An interactive ML field manual
Learn it by touching it.
Five stations from classical machine learning to diffusion models — the same road I took, rebuilt as instruments you can operate. Every page has one signature interactive: drag the points, slide the kernels, scrub the noise. The math sticks when your hands are on it.
lab note · why this site exists
It started with one confusing afternoon: a pixel-art workflow said nearest-neighbor, an ML course said nearest neighbors, and they turned out to be completely different ideas wearing the same name. Untangling that collision is the site’s founding story — it’s told in full at the interlude, in its rightful place on the journey. No experience required anywhere here: jargon wears a dotted underline you can hover for a plain-words translation.
01Classical ML— The Boundary Lab
● liveBias, variance, and why k changes everything.
02Neural Networks— The Descent
● liveWatch optimizers race down a loss surface.
··Two Kinds of Nearest— Interlude
● liveNearest-neighbor scaling is not K-nearest-neighbors.
03CNNs & Vision— The Convolution Bench
● livePaint pixels, slide kernels, watch features emerge.
04Transformers— The Attention Lens
● liveEvery token looks at every other token.
05Diffusion— The Denoising Deck
● liveFrom pure noise to an image, one step at a time.
start anywhere — 5 of 5 instruments are fully powered