An In-depth Study of Stochastic Backpropagation
This addresses memory efficiency for deep learning practitioners, but it is incremental as it builds on existing backpropagation and dropout techniques.
The paper tackles the problem of high GPU memory and computational cost in training deep neural networks by proposing Stochastic Backpropagation (SBP), which uses a subset of feature maps during backpropagation, resulting in up to 40% GPU memory savings with less than 1% accuracy degradation.
In this paper, we provide an in-depth study of Stochastic Backpropagation (SBP) when training deep neural networks for standard image classification and object detection tasks. During backward propagation, SBP calculates the gradients by only using a subset of feature maps to save the GPU memory and computational cost. We interpret SBP as an efficient way to implement stochastic gradient decent by performing backpropagation dropout, which leads to considerable memory saving and training process speedup, with a minimal impact on the overall model accuracy. We offer some good practices to apply SBP in training image recognition models, which can be adopted in learning a wide range of deep neural networks. Experiments on image classification and object detection show that SBP can save up to 40% of GPU memory with less than 1% accuracy degradation.