The big idea: An artificial neural network is layers of simple units. Each unit takes its inputs, multiplies each by a weight, adds them up, adds a bias, and passes the result through an activation function.
Nothing in one unit is complicated. The capability comes from having many, in layers.
Step through: one unit's arithmetic, the layers, what a hidden layer adds, and how training adjusts the weights.
Interactive diagram
Explore the labelled diagram, charts and maps for this topic in full study mode.
Free preview
This is the free notes preview
You're reading the free notes. Aimnova Pro unlocks the full study experience — and you can try it with your first topic free to keep:
- FlashcardsLock in vocabulary and key terms with spaced repetition.
- Practice questionsAnswer exam-style questions and get instant AI marking.
- Mock exams & past-paper vaultSit full mocks and see exactly how examiners award marks.
- Personalised study planA daily plan built around your exam date and weak areas.
Input layer
- One unit per feature
- Does no computation — it holds the values
- 28×28 pixels means 784 input units
Hidden layers
- Where the work happens
- Each layer builds on the patterns the previous one found
- "Deep learning" means several of them
Output layer
- One unit per class, or one for a regression
- Its activation turns the numbers into an answer
- Ten units for the digits 0–9
Why the activation function is not optional: Without it, every layer is a weighted sum of weighted sums — which collapses to one weighted sum. A hundred layers would have exactly the power of one.
The non-linear activation is what lets extra layers add anything at all.
Know your predicted grade
Take timed mock exams and get detailed feedback on every answer. See exactly where you're losing marks.
Forward pass
Loss
Backpropagation
Update
Training adjusts the weights, and nothing else: The structure — how many layers, how many units — is fixed before training and never changes. Those are hyperparameters.
What training produces is the set of weights and biases. That is the whole learned model.
It cannot explain itself: A trained network is millions of numbers. There is no rule to read out, no path to trace, no reason it can give.
That is why an accurate network is often rejected for lending or medicine, where a decision must be justified to the person it was made about.
How this is tested — you must describe the structure layer by layer and explain what training actually changes. It comes up two ways:
Paper 1 Section A
- Outline the structure and function of an ANN, 3-5 marks
- Name the three kinds of layer
- State what a weight does
Paper 1 Section B — case study
- Explain why multi-layer networks are more capable
- Judge an ANN against an interpretable model
The classic trap: Saying a network "works like the brain". It is an analogy in name only — a unit is a weighted sum and an activation. Describe the arithmetic, not the metaphor.
Outline the structure and function of an artificial neural network used to recognise handwritten digits, and explain why a hidden layer is necessary.
Model answer plan
See the mark-by-mark plan — for / against / judgement, with marking guidance — in study mode.