Nevin M. Darwish

Solution of systems of Boolean equations via the integer domain

This paper presents a new method for solving systems of Boolean equations. The method is based on converting the equations so that we operate in the integer domain. In the integer domain better and more efficient methodologies for solving equations are available. The conversion leads us to a system of polynomial equations obeying certain characteristics. A method is proposed for solving these equations. The most computationally demanding step is the repeated multiplication of polynomials. We develop a method for this problem that is significantly faster than the standard approach. We also introduce another variant of the method, the so-called hybrid approach, that leads to reduced memory requirements. Theoretical and experimental results indicate the superior performance of the proposed method and its variant compared to the competing methods. The proposed method is also validated by applying it to the problem of hardware verification.

Monitoring and synchronization for teamwork in GPGP

This paper addresses the problem of coordinating a group of agents involved in a team. To achieve flexible teamwork, agents should synchronize their work and monitor their performance to avoid redundant work. Generalized Partial Global Planning (GPGP) is one of the most common techniques used in coordinating cooperative agents, however, no technique is without limitations. Our work adopts some concepts of STEAM to overcome some of GPGP limitations. In particular, we suggest adding coordination mechanisms to GPGP and extending TAEMS, the model underlying GPGP, to facilitate such mechanisms. The work has successfully been implemented using JAF architecture. The coordination mechanisms are written as Soar rules where we implemented a JAF component that implements the Soar engine. Analysis of a case study is presented along with experimental results to illustrate the power of the proposed work.

Sketch recognition using particle swarm algorithms

Sketch recognition is defined as the process of identifying symbols that users draw using single or multiple strokes. Users draw strokes using a pen and the system immediately interprets their strokes as objects that can be easily manipulated. This paper uses Particle Swarm Optimization Algorithm (PSO) to divide the strokes the user draws into meaningful geometric primitives. These geometric primitives are grouped to formulate symbols which are further identified. The results show that using PSO improves segmentation results which guide the symbol recognition phase. This paper uses Support Vector Machines (SVM) classifier which further improves the final recognition accuracy.

A quantitative analysis of the alpha-beta pruning algorithm

Since the early 70s the question of efficiency of the alpha-beta pruning algorithm has been of interest to those who have been working on game-playing programs. In this paper, we present a quantitative study of the algorithm deriving estimates for its efficiency based on the scoring scheme suggested by Newborn [7].

Fuzzy-based centroid localisation in WSNs

One of the famous localisation methods in WSNs is the centroid method and its variants such as weighted centroid localisation WCL. This paper is considered as an extension to the WCL. The contributions of this paper are as follows: 1 the paper presents a fuzzy centroid algorithm as an extension to the weighted centroid localisation algorithm; 2 the paper extends the concept of weighted centroid localisation algorithm in WSN to include some of the new parameters such as power level PL, received signal strength RSS, and link quality indicator LQI; 3 the uncertainty principles are introduced to the WSNs localisation algorithms; 4 the paper implements a new fuzzy-based localisation algorithm using each of the newly added parameters for the purpose of analysis; 5 an extensive analysis has been done and recommendations to the network designers are given.

Bayesian based student knowledge modeling in intelligent tutoring systems

In this paper we present student knowledge modeling algorithm in a probabilistic domain within an intelligent tutoring system. The student answers to questions requiring diagnosing skills are used to estimate the actual student model. Updating and verification of the model are conducted based on the matching between the students and model answers. Three different approaches to updating are suggested, namely coarse, refined, and blended updating. In addition, different granularity levels are evaluated by changing the value of the updating step and the output of this parametric study is indicated. Results suggest that the refined model provides better approximation of the student model while utilizing blended model decreases the required trial numbers to model the student knowledge with limited reduction in accuracy.

FORN: First order rule based neural network

This paper presents FORN, the first automatic realization of mapping a first order rule based logic program onto a neural network. The work is based on Funahashis theorem and is of special interest to all applications where science advance necessitates the integration of new knowledge to existing systems. The proposal has been designed, implemented and successfully tested. Results are promising and suggest its future use for medical application.

Generating classification trees for small disjuncts using incremental gas

In this paper an Incremental GA techniques is proposed tosolve the problem of small disjuncts in classification trees. It is once applied on the disjuncts sorted in ascending orderand once in descending order with respect to their coverage. Both versions of the technique have been tested using benchmark datasets and the results are compared with thoseof other hybrid techniques.