Convolutional Neural Networks — CNN

In the previous article, we discussed the Convolution Operation. In this article, we discuss how the convolutional operation or in general the Convolutional Neural Network relates to Neural Networks and Deep Learning in general. This article covers the content discussed in the Convolutional Neural Networks module of the Deep Learning course and all the images are taken from the same module.

The relation between the convolutional operation and the neural networks:

Let say we are given an input image of size ‘30 X 30 X 3’(i.e 2700 pixels) and our task is to predict if this image contains a signboard or not(Binary Classification problem), if we unfold these 2700 values into one single vector then this vector would act as the input layer in the fully connected neural network depicted below where each neuron in this input layer is connected to all the neurons in the next layer:

After the input layer, we have a few hidden layers containing some neurons say 100 neurons each. And then we have the output layer containing only one neuron which would be a sigmoid neuron.

For each of the 2700 neurons(input values), we are multiplying these values with certain weights and computing the pre-activation value of the neuron highlighted in the below image: