NeuronCanvas

Curriculum

Learn Deep Learning

A growing curriculum, taught in order, each module building on the last. Every module follows the same shape: a short visual story with no assumed background, followed by a full playground you can train, break, and share. Five architectures are planned to start - more will be added over time.

New to deep learning entirely? Start with Neural Networks - it assumes nothing, not even that you know what a neuron is, and everything after it builds on what you learn there.

Figure 01 - Dependency spineOne root, four planned

01 / 04

Paths

Three ways through this

  1. 01

    Starting from nothing.

    Read module one from the top, in order, and don't skip the animations - they're doing the work the text isn't. It assumes no maths and no code, and every later module is written on the assumption you've been through it. About five hours, best taken over a few sittings rather than one.
  2. 02

    Filling one specific gap.

    Already in a course and stuck on a single idea - backpropagation, softmax, why initialization scale matters at all? Go straight to that page from the contents below. Each one re-establishes what it needs rather than assuming the previous sixteen.
  3. 03

    Teaching with it.

    Every playground encodes its full configuration in the URL. Set up a network that's about to fail, copy the link, and hand a class the exact same broken model to fix - no accounts, no setup, nothing to install on lab machines.

02 / 04

Available now

The module you can read today

  1. Module 01 - Available30–45 min

    Neural Networks

    The building block everything else is made of.

    Start from a single neuron and build up to a full trainable network - weights, bias, activation functions, loss, and backpropagation, each explained visually before it's explained mathematically.

    • Explain what a weight and a bias actually do to a number
    • Read a network diagram and know what the lines and colors mean
    • Describe why a network needs to be "trained" at all
    • Train a small classifier on four different toy datasets

03 / 04

Contents

Parts
6
Pages
17

Inside Neural Networks

Seventeen pages in six parts, meant to be read in order the first time through. If you're already in a course and just need one idea to click, every page also stands on its own - jump straight to it.

04 / 04

Roadmap

Why one module at a time

Four of the five modules say “coming soon,” and that’s the honest state of it. This is a solo project built alongside learning the material properly, and a half-finished module that teaches something subtly wrong is worse than a module that doesn’t exist yet. So each one ships complete - every concept page, every interactive, and the full playground - before the next one starts. Convolutional Networks is next.

  • 02

    Convolutional Networks

    Kernels sliding over an image, feature maps, pooling, and why that structure is what lets a network recognize a shape anywhere in a photo.Prerequisite: Neural Networks · 30–45 min
    Not built
  • 03

    Recurrent Networks

    Sequences, hidden state, and the vanishing gradient problem, plus why LSTMs were invented to fix it.Prerequisite: Neural Networks · 30–45 min
    Not built
  • 04

    GANs

    Generator vs. discriminator, and what it looks like when that adversarial game trains successfully, or collapses.Prerequisite: Neural Networks · 20–30 min
    Not built
  • 05

    Transformers

    Attention, tokens, and embeddings - the mechanism that lets a model decide which words in a sentence matter to each other.Prerequisite: Neural Networks · 40–60 min
    Not built

Start where everything else starts.

Neural Networks is the prerequisite for all four modules queued behind it, so it's the right first page whether you're here to learn from scratch or to check whether this site explains things the way you like. If you'd rather help build one of the modules that isn't written yet, that's open too.