Curriculum learning
Training a model on easy examples first, then gradually harder ones — like a school syllabus.
Think of it like
Like teaching arithmetic before calculus: the early, simple lessons build the footing the hard ones stand on.
Example
A code model trains first on short, simple snippets, then longer and trickier programs as it improves.
How it actually works
Curriculum learning orders training data from simple to complex so early, easy signal builds representations that make later, hard examples learnable. It can speed convergence and improve final quality, echoing how people learn. The tricky parts are defining “difficulty” and pacing the ramp — a badly designed curriculum can help little or even hurt versus plain random ordering.
For product teams
Sometimes teaching order matters — structuring the data can beat throwing everything in at once.
For engineers
Ordering training examples easy→hard to ease optimization; benefit depends on a good difficulty measure and pacing.
Related
- Training — A strategy within training.
- Learning signal — Shapes the learning signal.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome