CNN Introduction¶
A Convolutional Neural Network (ConvNet/CNN) is a Deep Learning algorithm which can take in an input image, assign importance (learnable weights and biases) to various aspects/objects in the image and be able to differentiate one from the other. The pre-processing required in a ConvNet is much lower as compared to other classification algorithms. While in primitive methods filters are hand-engineered, with enough training, ConvNets have the ability to learn these filters/characteristics.
Where can we use CNNs ?¶
CNNs can be used for image classification purposes.
Working¶
The image is resized to an optimal size and is fed as input to the convolutional layer.
There exists a filter or neuron or kernel which lays over some of the pixels of the input image depending on the dimensions of the Kernel size.
The Kernel actually slides over the input image, thus it is multiplying the values in the filter with the original pixel values of the image (aka computing element-wise multiplications).
Pooling layer is used which generally decreases the dimensions of the images but not the image channels.
Finally, an activation function generally ReLU is applied on it.
Advantages and Disadvantages¶
Advantages¶
Some of the advantages of a CNN over an MLP for images are that they are more location invariant due to the convolution meaning only a small portion of the image is paid attention to at a time and max pooling also helps with this. Also, CNN’s mean that you don’t need a ridiculous number of neurons which would take forever to train. Also, they are based off how the human brain’s image processing works and copying nature is normally good.
Disadvantages¶
CNN do not encode the position and orientation of object. Lack of ability to be spatially invariant to the input data. A ConvNet requires a large Dataset to process and train the neural network. If the CNN has several layers then the training process takes a lot of time if the computer doesn’t consist of a good GPU.
Reference¶
https://machinelearningmastery.com/convolutional-layers-for-deep-learning-neural-networks/
https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/