Parallel training of DNNs with Natural Gradient and Parameter Averaging
This work addresses scalability issues in speech recognition for researchers and practitioners using the Kaldi toolkit, though it is incremental as it builds on existing methods.
The paper tackles the problem of training deep neural networks with large datasets across multiple machines while minimizing network traffic, achieving this through a combination of periodic parameter averaging and an efficient natural gradient approximation that improves convergence.
We describe the neural-network training framework used in the Kaldi speech recognition toolkit, which is geared towards training DNNs with large amounts of training data using multiple GPU-equipped or multi-core machines. In order to be as hardware-agnostic as possible, we needed a way to use multiple machines without generating excessive network traffic. Our method is to average the neural network parameters periodically (typically every minute or two), and redistribute the averaged parameters to the machines for further training. Each machine sees different data. By itself, this method does not work very well. However, we have another method, an approximate and efficient implementation of Natural Gradient for Stochastic Gradient Descent (NG-SGD), which seems to allow our periodic-averaging method to work well, as well as substantially improving the convergence of SGD on a single machine.