Vote-based strategy for early exit is another example of ensemble-based early exit. This type of model will exit when there has accumulated more votes for exit than the predefined threshold.

New York University

One draw back of confidence-based early exit is wasted computation when the confidence indicator does not reach the threshold. On the contrary, ensemble-based early exit model makes use of predictions of multiple internal classifier and makes better prediction. 

Ensemble-Based Early Exit

Xu, C., & McAuley, J. (2022). A Survey on Dynamic Neural Networks for Natural Language Processing. ArXiv, abs/2202.07101.

A Survey on Dynamic Neural Networks for Natural Language Processing

Patience-based strategy for early exit is an example of ensemble-based early exit. This type of model will exit when a series of consecutive internal classifiers have suggested for exit.

Learn Before

Related