise - The Search for a More Efficient Method of Advising

Gregory G. Pengiel '94 — Tue, 24 May 2011 13:51:18 PDT

In order to choose courses both efficiently and properly, a student and
advisor must look at which courses have been taken, and also which ones are
necessary. They must then determine whether or not the student is qualified
to take the necessary courses. Unfortunately, there is no easy way to record
and maintain this work so that it may be used throughout the college career.
Hence, even though the student and advisor recently determined which
classes were headed, they must once again look up all relevant information
and determine which courses are best. This is exactly the type of problem
that a computer can solve. However, in order to properly design a software
application, various aspects of the situation must be extensively researched.
First, the solution presents a quandary in terms of the system design. The
language that is best suited to this type of application must be determined.
Secondly, in order to be useful, the user interface must be well constructed.
People do not like software that is not user friendly, so to make this a
worthwhile software application, the interface must be both intuitive and
accommodating. Finally, different development platforms must be considered
in order to properly solve the problem. If the previously mentioned items are properly devised and documented, the actual construction of the software
should not be difficult. Other points to consider are testing documentation,
unit and system testing, code reviews, scheduling, and graphic user interface
design,

The use of a genetic algorithm to evolve networks for a natural language processing task

Alexander E. Dimov '02 — Thu, 09 Oct 2008 14:11:19 PDT

In this project a novel approach was taken for performing a natural language task. The task requires a neural network to predict the grammatical category of the next word in a stream of sentences. There are two main reasons why this task is interesting. In natural language processing, it is sometimes very difficult to determine the grammatical category of a word in a sentence when that word could belong to different grammatical categories depending on the context. For example, the word "run" can either be a noun or a verb in a certain sentence. The ability to correctly determine the category of the word can help a computer process natural language. In addition, the approach taken here to solve this task can lead to insights about the way the human brain learns and/or understands language. A Genetic Algorithm, which is conceptually based on simple principles known from Genetics, was developed and utilized to evolve neural networks that were used to perform the task. Genetic Algorithms have been used with remarkable success to solve complex problems in a number of fields but not for this type of problem. In addition, networks were trained via a classic learning algorithm, called back-propagation, to perform the same task. Since a Genetic Algorithm has not been used for this type of task, an implicit goal of this project was to show that it can be used. One of the other main questions addressed is whether learning (as in the case of training a neural network via back-propagation) and a search for an optimal solution (as in the case of the use of a Genetic Algorithm to evolve neural networks) differ and if so, how. Also, the underlying properties of the two different types of networks (depending on the approach taken to obtain them) were compared. Finally, issues about the computational complexity of the Genetic Algorithm were studied and discussed. These issues included the relationship between the input size (for ex. 10000 sentences) and the perfonnance of the network developed via the Genetic Algorithm approach, as well as the way the network must change as the input changes in size and the task changes in complexity (i.e. as the grammar and lexicon change) while the optimal parameters (of the Genetic Algorithm) are used.

Course Scheduling Software: Reference Manual

Saa Gobir '91 — Thu, 09 Oct 2008 14:11:18 PDT

Welcome to version 1.0 of the Course Scheduling software. Course Scheduling is designed to meet needs of anyone who is tying to create a course schedule.

This software is a product that compiles, links, and runs in C++; although it has C extension. It has been specifically designed to work In C++ exclusively. I had mixed a little of C++ functions with C structure programming to complete this package and make it more efficient. It is versatile, quick, and efficient.

This instruction manual is brief, due to fact that the software is well documented. It only covers the basic functions and features; creating data, editing data, sorting and searching data.

To get started, you need to be a little familiar with DOS. If you are not familiar with DOS, please turn to page 1.

Financial Aid Budget Projection Methodology

Amy N. Baird '94 — Thu, 09 Oct 2008 14:11:15 PDT

The nature of this research project is to make a very strong contribution toward the final goal of completely automating the Financial Aid Office at Illinois Wesleyan University over the next five years.

