# What is Perceiver IO? Definition and meaning

A DeepMind architecture that cross-attends from a large input to a small learned latent array and decodes arbitrary outputs.

## Definition

**Perceiver IO**: A DeepMind architecture that cross-attends from a large input to a small learned latent array and decodes arbitrary outputs.

In short: Attention through a small latent array.

## When it is used

A global reasoning path whose cost does not grow quadratically with input size.

## Learn more

The full guide to **[Perceiver IO](/tools/perceiver-io)** explains how it works, when to use it and how it compares with the alternatives, with examples. The [glossary](/glossary) lists every term on the site from A to Z.

Previous term: [PEFT](/glossary/peft) · Next term: [pgvector](/glossary/pgvector)
