To Ensemble or Not Ensemble: When does End-To-End Training Fail?
This addresses a key problem for machine learning practitioners by identifying limitations in E2E training for ensembles, which is incremental but provides specific insights into training strategies.
The paper investigates failure cases of end-to-end (E2E) training for ensembles of deep networks, finding that over-parameterized models cannot be trained E2E and that optimal performance sometimes lies between independent training and full E2E.
End-to-End training (E2E) is becoming more and more popular to train complex Deep Network architectures. An interesting question is whether this trend will continue-are there any clear failure cases for E2E training? We study this question in depth, for the specific case of E2E training an ensemble of networks. Our strategy is to blend the gradient smoothly in between two extremes: from independent training of the networks, up to to full E2E training. We find clear failure cases, where over-parameterized models cannot be trained E2E. A surprising result is that the optimum can sometimes lie in between the two, neither an ensemble or an E2E system. The work also uncovers links to Dropout, and raises questions around the nature of ensemble diversity and multi-branch networks.