Machine Learning for AI›Extra· 40 min read

Neural Networks Explained Simply

A neuron multiplies inputs by weights, adds them up, and fires — stack millions and you get deep learning.

What you will learn

Describe a single neuron’s maths
Compute one neuron by hand
See how layers build deep networks

Inspired by the brain (loosely)

A neural network is built from tiny units called neurons. The name comes from the brain, but a neuron here is just a small piece of arithmetic — nothing mystical.

What one neuron does

A neuron takes some inputs, multiplies each by a weight (its importance), adds them up (plus a bias), and passes the total through an activation that decides whether it “fires”.

A single neuron: weighted sum + bias, then an activation

# One neuron deciding "should I go out?" from 2 inputs
inputs  = [1, 0]        # 1 = sunny(yes), 0 = free time(no)
weights = [0.6, 0.9]    # how much each input matters
bias    = -0.5

total = sum(i * w for i, w in zip(inputs, weights)) + bias
output = 1 if total > 0 else 0     # a simple activation (fire or not)

print('total =', total, '-> output =', output)

Note: Output: total = 0.1 -> output = 1 0.6×1 + 0.9×0 = 0.6, minus 0.5 bias = 0.1. That is above 0, so the neuron fires (1 = “go out”). Change the weights and the decision changes.

From one neuron to deep learning

Put many neurons side by side → a layer.
Feed one layer’s outputs into the next → a network.
Many layers between input and output → a deep neural network (deep learning).
Training = automatically adjusting all the weights until the network’s predictions match the data.

With enough neurons and layers, a network can learn astonishingly complex patterns — recognising faces, understanding speech, generating text. It is the same neuron maths, repeated millions of times.

Tip: You will not tune weights by hand — libraries like TensorFlow and PyTorch do it. But knowing a neuron is “weighted sum + activation” takes the mystery out of every headline about deep learning.

Watch out: Deep networks are powerful but hungry: they need lots of data and computing power, and they are hard to explain (“why did it decide that?”). They are not always the right tool — a small problem often needs only a simple model.

Q. What does a single neuron compute?

Answer: A neuron multiplies each input by a weight, sums them with a bias, and applies an activation to produce its output.

✍️ Practice

Recompute the neuron with inputs = [1, 1] — does it still fire?
Explain what the bias does if you make it +0.5 instead of −0.5.

🏠 Homework

In your own words, explain how a network “learns” (hint: it adjusts the weights).