Vanishing gradient problem

The vanishing gradient problem occurs when gradients become extremely small as they are propagated back through many layers, causing early layers of a deep network to learn very slowly or not at all.

Vanishing gradient problem sits in the Foundations part of the Agentik {OS} glossary, which defines the words used to build and run AI agent systems.

Also called Vanishing gradients, Exploding gradients.