You are on page 1of 18

Agenda of Presentation

• Introduction to CNN
• What is a Convolutional Neural Network?
• How CNN recognize images?
• Layers in Convolutional Neural Network
Introduction to CNN
• Yann LeCun is the Pioneer of Convolution Neural Network.
• He built the first CNN called LeNet in 1988
• It was used for character recognition tasks like reading zip codes,
digits.
What is CNN?
• CNN is a feed forward neural network that is generally used to analyze
visual images.
• Convolution operation form the basis of any CNN.
• In CNN every image is represented in the form of arrays of pixel
values.
Layers in CNN
1. Convolution Layer
2. ReLU Layer
3. Pooling Layer
4. Fully Connected Layer
5. Output Layer
Convolution Layer
• A convolution layer has a number of filters that perform convolution
operation
• Every image is considered as a matrix of pixel values
• Sliding the filter matrix over the image and computing the dot
product to detect patterns
ReLU Layer
• Once the feature maps are extracted, the next step is to move them
to a ReLU Layer
• ReLU Layer:
1. Performs element wise operation
2. Sets all negative pixels to 0
3. Introduces non-linearity to the network
4. The output is the rectified feature map.
Pooling Layer
• The rectified feature map now goes through a pooling layer. Pooling is
a down-sampling operation that reduces the dimensionality of the
feature map.
Pooling Layer
• Pooling layer is used to identify different parts of an image like edges,
corners etc.
Flattening
• Flattening is the process of converting all the resultant 2 dimensional
array from feature map into a single long linear vector.
Flattening
Flattening
• Structure of the neural network so far
Fully Connected Layer
• The flattened matrix from the pooling layer is fed as input to the fully
connected layer
Fully Connected Layer
• The flattened vector from the pooling is then connected to a few fully
connected layers which are same as Artificial Neural Networks and
perform the same mathematical operation.
• For each layer of the ANN the following calculation takes place

g(Wx + b)
Fully Connected Layer
g(Wx + b)
Where,
x– is the input vector
W- is the weight matrix
b-is the bias vector
g – is the activation function, which is usually ReLU
This calculation is repeated for each layer
Output Layer
• After passing through the fully connected layer, the final layer uses
the softmax activation function which is used to get probabilities of
the input being in the particular class
Softmax Function.
• The softmax function is a function that turns a vector of K real values
into a vector of K real values that sum to 1.
• The input values can be positive, negative, zero or greater than one,
but the softmax function transform them into values between 0 and 1
• So that they can be interpreted as probabilities.
Softmax Function
All the zi values are the
elements of the input
vector to the softmax
function

The input vector to The term in the


the softmax function K is the number denominator is the
of classes in the The standard normalization term. It
multi class exponential ensures that all the
classifier. function is output values sum to 1
applied to each i.e. in the range (0-1)
element of the
input vector. This
gives a positive
value above 0.
The End

You might also like