Improving the Learning Rate of the Back Propagation Algorithm by Aitkin Process
Khalil
K. Abbo
author
text
article
2013
eng
The Back Propagation Algorithm is used for training feed Forward Multilayer Neural Networks (FFMNN).But often this algorithm takes long time to converge since it may fall into local minimu, for this reason we need a long time to train the network. The suitable choice of the learning rate helps us to escape from slow convergent for the BP and reduce the time of learning. In this paper, we derived a new adaptive learning rate for the BP algorithm, our derivation is based on the Aitkin's process. The most important distinct feature of our approach is the computing of the learning rate needs only first order derivatives and is suitable for large training sets and large networks. Its efficiency is proved on the standard test functions including heart , XOR and function approximation problems .
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
1
10
https://stats.mosuljournals.com/article_75434_dce726e9feafe4d9280db92bd8bb3eab.pdf
dx.doi.org/10.33899/iqjoss.2013.75434
Using the Canonical Correlation Analysis Technique for Imaging Dimensionality Reduction in Multisource Land sat Images
text
article
2013
eng
The Canonical Correlations Analysis technique (CCA) was suggested in the dimensionality reduction images for the multivariate multisource data applied in remote sensing . These techniques transform multivariate multiset data into new orthogonal variables called Canonical Variates (CVs) . This research uses the LANDSAT-5 TM data for the set of multivariate multispectral correlation at fixed points in time . The results show maximum similarity for the low- order canonical variates and minimum similarity for the high- order canonical variates .
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
1
18
https://stats.mosuljournals.com/article_75422_994084f6beed8792826349c2b72da5ed.pdf
dx.doi.org/10.33899/iqjoss.2013.75422
A GENETIC ALGORITHM TO MINIMIZE THE MAKESPAN FOR THREE MACHINE FLOW SHOP SCHEDULING
MANAL
ABDULKAREEM ZEIDAN
author
text
article
2013
eng
The aim of this paper is to propose genetic algorithm to finding the optimal schedule with minimum makespan for n jobs in flow shop environment with three machines.
In this pape, a new approach for selection, a new crossover operation (3PLOX) and a new stopping criteria based on the lower bound of the makespan (LB) are proposed.Also, a new procedure of calculating the make span for n jobs at processing by three machine in flow shop environment is suggested. In order to examine the effectiveness of the proposed GA, a comparison was made with Johnson's algorithm. After the application on several problems which generated randomly by uniform distribution, the results showed that the proposed GA is better than Johnson's algorithm with rate 94% in finding the optimal sequence for scheduling jobs which gives optimal makespan.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
11
26
https://stats.mosuljournals.com/article_75435_e211a5d57d5d85cb177cdc21173844bc.pdf
dx.doi.org/10.33899/iqjoss.2013.75435
text
article
2013
eng
This paper addresses the important problem of prediction of the daily rainfall. The rainfall situation is modeled as a Markov chain of four states: the lack of rain, light rain, moderate rain, and heavy rain. After the presentation of the issue of prediction in general terms, two methods are suggested to predict the next rain situations by relying on the information that the real cases providevo. Two algorithms are suggested for prediction; the first algorithm relies mainly on the matrix of transition, and can be used to find predictors for any order of Markov chain, the second algorithm, the autoregressive models. These algorithms are applied on the time series of daily rain falling on the City of Mosul, after modeling this time series as a Markov chain of four states. Linear and non-linear autoregressive models are fitted. It is clear from the application of these two algorithms that they give reasonable predictors terms of ease of use.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
19
32
https://stats.mosuljournals.com/article_75423_94db82ffcf516708864c5655e9bf1d22.pdf
dx.doi.org/10.33899/iqjoss.2013.75423
text
article
2013
eng
This paper deals with the issue of link prediction in the analysis of Internet networks. In the theoretical side of this research the Markovian modeling of web log files is displayed and then the issue of compression and clustering of transition probability matrix is discussed. Then the problem of the use of the Markovian model in the link prediction is displayed . The practical side of this research includes the application of certain aspects of the theory discussed in the present research on the web links at the University of Mosul and the link prediction of them. Also a practical experiment is made to highlight the importance of compression of transition matrix.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
33
50
https://stats.mosuljournals.com/article_75424_2137bf540af9b2dbd0cea7409ea7f7ae.pdf
dx.doi.org/10.33899/iqjoss.2013.75424
Preparing a dynamic model for water filtering process in Mosul city
text
article
2013
eng
In this research the scientific dynamic process is studied concering
water filtering in Mosul city through employing the input and output variables represented by some tests which were performed on raw water
before the filtering process, to be treated later through the numerous filtering stages.
