E^2CM: Early Exit via Class Means for Efficient Supervised and Unsupervised Learning

Code

• github.com/alperengormez/class_means
  Officialpytorch★ 1

Abstract

State-of-the-art neural networks with early exit mechanisms often need considerable amount of training and fine tuning to achieve good performance with low computational cost. We propose a novel early exit technique, Early Exit Class Means (E^2CM), based on class means of samples. Unlike most existing schemes, E^2CM does not require gradient-based training of internal classifiers and it does not modify the base network by any means. This makes it particularly useful for neural network training in low-power devices, as in wireless edge networks. We evaluate the performance and overheads of E^2CM over various base neural networks such as MobileNetV3, EfficientNet, ResNet, and datasets such as CIFAR-100, ImageNet, and KMNIST. Our results show that, given a fixed training time budget, E^2CM achieves higher accuracy as compared to existing early exit mechanisms. Moreover, if there are no limitations on the training time budget, E^2CM can be combined with an existing early exit scheme to boost the latter's performance, achieving a better trade-off between computational cost and network accuracy. We also show that E^2CM can be used to decrease the computational cost in unsupervised learning tasks.

Reproductions