UHD REU Program > Project Description

Project Description

1. Experimentation with a Modified Dijkstra’s Algorithm -- Providing Physical Layer Security to an All Optical Network

Mentor: Dr. Shengli Yuan

Student: Daniel Stewart

The research explores a new routing algorithm that will provide physical security in an all optical network environment. This process includes research into the fundamentals of network routing, specifically Dijkstra’s algorithm. It begins by simulating Dijkstra’s algorithm. Further research was then needed to understand the inherent weaknesses with all optical networks (AONs). Lastly, I modified Dijkstra’s algorithm to simulate physical security of network traffic on AONs. The research begins with adapting Dijkstra’s algorithm to network routing problems. Dijkstra's algorithm produces a shortest path tree by solving the single-source shortest path problem for a graph with nonnegative edge path costs. This algorithm gives us the basis for network routing as well as the basis for the research. Next, we focused on the weaknesses of AONs. This research covered malicious high frequency disruptions, signal amplification and attenuation, and tapping. We continued the experimentation by adding security features to a modified Dijkstra’s algorithm. One modification allows the algorithm to protect certain paths by dynamically increasing their costs. A second modification allows the algorithm to remember the origin of routes. This ensures that overlapping routes of different origin are assigned higher costs. The remainder of the research will focus on the latest modification that allows a route’s origin to be remembered.  If successful, reserved routes will be assigned high costs, overlapping reserved routes will be assigned even higher costs, and crossing two or more reserved routes of different origin will be assigned the highest costs.

Final Presentation: Optical Network Security

2. Cascaded Classifier for Automatic Crater Detection

Mentor: Dr. Tomasz Stepinski, Dr. Wei Ding

Student: Henry Z. Lo

This project is to design a high-performance algorithm to automatically detect craters from satellite images of Mars. The idea is to adapt object recognition algorithms which were successful in classifying other objects for the task of crater detection. The performance of a cascaded Adaboost classifier which has been used successfully in face detection is evaluated. Initial results for our crater detection training and test sets show promising results. Some modifications to the features used by the classifier and the training set improved performance. One particularly effective technique was to train the classifier on images of craters which contain a little bit of the surrounding area as well as the crater itself; this provides the classifier with identifying information about the edges and rims of craters, which was not otherwise available. It is believed that further performance gains can be realized by improving the quality of the ground truth and creating custom features specifically for crater detections. Though the focus is currently on the cascaded Adaboost classifier, the performance gains that arise from optimizing the features and training set of this classifier is expected to generalize to other classifiers. Should the cascaded Adaboost classifier prove ineffective, other classifiers which have been useful for object classification tasks will be adapted for the class of crater detection.

Final Presentation: Cascaded Classifier for Automatic Crater Detection

3. Coreference Resolution in Clinical Texts

Mentor: Dr. Ping Chen

Students: David Hinote and Carlos Ramirez

The 2011 I2B2 challenge involves co-reference resolution in medical documents. Concept mentions have been annotated in clinical texts, and the mentions which co-refer in each document are to be linked by co-reference chains.  Normally, there are two ways of constructing a system to automatically discover co-referent links. One is to manually build rules for co-reference (a rule based approach), and the other is to use machine learning systems to build the rules automatically (a machine learning approach).  There have been systems developed for co-reference resolution by various organizations. The aim of this study was to use the systems which are publicly available, as well as build a system tailored for this challenge using a rule based system, and test these systems on the data provided for this challenge. The study shows the publically available systems do manage to find some of the co-referent links, and the rule based system developed for this challenge performs well finding the majority of the co-referent links.  The system that was used to provide the final outputs for the challenge had 89.6% overall performance average.

