2004 Projects
Intelligent Data Fusion and Error Reduction for Sensor Networks
Advisor: Dr. Elhajj
Sensor networks are attracting attention in several fields. However, the feasibility of such networks faces several challenges, two of which are data fusion and error reduction. The goal of this project was to develop data fusion and high level error correction algorithms for sensor networks. The algorithms developed are scalable and general, and thus can be applied to networks of any size using any type of sensors. The data fusion procedure devolved results in significant reduction of data sent without reducing the amount of information provided. This allows for real-time remote monitoring of information across low bandwidth connections such as the Internet. The high level error reduction is accomplished using a probability matrix and results in a significant amount of error elimination. A sensor network capable of tracking object motion is constructed to evaluate the performance of the two algorithms.
Lecture Notes |
Using Hybrid Neural Networks for In Silico Metabolic Modeling
Advisor: Dr. Zohdy
Neural networks are effective intelligent computational tools; simultaneously there is a need for better metabolic simulation programs which handle recent gene expression information from genome of microorganisms. This project focussed on developing a novel suite of Neural Network programs, which are designed to model metabolic pathways due to initial concentrations and environmental inputs. Five features were added to optimize these neural programs. They include: polymorphism, inheritance, encapsulation, context, and hints of available biological knowledge. With these key features accuracy in the program outputs increased to +/- 0.02. Therefore, the goal of establishing the suite of optimizing these neural programs was highly successful
Lecture Notes |
Handling Pose Estimation Using Distribution of Normals
Advisor: Dr. Li
In many aspects of modern medicine, dealing with 3D medical images has become the main source of information for the purpose of surgeries, diagnosis and treatment. Pose estimation has been an important aspect when dealing with the classification of 3D objects. Pose estimation is the primary problem that causes less accuracy during the research when classifying three dimensional objects. Even though the researchers have developed methods to solve pose estimation problem, the results tend to be less efficient which gives large error when dealing with the classification of 3D objects. This project developed a new method for handling the pose estimation problem. The efficiency and capacity of least possible error of this method will be compared with that of previously introduced methods. Even though this method can also be applied to 2D objects, since our research is based on automatic identification of 3D anatomical structures, our concentration is on 3D medical objects. This method works for any 2D or 3D objects with complicated structures.
Lecture Notes |
Verification of Cognitive Models
Advisor: Dr. Mili
Cognitive modeling is the creation of computer-based processes that mimic human problem-solving and task execution using existing cognitive theories. Creating correct cognitive models requires a lot of work and is very error-prone. The use of a specification and tool support would be very beneficial. The specification needs to be able to handle all cases of cognitive modeling. The tool in turn also has to be versatile enough to at least handle most if not all of the possible models. This project focussed on finding a specification methodology that will handle most of the possible models, then designing a tool that checked a specification for correctness. This is done by specifically checking for incoherence and inconsistencies. It also outputs a couple of sentences that can be evaluated for correctness. The sentences are generated to point out certain problems that are discovered as common mistakes.
Lecture Notes |
Speech Emotion Recognition
Advisor: Dr. Sethi
As part of a voice stream analysis system, the goal of this project is to automatically determine the emotion of a speaker. This determination can assist in various human interaction tasks, ranging from homeland security applications, to directing irate customers to the best customer service representative.
Lecture Notes |
