← Library · Advanced concept

Compositional Generalization

Compositional generalization refers to an AI model's ability to understand novel combinations of familiar components. For example, if a model understands 'red car' and 'blue bike', it should understand 'red bike' without explicit training on that exact combination. This is a crucial aspect of human intelligence that many current AI models struggle with.

In plain terms

It's like knowing individual words and rules of grammar, then being able to construct and understand new, never-before-heard sentences.

Why it matters

Achieving compositional generalization is key for AI to perform robustly in unseen scenarios and to move beyond literal pattern matching towards true understanding.

Learn one new AI thing every day.

Daily Deck sends you seven plain-English cards like this every morning. Free.

Start free