Compiler writers have developed various techniques, such as constant folding, subexpression elimination, loop transformation and vectorization, to help compilers in code optimization for performance improvement. Yet, they have been far less successful in developing techniques or cost models that compilers can rely on to simplify parallel programming and tune the performance of parallel applications automatically. This paper is the first of two-phase study to develop an analytical model that can be used to estimate the cost for sequential and parallel execution of array expressions on multicore architectures. While this paper discuss the possibility of developing a cost model to estimate the sequential execution of array expressions on a single CPU, the second part of the investigation shall focus on developing a model to estimate parallel execution of arrays on multicore platforms. The model presented in this paper is expected to be used by programming language compilers as complement component with the other model to estimate and subsequently decide whether to parallelize individual array expressions or not. The preliminary results which are presented here show that this model can give a satisfactory evaluation and high-precision estimation for the cost of executing a regular array expression on a single core processor.

Genetic algorithms (GAs) are based on techniques inspired by some aspects of natural science such as inheritance, reproduction and mutation, and they are used as optimization technique for searching large solution spaces. In computer science, for example, they could be used in data sorting and searching, circuit design and to improve application performance the quality of designed tools such as code generation. This paper looks at the possibility of using genetic algorithms to ameliorate the automatic construction of code generators. Experimental evidence is provided that the use of such algorithms can improve the quality of automatically constructed code generators.

neural network is considered as a nonlinear dynamic system consisting of a large number of simple processing elements interconnected in some manner with adjustable weighted strength. Neural networks provide qualitative and quantitative (analog, digital or logical) knowledge through information coding and decoding, and have powerful functions in learning and selforganization. These properties make neural networks considered to be more powerful in dealing with numerical data than other artificial intelligent systems like expert systems. However, the performance of neural networks depends deeply in number of factors including transfer function, number of hidden layers, number of nodes in hidden layers, input function, and weight function. In this article we present a comparative study of these factors and how they influence the performance of a system. 

This paper presents empirical methods for enhancing the accuracy of inductive learning systems. It addresses the problems of: learning propositional production rules in multi-class classification tasks in noisy domains, maintaining continuous learning when confronted with new situation after the initial learning phase is completed, and classifying an object when no rule is satisfied for it.

It is shown that interleaving the learning and performance-evaluation process allows accurate classifications to be made on real-world data sets. The paper presents the system ARIS which implements this approach, and it is shown that the resulting classifications are often more accurate than those made by the non-refined knowledge bases.

The core design decision that lies behind ARIS is that it employs an ordering of the rules according to their weight. A rule’s weight is learned by using Bayes’ theorem to calculate weights for the rule’s conditions and combining them. This model focuses the analyses of the knowledge base and assists the refinement process significantly.

The system is non-interactive, it relies on heuristics to focus the refinement on those experiments that appear to be most consistent with the refinement data set. The design framework of ARIS consists of tabular model for expressing rule weights, and the relationship between refinement cases and the rules satisfied for each case to focus the refinement process. The system has been used to refine knowledge bases created by ARIS itself, as well as to refine knowledge bases created by the RIPPER and C4.5 systems in ten selected domains.

Fuzzy decision forests aim to improve the predictive power of single fuzzy decision trees by allowing multiple views of the same domain to be modelled. Such forests have been successfully created for classification problems where the outcome field is discrete; however predicting a continuous output value is more challenging in combining the output from multiple fuzzy decision trees. This paper presents a new approach to creating fuzzy regression tree forests based upon the induction of multiple fuzzy regression decision trees from one training sample, where each tree will represent a different view of the data domain. The singular fuzzy regression trees are induced using a proven algorithm known as Elgasir which fuzzifies crisp CHAID decision trees using trapezoidal membership functions for fuzzification and applies Takagi-Sugeno inference to obtain the final predicted values. A modified version of Artificial Immune System Network model (opt-aiNet) is then used for the simultaneous optimization of the membership functions across all trees within the forest. A strength of the proposed method is that data does not require fuzzification before forest induction this reducing pre-processing time and the need for subjective human experts. Five problem sets from the UCI repository and KEEL repository are used to evaluate the approach. The experimental results have shown that fuzzy regression tree forests reduce the error rate compared with single fuzzy regression trees. 

In this paper we study the use of the Brill tagger [5,6,7] for tagging Modern Standard Arabic (henceforth MSA) text. The Brill tagger is a famous public domain part of speech tagger, originally designed for tagging English text by implementing machine learning approach through the method of transformation rules. IT had been adopted for other languages by many researchers [17,19,22]. Some modifications are needed on the learner and tagger that are written partly in perl and partly in C programming languages, and are run under the unix/linux operating system. The main change is done on the initial state tagger, which is used by both learner and tagger. A program is written using the lexical analyzer Lex to capture Arabic morphological structures, and then interfaced with both learner and tagger. The tagset used in this work is a revised version of that introduced by Khoja [11]. The revision included changing some of the tags for linguistic considerations and introducing some new tags to make the set more powerful, or to make up for limitations in the original tagset that hinder tagging some words. The corpus is obtained from two Jordanian magazines, and has to go through a series of editing steps. A collection of lexical rules and contextual rules are obtained and applied to Arabic text. The tagging accuracy of the resulting tagged text is measured now to be an average of up to about 84% for both known and unknown words, A rate, which is very promising for such a complex language and rich tagset. We still hope for better performance.

echnology enhanced Learning

This paper presents a method of generating realistic electricity load profile data for the UK domestic buildings. The domestic space features have been investigated excluding the heating and hot water systems. A questionnaire survey was conducted and the feedback were collected from a number of occupants at different intervals of times on daily bases in order to establish the probabilistic record of the estimated use of electrical appliances. The model concept of this study also considers the results of previous investigations such as that available in public reports and statistics as input data elements to predict the global domestic energy consumption. In addition, the daily load profile from individual dwelling to community can be predicted using this method. The result of the present method was compared to available published data and has shown reasonable agreement.