Baidu USA LLC

Joel Hestness

Hee Woo Jun

Sercan Omer Arik

Gregory Diamos

Kenneth W Church

Ryan Prenger

Adam Coates

Andrew Gibiansky

Christopher Fougner

Yanqi Zhou

Sharan Narang

Markus Kliegl

Rewon Child

Newsha Ardalani

Md Mostofa Ali Patwary

As deep learning application domains grow, a deeper understanding of the relationships between training set size, computational scale, and model accuracy improvements is extremely beneficial. Presented herein are large-scale empirical study of error and model size growth as training sets grow. Embodiments of a methodology for this measurement are introduced herein as well as embodiments for predicting other metrics, such as compute-related metrics. It is shown herein that power-law may be used to represent deep model relationships, such as error and training data size. It is also shown that model size scales sublinearly with data size. These scaling relationships have significant implications on deep learning research, practice, and systems. They can assist model debugging, setting accuracy targets, and decisions about data set growth. They can also guide computing system design and underscore the importance of continued computational scaling.

Predicting deep learning scaling

Described herein are systems and methods for word embeddings to avoid the need to throw out rare words appearing less than a certain number of times in a corpus. Embodiments of the present disclosure involve group words into clusters/classes for multiple times using different assignments of the vocabulary words to a number of classes. Multiple copies of the training corpus are then generated using the assignments to replace each word with the appropriate class. A word embedding generating model is run on the multiple class corpora to generate multiple class embeddings. An estimate of the gold word embedding matrix is then reconstructed from multiple pairs of assignments, class embeddings, and covariances. Test results show the effectiveness of embodiments of the present disclosure.

Embeddings with classes

Described herein are systems and methods for creating and using Convolutional Recurrent Neural Networks (CRNNs) for small-footprint keyword spotting (KWS) systems. Inspired by the large-scale state-of-the-art speech recognition systems, in embodiments, the strengths of convolutional layers to utilize the structure in the data in time and frequency domains are combined with recurrent layers to utilize context for the entire processed frame. The effect of architecture parameters were examined to determine preferred model embodiments given the performance versus model size tradeoff. Various training strategies are provided to improve performance. In embodiments, using only &#x2dc;230 k parameters and yielding acceptably low latency, a CRNN model embodiment demonstrated high accuracy and robust performance in a wide range of environments.

Convolutional recurrent neural networks for small-footprint keyword spotting

Growing community of inventors

Mountain View, CA, United States of America

Joel Hestness