Rapidly updated weight parameters that change during inference to capture task-specific or context-specific information.