LaTeX for machine learning papers without the notation mess

Most ML papers fail the same way in LaTeX: the math is fine on page three and inconsistent by page twelve. A loss is L in one section, \mathcal{L} in another, and bold vectors appear only when the author remembered. Reviewers notice.

Pick a notation sheet on day one

Before you write related work, write twenty lines of macros:

\newcommand{\vect}[1]{\boldsymbol{#1}}
\newcommand{\mat}[1]{\mathbf{#1}}
\DeclareMathOperator*{\argmax}{arg\,max}
\newcommand{\loss}{\mathcal{L}}
\newcommand{\expect}{\mathbb{E}}

Paste that block into every coauthor branch. When a model invents a new symbol, add it to the sheet first, then use it. The AI follow-notation lesson is the same idea with a language model in the loop.

Tables that survive camera-ready

Ablation tables grow until they fall off the page. Prefer booktabs with no vertical rules, siunitx for aligned decimals, and a caption that states the metric and the split. If the table spans pages, use longtable early instead of shrinking the font to 7pt.

A visual builder helps for the first draft: the table generator runs in the browser.

Algorithms and listings

Pseudocode belongs in algorithm2e or algorithmicx, not in a monospace paragraph. Real training scripts belong in listings or minted with a language set so reviewers can skim. Keep the main paper short; put full configs in the appendix or a repo link.

Figures from TensorBoard and matplotlib

Export PDF or SVG when you can. Raster PNGs at 300dpi are fine for photos of interfaces, not for line plots. Crop whitespace before \includegraphics. If a float flees to the next section, fix size and placement before you fight [H] for an hour: see float placement.

Git next to the paper

Every architecture change should be a commit you can name. Local-first tools like Oleafly treat the project as a folder with real Git history, which matches how ML code already lives. Pair that with the cheatsheet when you forget syntax under a deadline.