revamp reference file

Thesis_Configurations.tex

@@ -11,6 +11,7 @@
 \newtheorem{theorem}{Theorem}
 \newtheorem{acknowledgement}[theorem]{Acknowledgement}
 \newtheorem{definition}[theorem]{Definition}
+\usepackage[numbers,sort&compress]{natbib}
 %-------------------------------------------
 \usepackage{hyperref}
 \usepackage{multirow}
@@ -36,7 +37,7 @@
 \titlespacing*{\paragraph}
 {0pt}{3.25ex plus 1ex minus .2ex}{1.5ex plus .2ex}
 
-
+\usepackage{filecontents}
 % -------------------------------------------------------------
 \usepackage[dvipsnames, table]{xcolor}
 \usepackage{mathtools}

chapters/ch_literature.tex

@@ -6,13 +6,13 @@ \section{Literature Review}\label{Ch:Literature}
 %
 \subsection{Deterministic (Algorithmic) Approach}\label{sec:Determ_Algor_Appr}
 
-\cite{Abuata2016RuleBasedAlgorithmFor} present the most related work to our topic, classifying Arabic poetry according to their \textit{meters}. However, they have not addressed it as a \textit{learning problem}; they have designed a deterministic five-step \textit{algorithm} for analyzing and detecting meters. The first step and the most important is to have the input text carrying full diacritics; this means that every single letter must carry a diacritic, explicitly. The next step is converting input text into \textit{Arud writing} using \textit{if-else} like rules. \textit{Arud writing} is a pronounced version of writing; where only pronounced sounds written. Then metrical \textit{scansion} rules applied to the \textit{Arud writing}, which leaves the input text as a sequence of zeros and ones. After that they defined each group of zeros and ones as a \textit{tafa'il}, so now we have a sequence of \textit{tafa'il}. Finally, the input text classified to the closest meter to the \textit{tafa'il} sequence. 82.2\% is the classification accuracy on a relatively small sample, only 417 verse.
+\cite{Abuata2016RuleBasedAlgorithm} present the most related work to our topic, classifying Arabic poetry according to their \textit{meters}. However, they have not addressed it as a \textit{learning problem}; they have designed a deterministic five-step \textit{algorithm} for analyzing and detecting meters. The first step and the most important is to have the input text carrying full diacritics; this means that every single letter must carry a diacritic, explicitly. The next step is converting input text into \textit{Arud writing} using \textit{if-else} like rules. \textit{Arud writing} is a pronounced version of writing; where only pronounced sounds written. Then metrical \textit{scansion} rules applied to the \textit{Arud writing}, which leaves the input text as a sequence of zeros and ones. After that they defined each group of zeros and ones as a \textit{tafa'il}, so now we have a sequence of \textit{tafa'il}. Finally, the input text classified to the closest meter to the \textit{tafa'il} sequence. 82.2\% is the classification accuracy on a relatively small sample, only 417 verse.
 
 \cite{Alnagdawi2013FindingArabicPoemMeter} has taken a similar approach to the previous work, but it replaced the \textit{if-else} by \textit{regular expressions} templates. This approach formalized the \textit{scansion}s, \textit{Arud} based on lingual rules related to pronounced and silent rules, which is directly related to \textit{harakat} as \textit{context-free grammar}. Only 75\% from 128 verses were correctly classified. 
 
 \cite{Kurt2012AlgorithmForDetectionAnalysis} have taken a similar approach but worked on detecting and analyzing the \textit{arud} meters in Ottoman Language. They convert the text into a lingual form in which the meters appear. Their First Step, Converting Ottoman text transliterate to Latin transcription alphabet (LTA). After that, they feed the text to the algorithm which uses a database containing all Ottoman meters to compare the detected meter extracted from LTA to the closest meter found in the database which saved the meters.
 
-Both~\cite{Abuata2016RuleBasedAlgorithmFor} and~\cite{Alnagdawi2013FindingArabicPoemMeter} have common problems,
+Both~\cite{Abuata2016RuleBasedAlgorithm} and~\cite{Alnagdawi2013FindingArabicPoemMeter} have common problems,
 
 \begin{enumerate}
 \item \textbf{The size of the test data} which cannot measure the accuracy for any algorithms they have constructed because it is a very small dataset. Also, a 75\% total accuracy of 128 verses is even worse.
@@ -28,7 +28,7 @@ \subsection{Deterministic (Algorithmic) Approach}\label{sec:Determ_Algor_Appr}
     \textbf{Ref.}& \textbf{Accuracy}& \textbf{Test Size} \\
     \midrule
     \cite{Alnagdawi2013FindingArabicPoemMeter}   & 75\%     & 128\\
-    \cite{Abuata2016RuleBasedAlgorithmFor}      & 82.2\%   & 417  \\
+    \cite{Abuata2016RuleBasedAlgorithm}      & 82.2\%   & 417  \\
     This article   & 96.38\%  & 150,000 \\
     \bottomrule
   \end{tabular}

chapters/sec_background_deep_learning.tex

@@ -614,7 +614,7 @@ \subsection{Recurrent Neural Networks (RNNs)}\label{Sec:RNN}
  As we explained, RNN works on sequential data, and the idea is to predict new output not only based on the input data vector but also, other input vectors. Due to the recurrent structure in RNNs, it tends to suffer from long-term dependency to simplify this point let’s have an example, the following sentence \\
  \textit{Waleed Yousef who is Associate Professor at Helwan University and teaching Data Science courses and its dependencies \textbf{\underline{was}} got Ph.D. in Computer Engineering from GWU at 2006.}.
 
- In the previous example, to predict the word was is depending on long dependency to check if Waleed is singular or not to be consistent. Also, shows how some problems need the long-term dependencies handling.[Bengio et al.,1994]\cite{Bengio_ et_ al} showed that Basic RNNs has a problem in long-term dependency.  Another problem which may happen into basic Neural Networks is gradient exploding. One of the side-effects of gradient exploding is exponentially large gradient which causes our parameters to be so large. So, the Neural Networks parameters will have a server problem. Another fetal problem with Basic Neural Networks is overfitting problems [Zaremba et al., 2014]\cite{Zaremba_et_al}.
+ In the previous example, to predict the word was is depending on long dependency to check if Waleed is singular or not to be consistent. Also, shows how some problems need the long-term dependencies handling.[Bengio et al.,1994]\cite{Bengio_1994} showed that Basic RNNs has a problem in long-term dependency.  Another problem which may happen into basic Neural Networks is gradient exploding. One of the side-effects of gradient exploding is exponentially large gradient which causes our parameters to be so large. So, the Neural Networks parameters will have a server problem. Another fetal problem with Basic Neural Networks is overfitting problems [Zaremba et al., 2014]\cite{Zaremba_et_al}.
 
  So, to solve this learning problem [Hochreiter and Schmidhuber, 1997] introduced Long Short-Term Memory which helps to reduce the dependency problem using memory cell and forget gate.
 \begin{figure}[!t]