PowerFAIDS: Building a Road to Financial Aid Efficiency

Lauri Nichols '96 — Thu, 09 Oct 2008 14:11:14 PDT

To most students, the Financial Aid office is a small room in the basement of Holmes Hall where they are occasionally sent to sign over a paper or two. It has something to do with money, and every so often, these students get something in the mail that tells them just how much it's going to cost them to continue their education here at Illinois Wesleyan. They know that papers are filled out, usually by their parents, and then in a couple of months, a figure jumps out of nowhere and becomes "Your Financial Aid Package."

Improved Data Migration and Processing for Projecting the Financial Aid Budget

Jeffery L. Olson '96 — Thu, 09 Oct 2008 14:11:14 PDT

Last fall, I again resumed work on the budget projection model that encompassed five spreadsheets. Four of these sheets generated a set of statistical averages for each class. Each one consisted of 101 columns containing data for the four-to five-hundred students(rows) in each class. In addition, a fifth sheet used these averages to generate a highly accurate prediction for expenditures in the upcoming year. However, there were two main areas of improvement that became readily apparent: importing data and the sheets themselves.

Mapping Robotic Movement to a Three-Dimensional Coordinate System

Craig A. Materick '97 — Thu, 09 Oct 2008 14:11:13 PDT

The Illinois Wesleyan Intelligence Network on Knowledge (I.W.I.N.K.) is a project to design and implement an artificial"person'" named Shelley. Robotics, networking, and artificial intelligence will be the main topics ofthe preliminary work. For my research honors project I designed the three-dimensional coordinate system in which the robotic arms move and interact with objects. The anns we have constructed are based on an arrangement of six servos, each of which rotate approximately 185 degrees. The program takes in data about the location of an object in three-dimensional coordinates and moves each of the six motors in the ann to arrive at that point.Included in this work is a look at robotic arm developments through history, from Leonardo da Vinci through the Industrial Revolution, and beyond. Also discussed are the various joint and arm designs developed during these years of research and some robotics projects which employ these different designs. Next, we will investigate the various methods of control developed by other robotic arm research projects and apply one particular method to control Shelley, as briefly outlined above. Finally, we will highlight problems we faced during the implementation of this program, the solutions to these problems, and various ideas about future research possibilities.

Designing an Integrated Environment for Artificial Intelligence

Andrew B. Ritger '99 — Thu, 09 Oct 2008 14:11:10 PDT

The SHELLEY RESEARCH GROUP (part of the Illinois Wesleyan Intelligence Network on Knowledge -IWINK) has been in existence for several years, and has benefited immensely from various student contributors who have added such components as robotic arm control, cross platform networking, an artificially intelligent tic-tac-toe player, and an interactive teaching tool demonstrating the functionality of artificial neural networks. What is lacking, however, amidst these undergraduate contributions to the SHELLEY Project, is an effective means of integrating existing components into a single cohesive functional unit, let alone any easy means of making further contributions within a simple unified context. The focus of this research has been to design an all-encompassing structure for incorporating the different components of SHELLEY (both existing and future). Because we must operate under the assumption that we cannot predict what future contributions will be made to SHELLEY, nor how these components will be used, this integrated environment must be both flexible and expandable in such a way as to not confine future projects. The approach to artificial intelligence that the SHELLEY RESEARCH GROUP has taken relies heavily upon interaction with the surrounding environment. For this reason, many of the existing components are devices for receiving input from SHELLEY'S surroundings (such as vision cameras) or acting upon the surroundings (such as robotic arms). Thus, we can assume that future contributions will fall under two primary categories: additional devices (either cognitive, modules, such as neural networks, or interactive devices, such as cameras or arms), or intelligent agents (such as tic-tac-toe players, or navigation systems) that will use these devices. The environment must then be flexible in two manners -allowing for the addition of further devices, and providing a task management mechanism for accessing these devices. The solution is to use a modern operating system model where the devices that SHELLEY uses to interact with her environment correspond to computer hardware devices and their drivers, the intelligent agents are analogous to processes that run on the system and use the devices, and the administrator, which coordinates these agents and their usage of devices, can be compared to the kernel of the modern operating system.

