[CNN] What is Convolutional Neural Network?

A Convolutional Neural Network (CNN) is one of Deep Learning algorithms.

 

• Why Convolution Layer? not Multi-Layer Perceptron?

    To use multi-Layer percetron, it need to use fully connected method. It means input features lose their own local information. Then machine learning model will be in trouble to catch the exact information. This leads to poor performance.

    On the other hands, Convolution Layer observes data's local area! It helps to understand around information. 

Example bewteen FC and Convolution

• How to work Convolution Layer?

    Convolution Layer composes of kernel filters. The below figure shows how to work kernel filter on input data. To operate, there are some information such as kernel filter's size, stride. Kernel filter is overlapped to the input feature. It extracts local information of input data as one output. It moves upto stride's size. 

How to work convolution operation in CNN

 

• Real Convolution Filter Operation

Below figure, it shows how to operate convolution filter in real situation.

How to operate ML parameters in real