Adjusting for Dropout Variance in Batch Normalization and Weight Initialization
This solves a specific technical problem in deep learning for researchers and practitioners using dropout and Batch Normalization together, though it is an incremental improvement.
The paper addresses the variance mismatch between training and testing when using dropout with Batch Normalization, proposing corrections to weight initialization and Batch Normalization's variance estimates. This approach achieves state-of-the-art accuracy on CIFAR-10 and CIFAR-100 without data augmentation.
We show how to adjust for the variance introduced by dropout with corrections to weight initialization and Batch Normalization, yielding higher accuracy. Though dropout can preserve the expected input to a neuron between train and test, the variance of the input differs. We thus propose a new weight initialization by correcting for the influence of dropout rates and an arbitrary nonlinearity's influence on variance through simple corrective scalars. Since Batch Normalization trained with dropout estimates the variance of a layer's incoming distribution with some inputs dropped, the variance also differs between train and test. After training a network with Batch Normalization and dropout, we simply update Batch Normalization's variance moving averages with dropout off and obtain state of the art on CIFAR-10 and CIFAR-100 without data augmentation.