Final Presentation: Coreference Resolution

4. Brain State Analysis

Mentor: Dr. Hong Lin

Student: Tremaine A. Rawls

The primary objective for this summer research project is to analyze the Electroencephalography (EEG) data, brain waves statistics, in order to develop models to differentiate different brain states. This project aims to read these brainwaves when a person is in meditation state, allowing us to decipher the difference between a person in mediation and a person who is not. Since the brain is an electrochemical organ using electromagnetic energy to function, electrical activity emanates from the brain and can be displayed in the form of brainwaves when the data is rendered and configured accordingly. The brain wave analyzer contains 14 different nodes that are scattered across the four main sections of the brain (frontal lobe, thalamus, parietal lobe, and reticular formation) that picks up on different types of brain waves, alpha, beta, delta, and theta waves. The EEG data is being collected from our collaborators Dr. Wanli Ma and Dr. Dat Tran at the University of Canberra, Australia, and Dr. Beyongsang Oh at the University of Sydney. The Graphical system will include several elements that permit the information to be read into the program, and modified as needed. The brain wave analyzer will be the basis for distinguishing key distinctions in each node at any given frame to interpret brain states and analyze them.  Final Presentation: Brain State Analysis

5. UMLS Word Search

Mentor: Dr. Ping Chen

Student: Brian Ng

My objective for summer research is to find terms in documents based on human anatomical structures found in the human body. The terms in the documents will be compared to a database of terms that Unified Medical Language System (UMLS). The main focus for developing a log of the terms of the UMLS. The next stage was using a high level computer language like Java to do the simple comparison of words taken from an external source to the log of terms from the UMLS. When a first term of a UMLS word or phrase is located and they are equivalent. Tools to aid in searching for terms will be regular expressions and sample papers. These tools help to aid in accuracy for finding words and therefore improves the overall results to a person using the program. The next word for comparison is taken and compared to the next word of the phrase. When all words are equivalent a counter is updated and the word found is printed to the screen. The program will be refined to obtain better accuracy and recall throughout the summer or for a specific end product. After optimization, the rest of the time spent will be on how the words can be displayed either in another text document with tags for the words found or an interface showing connection of the word to a 3D model of the human body. By summers end, I will be more competent in coding and searching for expressions.

Final Presentation: UMLS Word Search

6. Information Extraction from Medical Extraction

Mentor: Dr. Ping Chen

Student: Alexander Barsky

Using GATE (General Architecture for Text Engineering), we are attempting to extract valuable information from patient's charts and organize the information in a coherent manner. Generally a patient's chart is heavily documented. This makes locating important information difficult and time consuming. Using the latest techniques in pattern recognition, data mining, co-referencing and other, we are attempting to pinpoint the information we need quickly and efficiently. We are utilizing JAPE grammar rules to establish the parameters that are necessary to extract information to obtain an accurate, concise picture of a person's medical information. After we have the final annotated documents, GATE outputs the document in XML format. We use C# in conjunction with XLST stylesheets to strip the XML documents of all superfluous data. This is done so that the data can be presented in an easily understandable format. Also it allows for simple and quick data manipulation. While information extraction through JAPE grammar allows us to retrieve an accurate portion of necessary information, it does have a few setbacks. No matter how many specific grammar rules are written for extraction, there will always be necessary information left out. This is not acceptable when a patient's medical history is to be considered. The best method to remedy this is through machine learning. We are using WEKA to add classifications to the project. This is only the first step to attempt to utilize machine learning as it is still in its elementary stages of practical usage.

Final Presentation: Information Extraction From Medical Records

7. Investigation of a 3D Simulation Model

Mentor: Dr. Ongard Sirisaengtaksin

Student: Melissa Greenlee

The goal of the project is to investigate a 3D model and create an object within the model that can navigate through the model. The object navigation will be determined by time and coordinates. The first part of the research included determining how to create a 3D model and which approach will suit these purposes best. Google SketchUp was chosen to create the model in that the model can be exported as an .fbx file and the ease of use of SketchUp. Then, we researched how to render the model using XNA software as well create the object within the 3D environment. The 3D simulation can also be created with keyboard and mouse controls to navigate the object. The advantage of using XNA is that it is easy to change models without having to drastically change the system.

Final Presentation: 3D Simulations