Decoder. plain-English AI glossary

Dropout

● Core

Randomly switching off part of the network during training so it can’t over-rely on any one path.

Think of it like

Making a team practice with random players benched, so no single star becomes a crutch.

Example

A model prone to memorizing its training data might use dropout to force more robust, spread-out learning.

How it actually works

During training, dropout zeroes a random fraction of activations each step, so neurons must learn redundant, general features instead of brittle co-dependencies. It’s a cheap, effective regularizer against overfitting. At inference it’s turned off and the full network runs.

For product teams

A standard anti-overfitting knob — it trades a little training speed for better generalization.

For engineers

Stochastically masks activations with probability p at train time; a regularizer disabled at inference.

Related

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