Notation is a language

Lesson 1 of 3 · 2 min

Maths in papers looks hard mostly because the symbols are new, not because the ideas are. Treat notation like words in a new language: build a small dictionary as you go.

Build a symbol table

Keep a short table while you read. Every time a new symbol appears, write what it means and where it was defined.

Symbol	Meaning	Defined in
x	one input example	Section 2
y	the correct label for x	Section 2
θ (theta)	the model's parameters	Eq. 1
L	the loss: how wrong the model is	Eq. 3

x: Meaning: one input example; Defined in: Section 2

y: Meaning: the correct label for x; Defined in: Section 2

θ (theta): Meaning: the model's parameters; Defined in: Eq. 1

L: Meaning: the loss: how wrong the model is; Defined in: Eq. 3

Common conventions in ML papers

Lower-case bold (x) is usually a vector; upper-case bold (W) a matrix.
A hat (ŷ) usually means a prediction or estimate.
Subscripts pick one item: x_i is the i-th example.
Σ means "add up"; E[...] means "the average over many cases".
argmin means "the value that makes this smallest".

Many books list these at the start. Goodfellow, Bengio and Courville's Deep Learning has a clear notation page that many ML papers follow.

When notation is inconsistent

Papers sometimes use one letter for two things, or define a symbol late. This is the authors' fault, not yours. Note it in your table. If notation changes between sections, it can be a sign the paper was put together in a hurry.

RememberKeep a symbol table for every hard paper. Most "I can't follow the maths" moments are really "I forgot what this letter means".