Is it expedient to use an ensemble of convolutional neural networks for pattern recognition tasks?
According to the article of Liran Chen (The University of Chicago) and this corresponding presentation, using ensemble of convolutional neural networks lead to dramatic error reduction:
It is shown that the predictors gain more and more accuracy as the number of models involve in the ensemble increases:
It is shown that ensemble learning may lead to accuracy gains along
with reduction in training time.
I could not find the publication date of the artice as well as the academic degree of the author. So, perhaps, this method of pattern recognition is outdated now, or have shown to be ineffective in certain areas.
For the moment, are there some worthwhile advantages of using an ensemble of convolutional neural networks over using a single network in pattern recognition tasks?
Can you help to clarify the matters?
Best Answer
Empirically, ensembles of convolutional neural networks (CNNs) often yield a few percentage points compared to a single CNN.
Example image classification for: Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." In Advances in neural information processing systems, pp. 1097-1105. 2012. http://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf
Example for machine translation (LSTM): Sutskever, Ilya, Oriol Vinyals, and Quoc V. Le. "Sequence to sequence learning with neural networks." In Advances in neural information processing systems, pp. 3104-3112. 2014. https://arxiv.org/pdf/1409.3215.pdf