What a neural network actually is
Take this if the phrase neural network still sounds like a brain to you, and you would rather it sounded like arithmetic.
What you’ll be able to do
You can draw a single neuron, say what its weights and bias do to the decision boundary, and explain why stacking two of them separates data one cannot.
What it assumes
You can plot a point on a graph and multiply two numbers. No coding and no linear algebra needed.
- Price
- Free
- Length
- 1 week
- Sessions
- 4
- Effort
- ~2 h/week · 2 h total
- For
- Anyone
Adds it to My courses on your dashboard. No card, nothing to cancel.
What’s taught, in order
- 01
One neuron, drawn by hand
A neuron is a weighted sum and a threshold. You set the weights yourself, plot what the neuron then believes, and discover you have drawn a straight line without being told you were going to.
- 02
What the weights and the bias each control
Changing one weight tilts the boundary; changing the bias slides it. Separating these two effects by hand is what stops weights from being a black box later.
- 03
The problem one neuron cannot solve
You are given two clusters that no single straight line can separate, and asked to try anyway. Failing at this on purpose is the reason the next module exists.
- 04
Two neurons, and the word hidden
Adding a second neuron and combining the two boundaries solves the problem you just failed. That combination is the whole idea of a hidden layer, met before the terminology.
What you build
Build a two-neuron network in the block editor that correctly separates the two interleaved clusters you were given in session three. Record the weights and biases you ended on, and a screenshot of the boundary it draws.
Check it yourself against this
- Your network classifies every point in the given set correctly.
- You can say which neuron is responsible for which part of the boundary.
- You changed one weight deliberately and correctly predicted which way the line would tilt.
- You can explain why one neuron alone could not do this, without using the word nonlinear.
Then answer this
Your two-neuron network draws a boundary neither neuron could draw alone — so where exactly does the extra capability come from?
Two or three sentences, in your own words. If you cannot, the course is not finished — go back to the module it came from.
What this course does not do
Nothing here is learned automatically. You set every weight by hand, which is precisely why it makes sense; the algorithm that finds those weights for you is backpropagation, and it is not in this course. Real networks also have activation functions, many layers and thousands of parameters, none of which appear here.
Where this leads
The single-neuron demo used throughout this course is free to play with on its own page, and it is worth ten minutes before you commit to anything — if setting weights by hand appeals to you, the rest of the curriculum works exactly like this.
FAQ
- Will I write any code in this course?
- Only in a block editor, where you drag pieces together rather than type syntax. That is a deliberate choice for a first course: typing errors are a distraction when the thing you are trying to understand is what a weight does to a boundary.
- Is a neuron here anything like a neuron in a brain?
- Barely, and the course says so early. The name is a historical accident from the 1940s. What you build is a weighted sum with a threshold, which is a piece of arithmetic — treating it as brain-like is the single most common reason people find deep learning mystifying.
- What should I take after this?
- If setting weights by hand was satisfying, the natural next step is the sprint on how a model finds those weights itself, or the Deep Learning phase course if you want the whole path from one neuron to a transformer.
Other courses
This is a standalone course. It does not add weeks to the 78-week curriculum or change your roadmap. If you want the whole path instead, the full programme is ₹399/month · ₹2,000/year — see what it covers.