Convolutional Neural Networks Architecture and Applications

In this blog, we will study Convolutional Neural Networks. Also, will learn it’s important terminologies and Convolutional Neural Networks architecture. Moreover, will study Convolutional Neural Networks applications and advantages.

 Introduction to Convolutional Neural Networks

In machine learning, CNN is a class of deep and feedforward learning. It has been successfully applied to analyze the visual imagery. CNN is made up of neurons that have learnable weights and biases. As it each neuron receives some inputs. Further, performs a dot product. Neural Networks still apply.

 History of CNN

As these convolutional networks were inspired by biological processes. The connectivity pattern between neurons is inspired by the animal visual cortex. Although, Individual cortical neurons are used to respond to stimuli that are only in a restricted region of the visual field known as the receptive field. Further, this field of different neurons partially overlap such that they cover the entire visual field.

 Important Terminologies of Convolutional Neural Networks

a. Shared Weights

In CNN, we have to replicate each filter across the entire visual field. Also, we use these replicated units to share the same parameterization and form a feature map. 

 We will show 3 hidden units belonging to the same feature map in the above figure. That unit are of weights of the same color that must be shared—constrained to be identical. We use gradient descent to learn shared parameters. Thus, we can say gradient of a shared weight is simply the sum of the gradients of the parameters being shared.

b. Details and Notation

Firstly we use convolution of the input image with a linear filter. Then will add a bias term and then apply a non-linear function to it.
If we denote the kth feature map at a given layer as, whose filters are determined by the weights and bias. Then the feature map is obtained as follows (for non-linearitie)
h^k_ij=tanh((W_k*x)_ij+b_k).

 Architecture of Convolutional Neural Networks

As we can say that CNN consists of a number of convolutional and subsampling layers. That layers are optionally followed by fully connected layers. As this input to a convolutional layer is a m x m x r image. Where, m is the height and width of the image and r is the number of channels. 

 For Example:
We have an RGB image that is having r=3. Also, this layer will have k filters of size n x n x q where:
n = is smaller than the dimension of the image ;
q = can either be the same as the number of channels r ;
As this size of filter gives rise to the locally connected structure. That are each convolved with the image to produce k feature maps of size m−n+1. Also, each map is then subsampled typically with mean or max pooling over p x p
In this contiguous regions, p ranges between 2 for small images and is usually not more than 5 for larger inputs.
After the convolutional layers- consider there may be any number of layers, one connecting another. Although, connected layers are identical to the layers in a standard multilayer network.

 Advantages of Convolutional Neural Networks

• Once trained, the predictions are pretty fast.
• With any number of inputs and layers, CNN can be train.
• Neural networks work best with more data points.

 Applications of Convolutional Neural Networks

a. Generally, we use CNN in Aircraft Ejection Seat Testing.

b. As effective Use of Word Order for Text Categorization with Convolutional Neural Networks.

c. Basically, semi-supervised Convolutional Neural Networks for Text Categorization via Region Embedding.

d. Generally, a Sensitivity Analysis of Convolutional Neural Networks for Sentence Classification.

e. Convolutional Neural Network for Modelling Sentences.

f. Deep Convolutional Neural Networks for Sentiment Analysis of Short Texts.