Skip to content
BENIBIX Learn

Semi-Supervised Learning

Also known as Semi-Supervised Machine Learning

The model learns from a few rows with labels and lots of rows with none.

Think of it like

Think of it like a teacher who marks only a few of your sums. You work out the rest from those.

Example

A clinic has 200 scans marked by doctors and 20,000 scans nobody has marked. It trains on both, and the model beats one trained on the 200 alone.

Why it matters

Labels are often the most costly part of a project. This gets more from each one.

Definition

Semi-supervised learning uses a few labelled examples and lots of unlabelled ones together. The labelled ones show what the answers look like. The rest show the shape of the data. Together they often beat the few labels used alone. It helps when labels cost a lot and raw data is cheap.
Semi-supervised learning sits between the first two panels. A few dots have colours. Most are grey.

Common mistake

Hoping the extra data will fix bad labels. If the few labels are wrong, the extra data spreads the mistake.

Key takeaways

  1. 1
    A few labels. Show what answers look like.
  2. 2
    Lots of raw data. Shows the shape of the rest.
  3. 3
    Saves labelling. More from each costly label.

Learn it properly