Convolutional neural networks is a class of neural networks that specializes in processing data that has a grid-like topology : images

In this figure, an image is represented as a grid that contains pixel values to denote how bright and what color each pixel should be

CNN architecture

Convolution Layer :

Performs a DOT between two matricees :

1. Matrix of learnable parameters : KERNEL

2. Restricted portion of receptive field

the kernel height and width will be spatially small, but the depth extends up to all three channels.

The kernel will slide across the height and width of the image to produce a two-dimensioal representation of the image know as the Activation map

• The activation map show the response of the kernel at any give spatial position of the image
• If we have an input size of WWD and
  • Spatial Size F
  • Stride S
  • Padding P
  • ⇒ Size of output padding

Trivial convolution networks use this method to represent the interaction between the input and output units VS Convolutional Neural Networks have sparse interaction

• Making the kernel smaller than the input
  • An image can have thousands of pixel but while processing it with the Kernel we can detect information using tens of pixels only.
  • Saves on memory and statistical efficiency of the model
• In a traditional neural network, each element of the weight matrix is used once and then never revisited, while convolution network has shared parameters. Weights applied to one paramterrs are the same elsewhere

Pooling