2014/10/27 by Daniel Povey, Xiaohui Zhang, Povey, Daniel +3
Computer Science · Mathematics · #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Neural Networks and Applications #Neural and Evolutionary Computing (cs.NE) #Speech Recognition and Synthesis #Speech and Audio Processing #cs.LG #cs.NE #stat.ML
paper · pdf · doi:10.48550/arxiv.1410.7455
Accepted as workshop contribution to ICLR 2015. 12 pages plus 16 pages of appendices, International Conference on Learning Representations (ICLR): Workshop track, 2015. [2 sets of minor fixes post-publication.]
openalex publication_date 2014/10/27 · arxiv created 2015/06/22 · arxiv updated 2015/06/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
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.