AI Talks 4: Neural Networks


Deep neural networks, convolutional neural networks, deep belief networks and deep stacking networks. We see these terms everywhere, but what do they mean? More importantly, what's a neural network?

In this post, we’re going to try explain in super simple terms how this side of machine learning works.


So, we have multiple images and we’d like the computer to automatically understand whether or not these images contain a cat, a dog or a face. To do that, we’ll build a neural network inside the computer. Think of a neural network as a machine with three different parts or layers. We call these layers the input layer, the middle layer and the output layer. Then, we add dots to each layer. These dots represent the neurons that will move the information through the network.

Screenshot (3).png

Here’s the tricky part: what’s a neuron? And what does it do in this machine? A neuron is what we call a processing unit. It’s connected to other neurons and propagates information. Let’s see how it works!

First, we connect neurons in each layer to the neurons in the following layer. Then, we find the desired output. Is the class of the image a cat, a dog, or something else?

Screenshot (4).png

Secondly, we connect the image to the input layer. What the input neurons do is look at the image from different perspectives.

Screenshot (5).png

Finally, the information starts propagating through the network and neurons get activated until the output layer is reached. In this case, the machine correctly classifies the image as a cat.

Screenshot (12).png

And the magic is done! Our neural network automatically recognised the subject of the image that we inputted into the machine.

So, to conclude - we learned three things:

  1. Neural networks are machines that process data with the help of multiple layers: input, middle and output layer.

  2. Neurons are simple processing units that can be tuned during the training phase.

  3. Neural networks can be trained by showing the machine the desired output and by adjusting the output of each neuron.


Thanks for reading this post. If you’d like to discover how neural networks can be used to speed up human labour-intensive tasks in your business, come talk to us!