Google Inc.

Baidu USA LLC

Vinay Rao

Anurag Gupta

Michael David McNally

Sanjeev Satheesh

Adam Coates

Zhenyao Zhu

Eric Dean Battenberg

Christopher Fougner

Anuroop Sriram

Hairong Liu

Jascha Sohl-Dickstein

Fei Xiao

Heewoo Jun

Markus Kliegl

Ioan Marius Pungaru

Ajay Kannan

Yashesh Gaur

Rewon Child

Atul Kumar

David Seetapun

Bill Davis

Jiaji Huang

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing inputs using a neural network system that includes one or more pre-normalized layers or one or more regularization normalization layers.

Neural networks with pre-normalized layers or regularization normalization layers

Described herein are systems and methods to identify and address sources of bias in an end-to-end speech model. In one or more embodiments, the end-to-end model may be a recurrent neural network with two 2D-convolutional input layers, followed by multiple bidirectional recurrent layers and one fully connected layer before a softmax layer. In one or more embodiments, the network is trained end-to-end using the CTC loss function to directly predict sequences of characters from log spectrograms of audio. With optimized recurrent layers and training together with alignment information, some unwanted bias induced by using purely forward only recurrences may be removed in a deployed model.

Systems and methods for principled bias reduction in production speech models

This specification describes technologies relating to detecting anomalous user accounts. A computer implemented method is disclosed which evaluates an unknown status user account. The method described compares features associated with a plurality of known anomalous user accounts stored in a database to features present in the unknown account. A correlation value corresponding to the probability of a specific feature occurring in a particular anomalous user account is calculated and a dependence value corresponding to the degree of dependence between the given feature and at least one other feature is also calculated. A subset of features in the unknown account is generated comprising those features that possess a correlation value less than a threshold value and a dependence value below a maximum correlation value. A risk score for the unknown account is calculated by selecting those features from the subset that maximizes the correlation value. The unknown account is then reviewed by an account reviewer if the risk score exceeds a threshold value.

Detection of anomalous instances through dynamic feature selection analysis

Growing community of inventors

Mountain View, CA, United States of America

Vinay Rao