Microsoft Technology Licensing, LLC

Google Inc.

Jui-Ting Huang

Jinyu Li

Yifan Gong

Dong Yu

Li Deng

Rui Zhao

Abhishek Gupta

Francoise Beaufays

Brian Strope

Yun-hsuan Sung

Hui Lin

Siegfried Joseph Bilstein

Systems and methods are provided for generating a DNN classifier by 'learning' a 'student' DNN model from a larger more accurate &#x201c;teacher&#x201d; DNN model. The student DNN may be trained from un-labeled training data because its supervised signal is obtained by passing the un-labeled training data through the teacher DNN. In one embodiment, an iterative process is applied to train the student DNN by minimize the divergence of the output distributions from the teacher and student DNN models. For each iteration until convergence, the difference in the output distributions is used to update the student DNN model, and output distributions are determined again, using the unlabeled training data. The resulting trained student model may be suitable for providing accurate signal processing applications on devices having limited computational or storage resources such as mobile or wearable devices. In an embodiment, the teacher DNN model comprises an ensemble of DNN models.

Learning student DNN via output distribution

System and methods for generating a shareability score in accordance with some example embodiments are disclosed. A social networking system receives a request to generate a shareability score for a list of content items for an organization. The social networking system identifies a plurality of members associated with the organization and analyzes past share data for the plurality of members to generate an organization sharing profile. The social networking system retrieves early sharing information for each content item in the list of content items. The social networking system generates a shareability score for each particular content item and ranks the list of content items based on the generated shareability scores. The social networking system then transmits the ranked list of content items to a client device, receives a selection of one or more content items, and broadcasts the one or more selected items to a plurality of client devices.

Shareability score

Described herein are various technologies pertaining to a multilingual deep neural network (MDNN). The MDNN includes a plurality of hidden layers, wherein values for weight parameters of the plurality of hidden layers are learned during a training phase based upon training data in terms of acoustic raw features for multiple languages. The MDNN further includes softmax layers that are trained for each target language separately, making use of the hidden layer values trained jointly with multiple source languages. The MDNN is adaptable, such that a new softmax layer may be added on top of the existing hidden layers, where the new softmax layer corresponds to a new target language.

Multilingual deep neural network

Speech recognition systems may perform the following operations: receiving audio; recognizing the audio using language models for different languages to produce recognition candidates for the audio, where the recognition candidates are associated with corresponding recognition scores; identifying a candidate language for the audio; selecting a recognition candidate based on the recognition scores and the candidate language; and outputting data corresponding to the selected recognition candidate as a recognized version of the audio.

Recognizing speech in multiple languages

Growing community of inventors

Fremont, CA, United States of America

Jui-Ting Huang