Computer Vision: Object Recognition

Michael Zalokar '00 — Thu, 09 Oct 2008 14:11:09 PDT

One of the growing fields in computer science is that of Artificial Intelligence or AI. Many theories have evolved to make a computer intelligent and so far no one has succeeded (Dreyfus 1992). One of the methods used by the Shelley Project in the past has been to use a back propagation neural network that is the backbone of the GNU Neural Network Visualizer (GNNV). GNNV uses a neural network to try to identify known objects, like faces, in the field of view. A different method, that is the focus of this research, is to identify objects in the image. These objects could be squares, circles or even blobs. Neural networks can work through changes in environment without changing the code provided appropriate training. However it is tough to know what the neural network is actually learning. One advantage this research has over neural networks is that as the programmer you know exactly what it knows. Instead, the problems are of the form, "How do I tell it what a circle is?" or "How do I have it determine what is noise that should be ignored?" The goal of this project is to create a program capable of taking in an image from a digital camera and identifying the tic-tactoe game. This is inspired from past work done for the Shelley Project which included playing tic-tac-toe (without the vision component) and the Shelley Integrated Environment (SIB). Various problems arose during implementation. The majority of these were system and API related. For others like the perspective correction and game detection, stepping away from the computer with pencil and paper in hand was invaluable. Through it all the goal of playing a game of tic-tac-toe against Shelley has finally become a reality.

Automated Annotation of Heegaard Diagrams

Dmitry Mogilevsky '03 — Thu, 09 Oct 2008 14:11:08 PDT

P-FASTUS: Information Extraction System Implemented in a Constraint Programming Language -SICStus Prolog

Rajen Subba '03 — Thu, 09 Oct 2008 14:11:07 PDT

P-FASTUS is an Information Extraction system developed in SICStus Prolog based on the implementation of FASTUS. It is program that extracts prespecified information such as the name of the companny, location and the position being advertised from" Job PostingIs'' in text files. The system is composed of different levels of processing phases that are implemented using finite-state transducers.

The State of Computer Science Facilities of Schools Across the United States that are Comparable to Illinois Wesleyan University

Sarah A. Bartz '93 — Thu, 09 Oct 2008 14:11:06 PDT

The purpose of this study is to compare Illinois Wesleyan's academic computer facilities with those at other comparable U.S. colleges and universities. This study also investigates factors affecting these facilities such as the methods other institutions use to deal with and plan for the constantly changing computer world. To obtain this information a survey was mailed to over 250 institutions across the U.S. that were considered comparable to IWU. These institutions include the Associated Colleges of the Midwest, the Oberlin Group, and some of the top colleges and universities in the U.S. as ranked in U.S. News and World Report's "America's Best Colleges". In order to facilitate a high return rate the survey was kept short and simple, letters were personalized whenever possible, a self-addressed and stamped return envelope was provided, and a copy of the results was promised to those who participated. Ninety-seven useable responses were received, which provided information such as: the number and type of computers and computer operating systems, the ability of students to remotely connect to the campus network, the facilities hours, the disciplines that use the facilities most, the number and type of staff, the reporting structure, the budget, the upgrading policy, and the education level of the director. These responses were then analyzed based on the size and budget of the school and compared with Illinois Wesleyan University.

