Decoder. plain-English AI glossary

Continual Learning

▲ Rising

Also called Lifelong Learning

Teaching a model new things over time without making it forget what it already knew.

Think of it like

Adding new recipes to a chef’s repertoire without them suddenly forgetting how to make the old ones.

Example

A support model needs this quarter’s product updates added on top, but must keep answering last year’s questions correctly.

How it actually works

The core tension is stability versus plasticity: learn new information fast enough to stay current, but not so aggressively that you overwrite old knowledge — the dreaded catastrophic forgetting. Techniques include replaying old data, freezing parts of the network, and adding small adapters. Truly seamless continual learning remains an open problem.

For product teams

The dream of updating a model with fresh knowledge without a full, expensive retrain — with real trade-offs today.

For engineers

Sequential training on new data while preserving prior performance, balancing plasticity against forgetting.

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