BASICS OF NEURAL NETWORKS

Hey there! This is your techie daydreamer who is going to explain the absolute basics of neural network.

When I first wanted to study Neural Network, I scavenged a lot of resources and the internet but it was only after some insights from my senior that I was able to finally understand this simple yet beautiful concept.

So let's begin.

The neural network is a concept (should say an integral concept) of deep learning models which is inspired by the human brain. So things are going to get lot easier if we first see how the human brain works.

Human brain consists of billions of neurons. Each neuron is connected to many other neurons, creating a complex network and this complexity of the network is the reason why our brain is so much superior to other species.

See, neurons have a simple function. They either fire or they don't. On or off. Zeros and ones. If we analyse these tiny switches, we notice that the signals are carried forward by undergoing some mathematical operations.

Let me break it down for you.

When a neuron fires a signal, it passes down a part of the signal to one of its many neighbouring neurons. It is passed through a component in neurons called dendrites. But we must understand that the other neuron is not only receiving signals from one neuron rather it is receiving signals from many other neurons. So, the neuron sums all these incoming signals in the cell body. Now, we have a consolidated signal value. This signal is then sent further to its neighbouring neuron. Here, it is important to understand that the component which sends the signals to other neurons (called Axon) allows the signal to go to certain neighbouring neurons only. Hence, the signal is not divided amongst all the neighbouring neurons rather it is only distributed to certain neurons.

Now, you might be thinking “That is some cool bio stuff, but where do neural networks sit amongst all this?”

For a long time, computer scientist were laying the foundation for machine learning but they were only able to make basic predictions and the accuracy fell drastically when they increased the complexity of the prediction. That is when in 1950s, they had a wild intuition. Why not recreate the structure of brain using lines of code? So they built an artificial neuron (they called it Perceptron) and used basic algebra to recreate what a neuron does.

The Dendrites (Inputs): Instead of biological signals, our digital neuron takes in data. If we are building an AI to predict house prices, the inputs might be the number of bedrooms, the square footage, etc.

The Cell Body (The Summing Machine): Just like your brain sums up all the incoming signals, the computer multiplies each piece of data by its importance and adds them all together.

The Axon (The On/Off Switch): This is where the neuron decides whether to pass the message forward. In AI, we call this the Activation Function. It’s just a mathematical rule that says: "If the total sum is big enough, fire a 1. If it’s too small, stay at 0."

But there is one massive secret ingredient that makes artificial neurons actually smart. It’s how the computer handles that "selective distribution" I mentioned earlier.

In code, we call these Weights.

Think of Weights as the neuron's personal opinion. Not all inputs are created equal. If you are buying a house, the number of bedrooms matters a lot to you (High Weight), but the colour of the front door probably doesn't matter at all (Low Weight). The network constantly tweaks these weights until it gets its predictions right.

When you chain thousands of these digital, opinionated switches together in layers, you get a Neural Network.

This was the beginning of deep learning in machine learning. What started as a simple biological inspiration has become the engine behind almost every major AI breakthrough today like ChatGPT.

The most beautiful part? At its core, a neural network is still just doing what our brain does naturally- learning from experience by adjusting the strength of connections.