Computer Program: General University Requirements Package

Abhishek Kejriwal '93 — Thu, 09 Oct 2008 14:11:05 PDT

The objective of my research project was to write a computer program in Turbo Pascal which would determine how many general university requirements a student has completed and what requirements he or she needs to complete in order to graduate. There are six degrees offered at Illinois Wesleyan University. They are: BA (Bachelor of Arts); BS (Bachelor of Science); BFA (Bachelor of Fine Arts); BSN (Bachelor of Science in Nursing); BM (Bachelor of Music) and BME (Bachelor of Music Education). There is a different set of criteria to be met for the completion of each of these degrees. The program processes student records and generates the appropriate check form.

Coding this problem and generating the output were extremely difficult because there are several classes, sub-classes, permutations and combinations possible to satisfy a requirement. Just to give a flavor of the complexity I will give an example. As stated earlier there are six degrees, each with different requirements. One of them is the BA Humanities is one of thirteen requirements a student has to meet to complete the BA degree. To meet the Humanities requirement the student must complete three courses from at least two of the following areas: literature, Philosophy and Humanities. There are seven successful ways to meet this requirement. A couple of these are two courses in literature and one in Philosophy or two in literature and one in Humanities and so on. Further, there are about 29 courses in literature, 23 courses in Philosophy and 5 courses in Humanities that qualify. In addition to this, the program has to check whether the course is valid. For a course to be valid, the course grade should not be Credit, No Credit, Withdrawn, Pass, Fail, Incomplete or Dropped and it should have a unit value of 0.7 or more. If the parts in the problem were mapped in a tree format there would be an incredible number of branches in the end. Ultimately there was the question of testing. To be sure that a program is working correctly one must perform a number of test runs. Some computer scientists describe testing as the most important part of the program. It was necessary to type in the records of students and generate results and then match the output to the results computed manually. Several such records had to be entered and any errors generated had to be ironed out. After a considerable amount of test data the package was finally generating outputs which exactly matched the results of outputs generated manually.

This program will be used in the Registrar's office at Illinois Wesleyan University starting this summer. After each semester the staff at the registrar's office will simply update the already existing data-base by adding any new students or adding courses to the records of the existing students. Copies of the form generated by the program after processing the checks will be sent to each student's advisor. Previously this entire process was accomplished manually and was extremely time consuming. With the help of this program the advisors will know at a glance where their advise's stand in terms of completing graduation requirements.

Limits of Diagonalization and the Polynomial Hierarchy

Kyle Barkmeier '06 — Mon, 21 Jul 2008 13:56:06 PDT

Determining the computational complexity of problems is a large area of study. It seeks to separate these problems into ones with "efficient" solutions, and those with "inefficient" solutions. Of course, the strata is much more fine-grain than this. Of special interest are two classes of problems: P and NP. These have been of much interest to complexity theorists for quite some time, because both contain many instances of important real-world problems, and finding efficient solutions for those in NP would be beneficial for computing applications. Yet with all this attention, there are still important unanswered questions about the two classes. It is known that P ⊆ NP, however it is still unknown whether P = NP or if P ⊂ NP. Before we discuss why this problem is so crucial to complexity theory, an overview of P, NP, and coNP is necessary.

The class P is a model of the notion of "efficiently solvable", and thus contains all languages (problems) that are decidable in deterministic polynomial time. This means that any language in P has a deterministic Turing Machine (algorithm) that will either accept or reject any input in n^k steps, where n is the length of the input string, and k is a constant. The class NP contains all languages that are decidable in nondeterministic polynomial time. A nondeterministic Turing Machine is one that is allowed to "guess" the correct path of computation, and seems to be able to reach an accept or reject state faster than if it was forced to run deterministically. It is unknown whether NP is closed under complementation because of this nondeterminism. It is quite easy to show a class of deterministically-solvable languages (such as P) is closed under complementation: we simply reverse the accept and reject states. This method is not viable for a nondeterministic machine, since switching the accept and reject states will result in machine that computes a completely different language. Thus the class coNP is defined as containing the complement of every language in NP.

