Machine-learning papers lean on a shared shorthand. The same Greek letter means a learning rate in one section and a noise schedule in the next, and most of the equations that matter are quoted without derivation. The atlas above is a reference for reading them.
How to use it
- Start from a symbol. Pick any symbol on the front page (𝔼, ∇, θ, β and so on) to see what it usually means and every equation in the atlas that uses it.
- Or start from a topic. Probability, linear algebra, optimisation, information theory, graphical models, HMMs, autoencoders, transformers, diffusion and distillation each have their own page.
- Open an equation for a plain-language reading, a legend of its terms and the paper or textbook it comes from.
- Search by symbol, name or idea — try
KL,Jacobianorattention.
It works best with a bit of room, so there is a full-screen link above the frame.