Pandya, Abhijit S.

Several neural network applications solving practical problems in communications are presented. A neural network algorithm to select paths through a three stage switching network is developed. An analysis of the dynamics of the neural network and a convergence proof are provided. With the help of computer simulations, a four dimensional region for the valid combinations of the neural network parameters was discovered. An analysis is performed to determine the characteristics of this region. The behavior of the neural network algorithm for different switching network configurations and varying traffic patterns were investigated. The effect of initial state of the neural network and heuristic improvements to the algorithm is provided. A comparative analysis of the neural network path selection algorithm against a sequential search method is also given.

The existing Group Method of Data Handling (GMDH) algorithm has characteristics that are ideal for neural network design. This thesis introduces a new algorithm that applies some of the best characteristics of GMDH to neural network design and develops a Pruning based Regenerated Network by discarding the neurons in a layer which don't contribute for the creation of neurons in next layer. Unlike other conventional algorithms, which generate a network which is a black box, the new algorithm provides visualization of the network displaying all the neurons in the network. The algorithm is general enough that it will accept any number of inputs and any sized training set. To show the flexibility of the Pruning based Regenerated Network, this algorithm is used to analyze different combinations of drugs and determine which pathways in these networks interact and determine the combination of drugs that take advantage of these interactions to maximize a desired effect on genes.

The fuzzy vault scheme introduced by Juels and Sudan [Jue02] was implemented in a fingerprint cryptography system using COTS software. This system proved to be unsuccessful. Failure analysis led to a series of simulations to investigate the parameters and system thresholds necessary for such a system to perform adequately and as guidance for constructing similar systems in the future. First, a discussion of the role of biometrics in data security and cryptography is presented, followed by a review of the key developments leading to the development of the fuzzy vault scheme. The relevant mathematics and algorithms are briefly explained. This is followed by a detailed description of the implementation and simulation of the fuzzy vault scheme. Finally, conclusions drawn from analysis of the results of this research are presented.

At the turn of the new millennium, the focus of Information Technology Management turned to Information and Systems Security, as opposed to competitive advantage investment. In catering to the security needs of various firms and institutions, it is seen that different entities require varying Information Security configurations. This thesis attempts to utilize Risk Analysis, a commonly used procedure in business realms, to formulate customized Firewalls based on the specific needs of a network, subsequently building an effective system following the "Defense in Depth" strategy. This is done by first choosing an efficient Risk Analysis model which suits the process of creating Firewall policies, and then applying it to a particular case study. A network within Florida Atlantic University is used as an experimental test case, and by analyzing the traffic to which it is subject while behind a single Firewall layer, a specific Security Policy is arrived at and implemented.

This research aims at proposing a model for visual pattern recognition inspired by the neural circuitry in the brain. Our attempt is to propose few modifications in the Alopex algorithm and try to use it for the calculations of the receptive fields of neurons in the trained network. We have developed a small-scale, four-layered neural network model for simple character recognition as well as complex image patterns, which can recognize the patterns transformed by affine conversion. Here Alopex algorithm is presented as an iterative and stochastic processing method, which was proposed for optimization of a given cost function over hundreds or thousands of iterations. In this case the receptive fields of the neurons in the output layers are obtained using the Alopex algorithm.

One of the major components of any pervasive system is its proactive behavior. Various models have been developed to provide system wide changes which would enable proactive behavior. A major drawback of these approaches is that they do not address the need to make use of existing applications whose design cannot be changed. To overcome this drawback, a middleware architecture called "Concord" is proposed. Concord is based on a simple model which consists of Lookup Server and Database. The rewards for this simple model are many. First, Concord uses the existing computing infrastructure. Second, Concord standardizes the interfaces for all services and platforms. Third new services can be added dynamically without any need for reconfiguration. Finally, Concord consists of Database that can maintain and publish the active set of available resources. Thus Concord provides a solid system for integration of various entities to provide seamless connectivity and enable proactive behavior.

With rapid growth of the World Wide Web, web performance becomes increasingly important for modern businesses, especially for e-commerce. As we all know, web server logs contain potentially useful empirical data to improve web server performance. In this thesis, we discuss some topics related to the analysis of a website's server logs for enhancing server performance, which will benefit some applications in business. Markov chain models are used and allow us to dynamically model page sequences extracted from server logs. My experimental studies contain three major parts. First, I present a workload characterization study of the website used for my research. Second, Markov chain models are constructed for both page request and page-visiting sequence prediction. Finally, I carefully evaluate the constructed models using an independent test data set, which is from server logs on a different day. The research results demonstrate the effectiveness of Markov chain models for characterizing page-visiting sequences.

In this thesis, we proposed a low power and high performance architecture for 1-bit full adder design. The proposed architecture was proven to offer a wide range of performance ability in terms of power consumption and speed. We implemented the architecture using a 2-input 2 multiplexers, an XOR and an XNOR gate. The proposed architecture and the Standard full adder were designed and simulated using Lasi, Winspice and Silos. Silos, a logic simulation environment was used in the design and verification of the proposed architecture and the standard full adder that were modeled with Verilog hardware description language. Lasi was used for the layout design of the proposed architecture and the standard full adder. After the layout, both the architectures were compiled separately using LASICKT and a corresponding .CIR file was generated. The .CIR file was imported and executed into WINSPICE3 for the simulation of the circuit.

In this thesis, importance of Intelligent Data Repository (IDR) and its real life applications are studied. We proposed an IDR for oncology applications which can handle large datasets and which can be used on both the intranet and the Internet. It is designed to provide one or multiple medical institutions on a global scale a common platform for patient care and consultation. The proposed application consists of two key models, Body Surface Area model and Search model, which are described in detail and their results are discussed. We have implemented the proposed IDR for oncology application using ColdFusion MX. Existing research in this area have been studied and compared. Framework of the proposed IDR, structure, front-end user interface and back-end database schema of the proposed oncology application are explained in this thesis.

K-means algorithm and Kohonen network possess self-organizing characteristics and are widely used in different fields currently. The factors that influence the behavior of K-means are the choice of initial cluster centers, number of cluster centers and the geometric properties of the input data. Kohonen networks have the ability of self-organization without any prior input about the number of clusters to be formed. This thesis looks into the performances of these algorithms and provides a unique way of combining them for better clustering. A series of benchmark problem sets are developed and run to obtain the performance analysis of the K-means algorithm and Kohonen networks. We have attempted to obtain the better of these two self-organizing algorithms by providing the same problem sets and extract the best results based on the users needs. A toolbox, which is user-friendly and written in C++ and VC++ is developed for applications on both images and feature data sets. The tool contains K-means algorithm and Kohonen networks code for clustering and pattern classification.