The elemental constructing block of CNNs

My recent articles have been a series on neural networks where we go from the straightforward perceptron to complicated architectures and the way to take care of common problems in deep learning. Should you have an interest, be at liberty to envision the series here:

Neural Networks

One exciting area neural networks have made significant strides in is computer vision. Think AI for self-driving cars and face recognition!

Nevertheless, the regular fully connected neural network that almost all people find out about will not be suitable for a lot of real-life image recognition tasks. It really works on the famous MNIST dataset, but it surely has small images of 28×28 pixels.

High-definition (HD) images have 1280×720 pixels. That’s roughly 1,000,000 pixels, which might mean 1,000,000 neurons within the input layer. Not to say the tens of millions of weights required for the hidden layers, rendering regular neural networks unsuitable attributable to the dimensional complexity.

So, what will we do?

Convolutional Neural Networks!

Convolutional neural networks (CNN) are the gold standard for nearly all of computer vision tasks today. As a substitute of fully connected layers, they’ve partially connected layers and share their weights, reducing the complexity of the model.

For example, for every neuron in a totally connected neural network layer, we might require 10,000 weight of a picture of 100×100 pixels. Nevertheless, a CNN can have only 25 neurons to process the identical image.

In this text, we’re going to dive into the basic constructing block behind CNNs, convolution.

Should you are having fun with this text, be certain that to subscribe to my YouTube Channel!

Click on the link for video tutorials that teach you core data science concepts in a digestible manner!