Simple explanation of convolutional neural network Understanding Convolutional Neural Networks CNN Made Easy Our deep learning series focuses on convolutional neural networks explained without complex mathematical jargons making it easy for even high school students to understand We are using an artificial neural network to handle the variation in this deep learning series To create a neural network we start by flattening the two dimensional representation of each hand written digit number into a one dimensional array We use the concept of a filter which consists of three components for the digit 9 head vertical line and diagonal filter These filters are used to detect tiny features in the images So how do we make computers recognize these tiny features Here comes the convolution operation or a filter operation When you see a number 1 or a number that is close to 1 it means there is a loopy circle pattern that indicates a particular feature In the case of a koala this feature could be an eye or a nose By applying a loopy pattern detector we get a feature map where the feature is activated Neural networks are used to classify a variety of inputs in a generic way The feature extraction is done using the convolution operation while the classification is carried out using the dense neural network The classification phase also involves a value operation Most commonly max pooling is used to achieve position invariant feature detection as it can detect features regardless of where they appear in the image People also use average pooling in some cases where the features can be averaged out The Convolutional Neural Network is a simple artificial neural network that detects features and reduces the dimension by using convolution There are three benefits of using the convolution operation It reduces the connection sparsity and overfitting It provides location invariant feature detection which means that the same feature can be detected even if it appears in different parts of the image It can detect various high level features by combining low level features A neural network has the ability to detect filters independently as part of its training using back propagation During the training of the convolutional neural network back propagation is used to determine the appropriate number and size of filters While you may specify the hyper parameters such as the number of filters and their sizes the network will learn the correct exit values for each filter on its own