Inductive Bias

What is Inductive Bias?

The set of assumptions built into a learning algorithm to help it generalize. CNNs have a 'strong' inductive bias for locality (pixels near each other matter); Transformers have a 'weak' bias, making them more data-hungry but flexible.

Where did the term "Inductive Bias" come from?

Fundamental concept in learning theory.

How is "Inductive Bias" used today?

Why CNNs beat Transformers on small datasets.

Related Terms