StatusNeo

How to understand Neural networks in 5 minutes?

Introduction

Are you getting bored of scrolling through your social media? Of course not!

Why do you think this happens, we scroll through Facebook, YouTube or Instagram for hours and we don’t get bored, somehow they manage to keep it interesting for every single individual even though everyone has different interests.

Let’s discuss about a very simple mathematical function which has revolutionised the whole World. It is called Neural Networks. It is the basic building block of Deep Learning. Let’s say you are on a vacation in Japan but don’t understand the language, Worry not! You have an amazing device in your hand and that’s your phone which you can use as a medium to interact with locals. It will make your vacation smooth with less hiccups.

There are already so many things which are possible because of neural networks, like Recommendation Systems, Voice recognition, Instagram filters, Facial Recognition, Signature Verification or handwriting analysis, Stock Market Prediction, Aerospace, Defense, Robotics, etc. I will not bore you with the details, that will make the introduction as long as the actual blog.

What are Neural networks?

As the name suggests, a network of neurons can be termed as a neural network, our brain is made up of motor neurons, sensory neurons and interneurons combined. Similarly, Artificial Neural Networks(ANN) is inspired from Brain Structure. ANN is simply a set of algorithms interconnected in a manner similar to a brain.

Here as you can see that there are a bunch of handwritten digits, We can identify and distinguish among them very easily, courtesy of our visual cortex. But let’s say we want to create a program which can identify handwritten digits without any human help. How do we do that? We can create an image dataset with respective annotations and train a model on that. Sounds very easy!

But wait, we are familiar with image data and annotations , but what is a model exactly and how does it work? Normally, models are treated as a black box where we just input the data and it gives a predicted output. We are going to open this black box and see what exactly happens in there. I will try to avoid too many mathematical calculations in order to make it less boring. In the next section, we will be taking a look into the classical case of handwritten digit recognition which is easy to understand.

How does it work?

Before getting into the neural network let’s first understand what is a neuron in this context. It is basically a number between 0 and 1, it is also called activation. Let’s look at an example to understand it better.

In the image shown above, we see a handwritten number 9. This image has 28 X 28 pixels. If you look more closely , you will notice that there are different numbers assigned to each pixel. If the pixel is black then 0 and if white it is 1 or if is not fully illuminated, the number will be between 0 and 1. There are total 784 pixels (28×28) in this image and each pixel has an activation. These 784 pixels will create the first layer of the neural network.

For every image, there is a pattern of activations, those activations will create some patterns in the next layer and so on. Since, there are only 10 digits , the last layer of the neural network will consist 10 activations one for each digit. The neuron with the highest activation will be the predicted number by the model. for instance , if 3 has the highest activation number , then the predicted number is 3.

The real question is how one single layer affects the next layer. It involves the sum of multiplication of weights with the activation and subtracting bias from it and then multiplying it with a sigmoid function. It sounds complicated I know , we will be covering this in the next blog.

How this neural network is related to the social media addiction?

How this neural network is related to the social media addiction?

As I mentioned earlier, neural networks are basic building blocks for deep learning. Social media giants or big e-commerce companies, they all use Recommender Systems . They basically study the data you consume and make suggestions or advertisements accordingly. For example, on youtube, if you start watching certain kind of videos , it will start suggesting you more videos of that kind. This is the reason why we are always glued to our phone screen.