Why should we add early exits to neural networks?

Open in new window