What is a Neural Network?

Neural networks are loosely modelled on how the neurons in a human brain behave. Their great advantage is that they extract features by themselves, without the programmer pointing anything out.

1What it is

Neural network

A neural network is essentially a system of organising machine learning algorithms to perform a task. It is a fast and efficient way to solve problems where the dataset is very large — images, for instance.

It is divided into layers, and each layer is divided into blocks called nodes. Every node has its own job to do, and passes its result on to the next layer.

2The three kinds of layer

Four layers, and where the work happens

Watch which nodes flare as the signal passes. Drag sideways to turn.

Data goes in at the left and the answer comes out at the right. Only the hidden nodes flare — because they are the only ones doing any computing.

The input layer only collects the data and passes it on. The hidden layers do all the processing, using weights and biases. The output layer just hands you the answer.

Input layer

The first layer. Its job is to acquire the data and feed it to the network. No processing occurs at the input layer.

Hidden layers

All of the processing happens here. They are called hidden because they are not visible to the user. Each node of a hidden layer runs its own machine learning algorithm on the data it received from the layer before.
The computation is done by means of weights and biases, and the value found is passed through a chosen activation function before it moves on.

Output layer

The last layer. The final hidden layer hands it the processed data and it gives that to the user as the final output. Like the input layer, it does not process the data — it is there for the user interface.

Three jobs, in order

A useful way to hold the layers in your head is to give each one a job: the input layer are the observers, the hidden layers are the thinkers, and the output layer are the reporters.
Be careful with that last one. It is tempting to call the output layer the decider, but the decision was worked out in the hidden layers — the output layer only announces it. The syllabus is explicit that no processing happens there, and a one-mark question can turn on exactly that.

Sorting photos of cats and dogs

Input layer: each pixel of the photo is fed into the network.
Hidden layers: the network looks for patterns — whiskers, floppy ears — passing the information through several steps.
Output layer: the answer comes out. Cat, or dog.
The more photos it practises on, the better it gets — just as you get better at a game the more you play it.

Two things the paper has already tested

The first layer is the input layer, never the output layer. And there is no fixed number of layers: there can be many hidden layers, and how many depends on how complicated the job is. The number of nodes in each layer can vary too.

3How it learns

Finding the right output begins with trial and error until the network finally learns. With each try the weights are adjusted, based on the error between the output that was wanted and the output the network gave.

A mixing desk with a great many sliders

Every connection has a slider on it — that is its weight. The network plays the track, hears that it is wrong, nudges the sliders a little, and plays it again. It does this thousands of times. Nobody ever decides where a slider should sit; the error decides.

Layer	How many	What it does
Input	One	Acquires the data and feeds it in. No processing.
Hidden	One or many — it depends on the complexity	All the processing, using weights and biases
Output	One	Hands the final answer to the user. No processing.

Where you have met one

Real-world neural networks: face recognition, customer support chatbots, vegetable price prediction.

4From the board papers

Statement 1: There are four layers in a neural network.
Statement 2: The first layer of the neural network is known as the output layer.CBSE sample paper 2024–25

1 mark

▶Show the answer

Both statements are incorrect. There is no fixed number of layers — there can be many hidden layers, depending on the complexity of the problem. And the first layer is the input layer; the output layer is the last one.

Assertion (A): Information processing in a neural network relies on weights and biases assigned to nodes.
Reason (R): These weights and biases determine how strongly a node is influenced by its inputs and its overall contribution to the next layer.Class X handbook

1 mark

▶Show the answer

Both (A) and (R) are true, and (R) is the correct explanation of (A). Saying the processing “relies on weights and biases” and saying they “determine how strongly a node is influenced by its inputs” are the same fact, stated twice.

What is a neural network? Give the functioning of its three layers.Class X handbook, Reflection Time

2 marks

▶Show the answer

A neural network is a system of organising machine learning algorithms to perform a task, loosely modelled on how neurons in the human brain behave. It is divided into layers of nodes, and can extract features automatically without the programmer’s input.
Input layer — acquires the data and feeds it to the network; no processing happens here.
Hidden layer — where all the processing happens, by means of weights and biases; each node runs its own machine learning algorithm.
Output layer — receives the processed data from the last hidden layer and gives it to the user; it does not process the data.

5Check yourself

Quick Check

Which layer of a neural network does no processing at all?

Quick Check

Where does all the processing in a neural network happen?

Quick Check

The process of adjusting the weights in a neural network to improve performance is called:

Quick Check

Neural networks are particularly well suited to:

← Previous

🧠 Deep Learning: ANN and CNN

How AI Makes a Decision 🎚️