Local Critic Training for Model-Parallel Learning of Deep Neural Networks
This method aims to improve the efficiency of training deep neural networks for researchers and practitioners by reducing training time and memory consumption through model parallelism.
This paper introduces local critic training, a model-parallel learning method that divides neural networks into layer groups, each updated by a dedicated local critic network. This approach decouples the update process for CNNs and RNNs, leading to satisfactory performance, reduced training time, and decreased memory consumption per machine.
In this paper, we propose a novel model-parallel learning method, called local critic training, which trains neural networks using additional modules called local critic networks. The main network is divided into several layer groups and each layer group is updated through error gradients estimated by the corresponding local critic network. We show that the proposed approach successfully decouples the update process of the layer groups for both convolutional neural networks (CNNs) and recurrent neural networks (RNNs). In addition, we demonstrate that the proposed method is guaranteed to converge to a critical point. We also show that trained networks by the proposed method can be used for structural optimization. Experimental results show that our method achieves satisfactory performance, reduces training time greatly, and decreases memory consumption per machine. Code is available at https://github.com/hjdw2/Local-critic-training.