Principal Component Analysis
Finding the directions along which data varies most, so it can be described with fewer numbers.
Centre the columns, then find the direction of greatest spread — that's the first principal component. The second is the direction of greatest remaining spread at a right angle to it, and so on. Formally they're the eigenvectors of the covariance matrix, sorted by eigenvalue, and each eigenvalue's share of the total is that component's explained variance.
Neither input requirement is optional. Centring is required because variance is measured about the mean, and scaling is required because variance carries units — a column in centimetres outvotes the same column in metres by a factor of 10,000.
Two things distinguish it from t-SNE and UMAP: it gives you a transform you can save and apply to a new row, and its distances and directions are meaningful. It's also fitted, so fit it on training rows only.