Rosalina Abdul Salam

Harmony Search Based Supervised Training of Artificial Neural Networks

This paper presents a novel technique for the supervised training of feed-forward artificial neural networks (ANN) using the Harmony Search (HS) algorithm. HS is a stochastic meta-heuristic that is inspired from the improvisation process of musicians. Unlike Backpropagation, HS is non-trajectory driven. By modifying an existing improved version of HS & adopting a suitable ANN data representation, we propose a training technique where two of HS probabilistic parameters are determined dynamically based on the best-to-worst (BtW) harmony ratio in the current harmony memory instead of the improvisation count. This would be more suitable for ANN training since parameters and termination would depend on the quality of the attained solution. We have empirically tested and verified our technique by training an ANN with a benchmarking problem. In terms of overall training time and recognition, our results have revealed that our method is superior to both the original improved HS and standard Backpropagation.

Enhancing the low quality images using Unsupervised Colour Correction Method

Underwater images are affected by reduced contrast and non-uniform colour cast due to the absorption and scattering of light in the aquatic environment. This affects the quality and reliability of image processing and therefore colour correction is a necessary pre-processing stage. In this paper, we propose an Unsupervised Colour Correction Method (UCM) for underwater image enhancement. UCM is based on colour balancing, contrast correction of RGB colour model and contrast correction of HSI colour model. Firstly, the colour cast is reduced by equalizing the colour values. Secondly, an enhancement to a contrast correction method is applied to increase the Red colour by stretching red histogram towards the maximum (i.e., right side), similarly the Blue colour is reduced by stretching the blue histogram towards the minimum (i.e., left side). Thirdly, the Saturation and Intensity components of the HSI colour model have been applied for contrast correction to increase the true colour using Saturation and to address the illumination problem through Intensity. We compare our results with three well known methods, namely Gray World, White Patch and Histogram Equalisation using Adobe Photoshop. The proposed method has produced better results than the existing methods.

Protein Conformational Search Using Bees Algorithm

Proteins perform many biological functions in the human body. The structure of the protein determines its function. In order to predict the protein structure computationally, protein must be represented in a proper representation. To this end, an energy function is used to calculate its energy and a conformational search algorithm is used to search the conformational search space to find the lowest free energy conformation. In this paper, the Bees Algorithm, i.e. a Swarm Intelligence based algorithm inspired by the foraging behaviour of honey bees colony, is adapted to search the protein conformational search space. The algorithm was able to find the lowest free energy conformation of Met-enkephaline using ECEPP/2 force fields.

Protein Tertiary Structure Prediction Using Artificial Bee Colony Algorithm

Proteins are essential for the biological processes in the human body. They can only perform their functions when they fold into their tertiary structure. Protein structure can be determined experimentally and computationally. Experimental methods are time consuming and high-priced and it is not always feasible to identify the protein structure experimentally. In order to predict the protein structure using computational methods, the problem is formulated as an optimization problem and the goal is to find the lowest free energy conformation. In this paper, Artificial Bee Colony algorithm (ABC) is a swarm intelligence based optimization algorithm inspired by the behaviour of honey bee foraging. This algorithm is adapted to search the protein conformational search space to find the lowest free energy conformation. Interestingly, the algorithm was able to find the lowest free energy conformation for a test protein (i.e. Met enkephaline) using ECEPP/2 force fields.

Removing salt-and-pepper noise from binary images of engineering drawings

Removing noise in engineering drawing images is important before applying image analysis processes. Noise should be removed while keeping the fine detail of the image intact. A noise removal algorithm that can remove noise while retaining fine graphical elements is presented in this paper. The algorithm studies the neighborhood of thin lines before choosing to remove or retain it. Real scanned images from GRECpsila03 and GRECpsila05 arc segmentation contests corrupted by 15% uniform salt/pepper noise are used in this experiment. Objective distortion measurements including PSNR and MSE show that our algorithm gives better quality images compared with other methods.

Parallel K-Means Clustering Algorithm on DNA Dataset

Clustering is a division of data into groups of similar objects. K-means has been used in many clustering work because of the ease of the algorithm. Our main effort is to parallelize the k-means clustering algorithm. The parallel version is implemented based on the inherent parallelism during the Distance Calculation and Centroid Update phases. The parallel K-means algorithm is designed in such a way that each P participating node is responsible for handling n/P data points. We run the program on a Linux Cluster with a maximum of eight nodes using message-passing programming model. We examined the performance based on the percentage of correct answers and its speed-up performance. The outcome shows that our parallel K-means program performs relatively well on large datasets.

Multiple sequence alignment using genetic algorithm and simulated annealing

This paper presents the combination of genetic algorithm and simulated annealing to solve multiple sequence alignment (MSA) assignment. Genetic algorithm will try to find a new region of feasible solution while simulated annealing will act as an aligning improver. There are several aspects that must be taken into consideration such as the representation, evaluation function and operator. Simulated annealing also helps to prevent local minima problem. Sequence similarity plays a major role in Bioinformatics and molecular biology. Significant results were produced from the prealignment and genetic algorithm phase.

Blood cell image segmentation using hybrid K-means and median-cut algorithms

In blood cell image analysis, segmentation is crucial step in quantitative cytophotometry. Blood cell images have become particularly useful in medical diagnostics tools for cases involving blood. In this paper, we present a better approach on merging segmentation algorithms of K-means and Median-cut for colour blood cells images. Median-cut technique will be employed after comparing best outcomes from Fuzzy c-means, K-means and Means-shift. We used blood cell images infected with malaria parasites as cell images for our research. The result of proposed method shows better improvement in terms of object segmentations for further feature extraction process.

Training Feed-Forward Neural Networks Using a Parallel Genetic Algorithm with the Best Must Survive Strategy

Feed-Forward Artificial Neural Networks (FFANN) can be trained using Genetic Algorithm (GA). GA offers a stochastic global optimization technique that might suffer from two major shortcomings: slow convergence time and impractical data representation. The effect of these shortcomings is more considerable in case of larger FFANN with larger dataset. Using a non-binary real-coded data representation we offer an enhancement to the generational GA used for the training of FFANN. Such enhancement would come in two fold: The first being a new strategy to process the strings of the population by allowing the fittest string to survive unchanged to the next population depending on its age. The second is to speed up fitness computation time through the utilization of known parallel processing techniques used for matrix multiplication. The implementation was carried on master-slaves architecture of commodity computers connected via Ethernet. Using a well-known benchmarking dataset, results show that our proposed technique is superior to the standard in terms of both the overall convergence time and processing time.

A risk identification architecture pattern based on Bayesian network

Web projects continually face a high degree of visible developmental failure. Such inefficiencies in web projects cost losses in terms of money and time thus negatively impacting on business growth. Web development projects have an array of unique risks different from those found in traditional software development projects. Many failures associated with web projects are the consequences of poor awareness of the risks involved and the weak management of these risks. Although many approaches have been proposed to overcome this shortcoming, there is still a huge gap between these approaches and actual industry needs. This research aims to improve risk management through the design of a risk identification architecture pattern based on Bayesian network to help avoid risks in the web project, and improve the chances of managing critical risks beforehand. The paper presents a case study of the practical applications of the proposed model in an actual web project.