In the rest of this paper we will present structural definitions of P and NP as well as present example languages from each. These structural definitions will give insight into the arrangement of the polynomial hierarchy, which is discussed in section 3. A diagonalization proof is presented in section 4, and an explanation of the general usage of diagonalization follows. In section 5, universal languages are defined and an important result from Kozen is given. In the final section, the limits of diagonalization as they pertain to P and NP are outlined, as well as the same limits for relativized classes.

The Virtual Beta: An Interactive Fish Using Java Script and CSS

Lauren B. Carroll '03 — Mon, 21 Jul 2008 13:56:06 PDT

My first version of the fish was written in C, because it was the language I was studying at the time. I wanted something that more people could view easily, however and one that could manage images more simply than C. I ended up choosing ]avaScript. At first I tried to translate my C program directly to ]avaScript, but I soon found this was too complex a task.

Unsupervised Learning to Improve Anomaly Detection

Daniel H. Garrette '06 — Mon, 21 Jul 2008 13:56:05 PDT

An intrusion detection system (IDS) is used to determine when a computer or computer network is under attack. Most contemporary IDSs operate by defining what an intrusion looks like and checking traffic for matching patterns in network traffic. This approach has unavoidable limitations including the inability to detect novel attacks and the maintenance of a rule bank that must grow with every new intrusion discovered. An anomaly detection scheme attempts to define what is normal so that abnormal traffic can be distinguished from it. This thesis explores the ways that an unsupervised technique called "clustering" can be used to distinguish normal traffic from anomalous traffic. This thesis will also explore an attempt to improve upon existing clustering algorithms to improve anomaly detection by adding in limited amounts of a posteriori knowledge.

Using Binary Space Subdivision to Optimize Primary Ray Processing in Ray-Tracing Algorithms

Mark Portolese '05 — Mon, 21 Jul 2008 13:56:04 PDT

Ray-tracing algorithms have the potential to create extremely realistic three-dimensional computer graphics. The basic idea is to trace light rays from the user through the computer screen into the hypothetical three-dimensional world. This is done to determine what objects should be displayed on the screen. Furthermore, these rays are traced back to the light sources themselves to determine shading and other photorealistic effects. However, without optimization these algorithms are slow and impractical. This paper explores the use of the classic binary space subdivision algorithm in order to speed up the process. Binary space subdivision is the use of binary trees to recursively partition the screen into rectangular areas which are then rendered separately. The algorithms were implemented using C++. The use of binary space subdivision dramatically improved the speed of the implementation in most cases, resulting in a doubled or tripled frame rate under favorable circumstances.

Rapid Face Detection using Independent Component Analysis

Aditya Rajgarhia '07 — Mon, 21 Jul 2008 13:56:02 PDT

Face detection is the task of determining the locations and sizes of human faces in arbitrary digital images, while ignoring any other objects to the greatest possible extent. A fundamental problem in computer vision, it has important applications in fields ranging from surveillance-based security to autonomous vehicle navigation. Although face detection has been studied for almost a decade, the results are not satisfactory for a variety of practical applications, and the topic continues to receive attention.

A commonly used approach for detecting faces is based on the techniques of "boosting" and "cascading", which allow for real-time face detection. However, systems based on boosted cascades have been shown to suffer from low detection rates in the later stages of the cascade. Yet, such face detectors are preferable to other methods due to their extreme computational efficiency.

In this thesis we introduce a novel variation of the boosting process that uses features extracted by Independent Component Analysis (ICA), which is a statistical technique that reveals the hidden factors that underlie sets of random variables or signals. The information describing a face may be contained in both linear as well as high-order dependencies among the image pixels. These high-order dependencies can be captured effectively by representation in ICA space. Moreover, it has been argued that the metric induced by lCA is superior to other methods in the sense that it may provide a representation that is more robust to the effect of noise such as variations in lightening. We propose that features extracted from such a representation may be boosted better in the later stages of the cascade, thus leading to improved detection rates while maintaining comparable speed. We present the results of our face detector, as well as comparisons with existing systems.