Module 01 - Live
See how neural networks actually think.
NeuronCanvas is a free (for noncommercial use), source-available, interactive explanation of deep learning, built for people who have never heard the words "gradient descent" before, as much as for people who already ship ML in production. Drag a weight, watch backpropagation happen step by step, and train a real model entirely in your browser.
- No account
- Nothing to sign up for. No email, no password, no paywall, no "free trial."
- No install
- It's a browser tab. Every playground on this site runs on a phone.
- No prerequisites
- No calculus and no Python. Module one starts from what a number is doing when you multiply it.
- Nothing leaves your device
- Models here train in your own browser. No analytics, no trackers, no third-party cookies.
01 / 07
Playground
Everything below trains in this tab. Nothing is sent anywhere.
Try it right now
This is a real, running network, not a screenshot, not a recording. Pick a dataset, press play, and watch it learn.
02 / 07
Anatomy
What's inside a module
Every concept page on this site follows the same four beats, in the same order. The Neural Networks module runs seventeen of them, from “what is a neuron” through to a full trainable playground, so you always know the shape of what you're about to read.
- 01
A clip, before any notation.
The page opens with a short, silent animation of the thing itself: data moving through a layer, a curve bending, a gradient rolling downhill. There's nothing to read yet, and nothing to understand, just something to watch. - 02
A canvas you drive.
Then the idea becomes a control in your hand. Drag the weight, switch the activation, break the initialization on purpose. The diagram responds within a frame, so cause and effect stay attached to each other. - 03
The math, next to the thing it describes.
Only once the behaviour is familiar does the formula appear, placed beside the exact diagram it's describing, so every symbol in it has something on screen to point at. - 04
A challenge you can't read your way through.
Each page closes with a small task: get the spiral to separate, make this network fail in a specific way. If you skimmed, it won't work, and that's the point.
03 / 07
Method
How NeuronCanvas teaches
Same three-step rhythm in every module, whether you know what a gradient is or you’ve never heard the word "neural network" before today.
- 01
Watch it happen
Every concept starts as a short animation - data flowing through a network, a weight tilting a decision, a loss curve falling. You see the shape of the idea before any notation. - 02
Touch it yourself
Then you take over. Drag a weight, click a neuron, edit a kernel - every diagram on this site responds within a frame. Intuition comes from your own hands, not from re-reading a paragraph. - 03
Train a real model
Every module ends in a full playground: pick a dataset, shape the network, press play, and watch an actual model, running entirely in your browser, learn in real time.
04 / 07
Audience
Who this is for
Built so that a total beginner and a working engineer can open the same page and both get something out of it.
- A
Complete beginners
Never written a line of code or seen the word "gradient" before? Every module starts from a plain-language analogy, no prior math or programming assumed, before it ever shows an equation. - B
Students and self-learners
Studying deep learning from a course or textbook and want the concept to finally click? Use NeuronCanvas alongside your material - drag the exact weight your lecture is describing. - C
Teachers
Every playground can be projected, screen-shared, or linked with an exact configuration baked into the URL, so a class can look at the same trained (or deliberately broken) network together. - D
Engineers going deeper
Already ship ML in production but want sharper intuition for why a network behaves the way it does? The playgrounds make failure modes, like a stuck spiral or a dead ReLU, visible in seconds.
05 / 07
Curriculum
The four unbuilt modules all list Neural Networks as their prerequisite.
One architecture at a time
Each module is a scroll-driven visual story followed by a full playground you can break, fix, and share. Neural Networks is live today; everything else is being built in the open, in order, one module at a time.
- 01
Neural Networks
The building block everything else is made of.Live - 02
Convolutional Networks
How machines learn to see.Not built - 03
Recurrent Networks
Giving a network memory.Not built - 04
GANs
Two networks, playing a game against each other.Not built - 05
Transformers
The architecture behind modern AI.Not built
06 / 07
Open
- 17
- Concept pages, module 01
- 18
- Hand-built interactive demos
- 0
- Analytics scripts and trackers
Built in the open, one module at a time
NeuronCanvas is one person's project, and it doesn't pretend otherwise. The source is public and readable - every explanation, every animation, and every line of the training engine that draws the pictures. That's deliberate. A teaching site you can't inspect is asking for trust it hasn't earned, and the fastest way to check whether an explanation here is honest is to go and read the code that produced the diagram.
It's free for noncommercial use under the PolyForm Noncommercial License, with a separate paid licence for commercial use. There's no funding behind it, no team, and no launch date to hit, which is exactly why modules ship one at a time and finished, rather than five at once and half-built. Neural Networks is finished; Convolutional Networks is next.
07 / 07
FAQ
Frequently asked questions
01Do I need to know math to start?
No. Every module opens with a plain-language explanation and an animation before any formula appears. The math is there when you're ready for it - each equation is shown next to the exact interactive diagram it describes, so you can see what each symbol means by dragging it.
02Do I need to know how to code?
No. NeuronCanvas is something you read and play with in a browser tab, not something you set up. Nothing to install, no account, no code editor.
03Is my browser powerful enough to train a neural network?
Yes, the networks in these playgrounds are small (a handful of neurons over a few hundred 2D points), and the training engine is a lightweight hand-written implementation built for exactly this, so it runs smoothly on a phone.
04Is it free?
Yes, for noncommercial use - personal learning, teaching, research, and hobby projects are all free under the PolyForm Noncommercial License. The source is publicly readable and forkable too. Commercial use requires a separate paid commercial license.
05How is this different from TensorFlow Playground?
TensorFlow Playground covers one architecture, a small feedforward network, extremely well, and NeuronCanvas's Neural Networks module is built in that same spirit. The goal here is to extend that same drag-it-and-see-it approach to the architectures Playground doesn't cover: CNNs, RNNs, GANs, and transformers.
06What's a neural network, in one sentence?
A neural network is a small number of simple decisions (each one just a weighted sum passed through a squashing function) chained together and adjusted automatically until, collectively, they get better at a task - that whole idea is what the Neural Networks module walks through from scratch.
Start with a single neuron.
Module one assumes nothing - not that you've seen a matrix, not that you've written code, not that you know what the word “neuron” is doing in a computing context. It takes about forty minutes, and everything else on this site is built on it.