Experiments on Parallel Training of Deep Neural Network using Model Averaging

TL;DR

The paper addresses accelerating deep neural network training for speech recognition by employing data-parallel training with frequent parameter averaging across GPUs via MPI. It analyzes the role of Natural Gradient SGD and Restricted Boltzmann Machine pretraining within this framework, and empirically demonstrates substantial speedups on Switchboard-300h with manageable degradation in decoding accuracy. The study systematically explores averaging frequency, minibatch size, and learning-rate schedules, showing that NG-SGD and RBM pretraining substantially improve convergence and efficiency. It also outlines future directions, including convergence theory and hardware-aware optimizations, to further enhance parallel training performance.

Abstract

In this work we apply model averaging to parallel training of deep neural network (DNN). Parallelization is done in a model averaging manner. Data is partitioned and distributed to different nodes for local model updates, and model averaging across nodes is done every few minibatches. We use multiple GPUs for data parallelization, and Message Passing Interface (MPI) for communication between nodes, which allows us to perform model averaging frequently without losing much time on communication. We investigate the effectiveness of Natural Gradient Stochastic Gradient Descent (NG-SGD) and Restricted Boltzmann Machine (RBM) pretraining for parallel training in model-averaging framework, and explore the best setups in term of different learning rate schedules, averaging frequencies and minibatch sizes. It is shown that NG-SGD and RBM pretraining benefits parameter-averaging based model training. On the 300h Switchboard dataset, a 9.3 times speedup is achieved using 16 GPUs and 17 times speedup using 32 GPUs with limited decoding accuracy loss.

Figures (2)

• Figure 1: All-reduce network
• Figure 2: Speedup factor v.s. number of gpus