How to Explain Neural Network Concepts to a Sixth Grade Student

This article is an attempt to explain the concepts of deep learning in a comprehensible way that even school going students can understand

In fact, Deep Learning is not a difficult concept to understand. For complete beginners, I have already explained this Artificial Intelligence, Machine Learning and Deep Learning in this five minutes article. This will help you to understand some basic things in this topic.

Things we are going to learn in this article.

1. What is Neural Network?
2. How it works?
3. What is Activation Function?

Deep Learning is like an ocean. I will cover all the other topics related to this in my subsequent articles.

What is Neural Network?

Let’s interpret it word by word. ‘Neural’ means ‘Neuron’. Neuron is a cell in our body and we have approximately 10¹⁵ neurons in our whole body. Neuron is responsible for passing information within our whole body system. For instance, if you touch something, a neuron from your hand will pass the information to the neurons in brain and says that you are touching something. It passes the information through the network, it means there is connection between your neuron in your hand and neuron in your brain. In biological term, we call that network as ‘Axon’. Same like that, all the neurons in our body are interconnected. ‘Network’ is the fancy of calling connection between the two or more things.

Imagine cells as humans inside our body in the size of less than one mm, they are working inside our body to keep our body functioning all the time in a smooth manner. Neurons are workers sent by our creator to work for you(work for your body, more precisely). They all are interconnected and working with a great coordination. You don’t need to pay them.

Okay, but what this neuron is doing in deep learning?

I researched about this and got to know lots of new things. Let’s here this story, Once scientists were fed up of this machines programmed activities. It was not doing anything other than the instructions they have passed to it(in the form of code). But these greedy scientists expected more than that from this innocent machines. They thought it would do more if it had some intelligence like human beings. They compared these machines with human beings. Machines are faster and precise than humans. But we don’t need to instruct humans all the time like we do for machines. They thought, if something which has some intelligence and at the same time faster and precise as well, it would do anything in this world. So they decided to inject intelligence into these machines artificially. But at that times, they didn’t know how to do that. So they researched how this human brains works and they came to know about this neural network in our brain. They just wanted to make one like this, so they just emulated this neural network mechanism and injected it artificially into machines. They achieved success after a long process. That’s how this deep learning born and doing lots of unbelievable stuffs in this world nowadays.

How Neural Network Works?

Before knowing about the working mechanism of neural networks, hear this story. There were twenty students in a sixth grade classroom. One day, their teacher was in a good mood, so he just wanted to teach something to his students in an innovative way. He brought all his students to the playground of the school and make everyone standing in a queue. Now students are standing in a queue with one meter distance between each of them. That teacher picked a word from the dictionary and he ensures that word shouldn’t be familiar to a sixth grade student. He pronounced that word in a proper manner to the last guy in the queue. Now the game is, each student should pass that word to the person who is standing in front of them till it reaches the first guy in the queue. After it reaches the first guy, the first guy should try to write that word with exact spelling in a paper. That teacher will collect the paper from that first guy and compare that spelling with the spelling of that word in the dictionary. Finally he will rate it based on how the written spelling of the word in the paper matches with the original one. After the comparison, he writes his mark in that paper and give it to the first guy. In the first attempt, he rated that word spelling with 4 out of 10. Now the first guy should pass this paper to the student behind him and every one in the queue should keep passing it till it reaches the last guy again. Now every one in the queue wants to increase the mark in the second chance. That last guy again passes the word with some clarity. This time everyone performs well in this passing game to increase the rating score. Everyone gets the input with utmost care and pass it with same clarity to the student standing before him. Now they get 7 out of 10. They again taking chance to increase the score even better in the same way till they gets the perfection.

This is a story happened in a school. Why I told this story in this context is, this is the way how neural network exactly works.

Let’s see the architecture of a neural network, there is an input layer and output layer. Inside there are lots of hidden layers. That input layer is the last guy in the queue and output layer is the first guy in the queue. Hidden layers are other students in the classroom. In neural networks, these all hidden layers are extremely smart persons. This below picture will help you to understand Neural Network better.

Above story is a base to easily understand how neural network works. Let’s see the actual working mechanism of neural network works. Input layer is the one which gets input from the user. For example, if you want to know the price of a house in some location, think what are the possible ways to find the actual worth of that building.

1. Assessing the building condition.
2. Counting the number of bedrooms.
3. Considering the location of the building.

There are a lot of parameters. In machine learning, we usually call it a features. Because these are things which are going to assist us to assess the actual worth of the building. These features are independent, because the value of the location will not going to affect the value of the bedroom count. But what we are going to find is also a feature which is dependent. Because, we wants to know the price of the house, it changes based on the values of these independent features.

But here is one thing we have to think about. These all features are going to contribute in the same level to calculate the price of the house? No…

Here I tell you the reason. If we want to know how good the recently released movie is, we ask the different persons in our gang about the movie who already watched it. Consider, if you get a mixed reviews from all of them, but what if a person in your gang is a cinephile who gives positive review about that. We value his words more than the other persons, right? In machine learning jargon, it means you are giving more weightage to his words/review.

In the same manner, some independent features affect your dependent feature more than the rest of the features. In first, our neural network doesn’t know how to set weightage for all the features. It is very much naïve to do that in the first shot itself. So initially it will fix some equal weightage to all the features in random manner, exactly like our perception about all the students in the classroom being neutral in the first day of a new school. It will get all these input features from input layer, it will send it to all the hidden layers in the neural network to process it. Same like the last student in the queue got the input word from the teacher and sent it to all the students in the queue.

The feature with weightage gets add and gives a final result to the activation function at the end of hidden layers(it presents before output layer). He is the smartest one in the neural network who changes the processed data into a meaningful form . Consider him as the 1st guy in the standing in the queue of twenty persons who writes the spelling of the word in paper(we give input to 20th guy). This is called forward propagation in neural network. In students passing game, after getting rating from teacher, students pass the result in the backward direction till it reaches the last guy right? That’s how exactly Back propagation of error in neural network works.

What is Activation function?

Yes, this is the important function in the complete neural network process. It gives us the output in a reasonable format how we want to get the output.

For example, consider you buy some vegetables from a market and give it to your chef and you ask him to make a vegetable soup. After collecting all the vegetables from you, your chef giving that to his subordinates and asking them to make a good vegetable soup for you. After they finished cooking, he will present that soup in a neat way in your table. Now this chef is activation function in this whole cooking process. You don’t need to care about the inside process as long as the activation function present in your deep learning model, he will present the output in a way how you want.

If you are working in a binary classification problem, you just want to get the final output either in ‘Yes’ or ‘No’. It will do some calculation inside based on the values of the input features and give the final result of the calculation to the activation function. It may be a continuous numbers, we need not to worry about that, it will take that continuous numbers and map the values to 0s and 1s and give you a outcome in binary form. Because that’s what we want.

There are lots of Activation function and I will write about that in a separate article.

In this article, I have given an introduction to some of the basic concepts in neural network and try to write how to implement all these things in python in my subsequent articles as soon as possible.