The tests involved: the electrical conduction, turbidity, water PH, and water temperature The same tests were conducted after the filtering process to guarantee that water is good for human consumption and to guarantee the validity of these tests as well and also to ensure that these tests meet the Iraqi Standard specifications before adding the chlorine. Four models were constructed through this study concerning the dynamic filtering process of the transfer functions with a single input for the four tests. And then models were built for the four transfer functions as input multiple series before filtering. Each of the four tests represented the output series by itself and consequently four models of transfer functions were obtained with multiple inputs and a single output variable in order to observe the range of impact for the output variable with the multiple input variables and using these models in Forecasting and then comparing the single input models and the multiple input models by means of using the standards of Forecasting control accuracy. The results showed that the electrical conduction model and the multiple input PH value were better than the results of the single input and the turbidity model of single input was better in comparison with the results of the multiple input. In addition to that, results showed that the temperature model for the single and the multiple input are similar.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
51
74
https://stats.mosuljournals.com/article_75425_1c3a18a134c376d4c56d65bfe9131047.pdf
dx.doi.org/10.33899/iqjoss.2013.75425
text
article
2013
eng
In this paper we derive the reliability of a series system (which consists of k-components)in a new way based on the following definition of a series system " a series system fails if any component fails" .
Because components failure ( or number of failures) follow Poisson distribution , and the main principle of this distribution is that in a very small interval ∆t ,
the probability of more than one failure is zero. Therefore the failure of a series system at the time of failure would be due to only one component.
In the application side a simulation run for a series system for hypothetical values of the parameters of two parameter Wiebull distribution , and the reliability of such system is calculated.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
75
98
https://stats.mosuljournals.com/article_75427_0734dd683c7fa2e068a87f0e5e829de3.pdf
dx.doi.org/10.33899/iqjoss.2013.75427
Study Series Stocks Exchange by using PMRS , ANN , and ARIMA
text
article
2013
eng
It is well known that the processes of countries development is the process of planning and setting future plans. This demands the use of advanced statistical styles. So , we have conducted a comparison between three methods,namely;Box Jenkins (ARIMA),and Pattern Modeling Recognition System (PMRS)which is a modern to know the relation between the current values and the previous ones and the Artificial Neural Networks (ANN).These three methods were application on the Iraqi monetary market for the year 2006.The Results showed superiority of artificial neural networks through error criteria used.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
99
118
https://stats.mosuljournals.com/article_75428_26021a98d60cb5f3bda99674ccc5d2e6.pdf
dx.doi.org/10.33899/iqjoss.2013.75428
Proposed Method for Bootstrap Technique to Estimate the Results of some Men’s Competitions in the Coming Olympic Games (2012, 2016) in case of Using Incomplete Data
text
article
2013
eng
In this paper, the application of Bootstrap Method in predicting the results of Skipping and Jumping competitions for men in the next two sessions of Olympic Games 2012 and 2016. The researcher tried to use the results achieved by all holders of Olympic medals (top three) in the competitions of Skipping and Jumping (wide, triple, high, and Pole vault) in six consecutive Olympic Games Seoul Olympics (1988) is selected as a starting point since it did not face Olympic sanctions from some countries that took place in two sessions of Moscow (1980) and Los Angeles (1984), and here arose the idea to use the method of bootstrap since the time series is too short and it is not easy to determine its probability distribution, and therefore it is difficult to use classical methods in the estimation process as well as the sample size is too small.
This paper will illuminate the way for those interested in the preparation of the teams participating in the four Skipping and Jumping competitions for well planning based on solid scientific base of the number and level of participants in these competitions based on the expected accomplishments brought through the paper in the two Olympic Games, where the first one will start within days.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
119
137
https://stats.mosuljournals.com/article_75429_f8e47c6b59f67db9b39d1beba5d9b3ea.pdf
dx.doi.org/10.33899/iqjoss.2013.75429
Creating a Genetic Algorithm to Test the Optimum of the
text
article
2013
eng
The standard number is considered an important statistical tool to measure change quantity in economic phenomena through two different periods or places ; the first is named the basis period and the second is the comparison period (or basis place , comparison place ) . The optimal standard number is that number which performs the two tests of chronological reflex in the basis and the reflex in the agent . In this research a genetic algorithm was created to test the standard numbers which were programmed by MATLAB9b language which saved a lot of time and efforts done annually to test the optimum of these numbers.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
138
149
https://stats.mosuljournals.com/article_75430_faf8168881a4bc2777c5e22f241f0322.pdf
dx.doi.org/10.33899/iqjoss.2013.75430
Person recognize using Elman neural network by face image
text
article
2013
eng
The purpose of this system is to recognize aperson`s using face image, and the uses of this system at airports and other offices which use surveillance cameras that take pictures of views and different angles and there are qualities in every person face which do not change and are not affected by the way the picture is taken. In this paper, the remittance waves (Wavelet)are used in order to withdraw more accurate details of
the image, and then features were extracted based on the seven resolution and the four statistical properties (location measurement mean, standard deviation and skewness and kurtosis), which address the problems of image capture by the surveillance cameras.
This system has been in the database configuration to many people's each person has seven different images in order to configure a database of training. I have been using Elman neural network as a tool to identify people, were trained network (Elman) on the line ratio(0.001)and the number of training courses(1079)and by know(92%). And used the language of MATLAB R2010a in this system with a database of the type of Microsoft Excel.
IRAQI JOURNAL OF STATISTICAL SCIENCES
University of Mosul
1680-855X
13
v.
1
no.
2013
150
167
https://stats.mosuljournals.com/article_75433_b55383aa7facb4a1308af5f3c83a7913.pdf
dx.doi.org/10.33899/iqjoss.2013.75433