Mohammad R. Sayeh

A neural network approach to robust shape classification

Abstract A neural network approach for the classification of closed planar shapes is described. The primary foci are the development of an effective representation for planar shapes which may be used in conjunction with neural nets, the selection of a suitable neural network structure, and determining training methods to increase the degree of robustness in classification. A three layer perception using backpropagation is initially trained with contour sequences of noisefree reference shapes and its generalization capability is demonstrated. The network is then gradually retrained with increasingly noisy data to improve the robustness of the classifier. The advantages and improvement in robustness using this extended training scheme are shown and typical classification results are presented.

Convergence and limit points of neural network and its application to pattern recognition

A novel neural network model, based on the gradient system theory, is introduced. The proposed design approach solves the problem of parasitic limit points. This could have significant impact on many potential applications, particularly in the area of pattern classification/recognition. The design approach, the development of the Lyapunov function, the stability analysis, and the convergence characteristics of the neural network are discussed in detail. Design examples and simulation results are presented to illustrate the design process and the convergence characteristics of the proposed neural network. One example shows its application in pattern recognition. >

A self-organizing map and its modeling for discovering malignant network traffic

Model-based intrusion detection and knowledge discovery are combined to cluster and classify P2P botnet traffic and other malignant network activity by using a Self-Organizing Map (SOM) self-trained on denied Internet firewall log entries. The SOM analyzed new firewall log entries in a case study to classify similar network activity, and discovered previously unknown local P2P bot traffic and other security issues.

A Web-based information system for stock selection and evaluation

We have developed an expert system, called PORSEL (PORtfolio SELection system), which uses an effective set of rules to select and evaluate stocks on the Internet. At present, the PORSEL consists of three components: the information center, the fuzzy stock selector, and the portfolio constructor. The purpose of the information center is to provide representation of several technical indicators such as candlestick charts, moving average of closing prices, and price trends. The fuzzy stock selector evaluates the listed stocks and then assigns a composite score for each stock. The portfolio constructor generates the optimal portfolios for the selected stocks. A client/server model is implemented which allows users to communicate with the PORSEL program on a server computer at a remote site in a user-friendly manner. The results of simulation show that PORSEL outperformed the market almost every year during the testing period.

Real-Time Vector Quantization and Clustering Based on Ordinary Differential Equations

This brief presents a dynamical system approach to vector quantization or clustering based on ordinary differential equations with the potential for real-time implementation. Two examples of different pattern clusters demonstrate that the model can successfully quantize different types of input patterns. Furthermore, we analyze and study the stability of our dynamical system. By discovering the equilibrium points for certain input patterns and analyzing their stability, we have shown the quantizing behavior of the system with respect to its vigilance parameter. The proposed system is applied to two real-world problems, providing comparable results to the best reported findings. This validates the effectiveness of our proposed approach.

Adaptive competitive self-organizing associative memory

This work presents the design of an adaptive competitive self-organizing associative memory (ACSAM) system for use in classification and recognition of pattern information. Volterra and Lotkas models of interacting species in biology motivated the ACSAM model; a model based on a system of nonlinear ordinary differential equations (ODEs). Self-organizing behavior is modeled for unsupervised neural networks employing the concept of interacting/competing species in biology. In this model, self-organizing properties can be implicitly coded within the systems trajectory structure using only ODEs. Among the features of this continuous-time system are: 1) the dynamic behavior is well-understood and characterized; 2) the desired fixed points are the only asymptotically stable states of the system; 3) the trajectories of ACSAM derived from the weight activities of the gradient system have no periodic or homoclinic orbits; and 4) the heteroclinic orbits that exist between equilibrium states are structurally unstable and can be removed by small perturbations.

Operation of an associative memory

Abstract This paper describes the operation of an associative memory (LYAM) governed by only ordinary differential equations, useful for pattern clustering. Several computer simulations illustrate its operation as an unsupervised classifier, vector quantizer, and content-addressable memory.

Comparison of wavelength conversion in SOA and EDFA fiber ring systems

We report on two types of wavelength conversion techniques that are based on gain saturation effect in semiconductor optical amplifier (SOA) and erbium doped fiber amplifier (EDFA). In these amplifiers the gain saturation occurs when the optical density at the gain medium is high enough to result in depletion of the population inversion by stimulated emission. In each case, the fiber ring laser is assembled using a variable fiber coupler, a narrowband optical filter and the gain medium. For external input power values higher than the determined threshold value of the ring resonator, the gain will be saturated. Because the wavelength of the external laser is different from the oscillating wavelength of the ring resonator, the optical power at the output of the resonator is drastically decreased (low-state). On the other hand, when the input of the external laser is below the threshold value the output power of the resonator increases (high-state). In our experiment the operating wavelengths of the ring resonators are 1314 nm and 1553 nm for the SOA and EDFA respectively. The input signal is modulated around the threshold value for frequencies of 20 MHz and 1 MHz and resonator lengths of around 8 m and 16 m for the SOA and EDFA cases respectively. Both systems exhibit high contrast modulation of 41 dB and 33 dB at the output port for the low/high states of the SOA and EDFA ring lasers respectively.

Nonlinear dynamics of two-wave coupling process

We present a closed-form solution for the two-wave coupling process, resulting in a solution with an interesting part, the arbitrary function Lambda(t), not being considered in the previous works. In the photorefractive medium, it has been shown that a moving grating will produce a phase shift in the coupling constant. To our knowledge, this has not been derived directly from the grating dynamic equations. We will also show that the temporal dynamics does not have any asymptotically stable equilibrium points of an exponential form. As an example, we have developed resonance conditions resulting in a new expression for the frequency detuning of a resonator with a photorefractive gain medium.

Neural network-based multimode fiber-optic information transmission

A new technique for transmitting information through multimode fiber-optic cables is presented. This technique sends parallel channels through the fiber-optic cable, thereby greatly improving the data transmission rate compared with that of the current technology, which uses serial data transmission through single-mode fiber. An artificial neural network is employed to decipher the transmitted information from the received speckle pattern. Several different preprocessing algorithms are developed, tested, and evaluated. These algorithms employ average region intensity, distributed individual pixel intensity, and maximum mean-square-difference optimal group selection methods. The effect of modal dispersion on the data rate is analyzed. An increased data transmission rate by a factor of 37 over that of single-mode fibers is realized. When implementing our technique, we can increase the channel capacity of a typical multimode fiber by a factor of 6.