Foundations · MLD-0691

Machine learning

Methods that use data or experience to fit models rather than specifying every decision rule by hand.

Expanded guide · editorial review 2026-11-01. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.

Prerequisites

How it works

Machine learning chooses a model from data using an objective and a training procedure. The objective measures success on examples; the experiment must separately measure success on unseen cases. Deep learning is one family within machine learning. A language model is a particular kind of model, rather than a synonym for the entire field.

Worked example

For a campus helpdesk, use past tickets to predict a routing category. Compare a majority-class baseline, a linear classifier and a neural model. Split by time before fitting text transformations so later tickets cannot influence training. This is an illustrative experiment, not a measured result.

A useful objective

Choose parameters theta to minimize average training loss: J(theta) = (1/n) sum_i loss(f_theta(x_i), y_i). Minimizing J alone does not establish generalization.

What to test

Report held-out performance, class-specific errors and variation across splits. Test a new department separately to reveal distribution shift. A more complex model is useful only when its improvement justifies its operational cost.