Swapout: Learning an ensemble of deep architectures
This addresses the challenge of efficient and regularized training for deep learning practitioners, offering a novel ensemble approach that is incremental over existing methods like dropout and stochastic depth.
The paper tackles the problem of improving deep neural network training by introducing Swapout, a stochastic method that outperforms ResNets with identical structure, achieving state-of-the-art accuracy on CIFAR-10 and CIFAR-100, with a 32-layer model matching the performance of a 1001-layer ResNet.
We describe Swapout, a new stochastic training method, that outperforms ResNets of identical network structure yielding impressive results on CIFAR-10 and CIFAR-100. Swapout samples from a rich set of architectures including dropout, stochastic depth and residual architectures as special cases. When viewed as a regularization method swapout not only inhibits co-adaptation of units in a layer, similar to dropout, but also across network layers. We conjecture that swapout achieves strong regularization by implicitly tying the parameters across layers. When viewed as an ensemble training method, it samples a much richer set of architectures than existing methods such as dropout or stochastic depth. We propose a parameterization that reveals connections to exiting architectures and suggests a much richer set of architectures to be explored. We show that our formulation suggests an efficient training method and validate our conclusions on CIFAR-10 and CIFAR-100 matching state of the art accuracy. Remarkably, our 32 layer wider model performs similar to a 1001 layer ResNet model.