Towards an Architecture-Agnostic Definition of "Attention"

Researchers propose a definition of 'attention' in AI that is independent of specific architectures, aiming to unify understanding across models. This approach seeks to clarify how attention mechanisms function regardless of implementation details.