1.1.4.5 Deep Neural Network Variants
Feed-forward neural networks: it is the most basic form of
neural networks where the flow only occurs from the input layer, they only have
one layer ,or at most one hidden layer,in this architecture there is no
back-propagation technique,they are usually used in face recognition
applications1.4
Figure 1.4: feed forward neural network /
Source[14]
Radial basis function neural networks: this networks have
preferably two layers,the relative
distance from any point to the center is calculated and the same
is passed to the next layer 1.5
8
1.1 Machine Learning
Figure 1.5: radial basis neural network
[20]
Multi layer perceptron(MLP): these networks usually have more
than three layers with fully connected nodes this architecture is usually used
for classifying data and speech recognition and various other applications
1.3
Modular neural networks: this architecture is a combination
of smaller networks that serve to achieve a common target ,which is very
helpful in breaking a big problem into small pieces 1.6
Figure 1.6: modular neural network /
Source[19]
Recurrent Neural Network: This architecture is unique for
it's use of loops where the output of one neuron is fed back to the same neuron
as an input allows the predicting of the output and the creation of small state
memory which is useful for video and audio applications 1.7
1.2 Natural Language Processing
Figure 1.7: recurrent neural network / Source
[3]
1.2 Natural Language Processing
Natural language processing (NLP) is a sub-field of
linguistics, computer science, and artificial intelligence concerned with the
interactions between computers and human language, in particular how to program
computers to process and analyze large amounts of natural language data. The
goal is a computer capable of "understanding" the contents of documents,
including the contextual nuances of the language within them.
Definition 1.5
4
1.2.1 What is Language?
Noam Chomsky gives the following definition to languages:
Definition 1.6
language is the inherent capability of native speakers to
understand and form grammatical sentences. A language is a set of (finite or
infinite) sentences, each finite length constructed out of a limited set of
elements. This definition of language considers sentences as the basis of a
language. -Noam Chomsky-
4
9
1.2.2 Why Natural Language Processing?
Natural language processing helps computers communicate with
humans in their natural language,NLP makes it possible for computers to read
text, hear speech and interpret it.
10
1.3 Computer Vision
1.2.2.1 Communication
communication can be defined as the act of interaction between
two entities , in the context of Natural Language processing it's the
interaction between humans and machines
1.3 Computer Vision
1.3.1 what is computer vision
inspired by the architecture of the vision systems in humans and
animals ,we create computer vision by using a sensing device and a interpreting
device as illustrated in figure 1.8 in the scope of
Figure 1.8: computer vision architecture /
Source[7] this project we will focus on the interpreting part.
· Note traditional Multi layer
perceptron network have are fully connected ,means each node is connected to
every and each neuron in the next and previous layer with can lead to an
explosion in the number of weights when the number offeatures is height ,this
will be a problem when we apply it on computer vision. each pixel in an image
will be a feature ,in an grey scale 256*256 image will produce 65,536 feature
meaning millions of weights ,this will only increase exponentially when we add
RGB images with more dimensions,for this exact purpose we use Convolution
neural network(CNN).
| 
