Google Inc.

Amazon Technologies, Inc.

Webtv Networks, Inc.

Wei Huang

Shiladitya Roy

Sarthak Anand

Ravi Chikkanayakanahalli Mallikarjuniah

Fabio Soldo

Matthew Nirvan Sharifi

Mark Wagner

Julian James Odell

Daniel Ramage

Tim Phillip Wantland

Abhanshu Sharma

Robert Mark Waugh

Nitin Kesarwani

Alexa Greenberg

Muhammad Mohsin

Andrew E Rubin

Timo A Bruck

Bruce A Leak

Dean J Blackketter

Randall Sarafa

Andrew T McFadden

Samuel Thomas Scott, Iii

Richard Eugene Fulcher

Zhenyu Liu

Priya Rao Chagaleti

Cole Ira Brendel

Christopher Forbes Will

Jing Cao

Yanchao Su

Seungyeon Kim

Kaikai Wang

Jacek Slawosz Jurewicz

Surbhi Maheshwari

Nicholas Kong

Benjamin Sidhom

Iulia Turc

Joshua Patrick Gardner

Michael D Dodd

Emmanuel Pierre Devillard

Alan Bishop

Alan Cohn

Andy Rubin

Michael A Killianey

Mayank Thakore

Michael Daub

Hasan Nuzhet Atay

Shuo Li

Arne Mauser

Alexander E Mayorov

George Xie

Evan Miller

Matt Nakoneczny

Eric M Tucker

Yan Cui

Yingwei Li

Longhua You

Ross William Snider

Chunzhang Mo

Tianbin Jiang

Tianyi Pan

Sharon Xuelun Yang

Matthew Leach

Bhawana Goel

Li He

The present disclosure describes techniques for training a model using cross-domain adaptation to classify content requests from client devices having unknown attributes. The system can obtain requests for content from client devices of a first domain, and requests for content from client devices of a second domain. The system can train a model by propagating request attributes of the first domain through the model to generate first internal data from an internal layer of the model and a first output vector an output layer of the model. The system can propagate request attributes of the second domain to generate second internal data, and determine a difference between the first internal data and the second internal data. The system can update the model based on the difference and the output vector, and classify a third client device of the second domain using the model.

Cookie space domain adaptation for device attribute prediction

A system and method are disclosed for identifying a media item to be provided to a group of users of a content sharing platform, wherein the media item is associated with a category, and wherein each user in the group of users is associated with a weight indicating a probability of a correspondence between a respective user and the category associated with the media item, receiving a request to change a size of the group of users from a first level to a second level, the first level corresponding to a first weight threshold for the group of users, and calculating, based a value indicating a difference between the first level and the second level, a second weight threshold for the group of users corresponding to the second level, the second weight threshold to be subsequently used to determine whether the media item is to be provided to a user requesting content from the content sharing platform.

System and methods for changing a size of a group of users to be presented with a media item

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for reducing the difference in performance of a model across groups and sub-groups within the same group of users with similar characteristics for providing digital components. Methods can include identifying, a loss function that generates a loss representing a measure of performance the model seeks to optimize during training. The loss function is modified by adding an additional term to the loss function. The model is trained using the modified loss function. A request for digital component is received that includes a user group identifier. The model generates one or more user characteristics based on which one or more digital components are selected and transmitted to the client device of the user.

Robust model performance across disparate sub-groups within a same group

This document describes techniques for expanding user groups while preserving user privacy and data security. In one aspect, a method includes receiving, by a content platform and from a client device of a user, a request for a digital component that also includes a user identifier. A determination is made that the user identifier is included in a user list that includes multiple user identifiers respectively corresponding to multiple users in a user action group. In response to determining that the unique identifier is included in the user list, a digital component of the entity for which the user list is generated is selected and provided to the client device of the user for display to the user of the client device.

Privacy preserving user group expansion

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for selecting and distributing digital components based on predicted user attributes of users are described. In one aspect, a method includes obtaining data indicating content categories of content of the content pages accessed by the user during the user visits. A determination is made for an aggregate measure of each content category based on a quantity of user visits to content pages of the electronic resource of the publisher that included content classified as belonging to the content category. User attribute prediction data indicating previously predicted user attributes of the user is obtained. User attributes are predicted for the current visit of the user to the electronic resource of the publisher that is further used to select digital components for display with the electronic resource on a client device during the current visit.

Distributing digital components based on predicted attributes

The present disclosure is directed to input suggestion. In particular, the methods and systems of the present disclosure can: receive, from a first application executed by one or more computing devices, data indicating information that has been presented by and/or input into the first application; generate, based at least in part on the received data, one or more suggested candidate inputs for a second application executed by the computing device(s); provide, in association with the second application, an interface comprising one or more options to select at least one suggested candidate input of the suggested candidate input(s); and responsive to receiving data indicating a selection of a particular suggested candidate input of the suggested candidate input(s) via the interface, communicate, to the second application, data indicating the particular suggested candidate input.

Methods and systems for input suggestion

Techniques for improving systems for tracking data access are provided. A system for accessing and encoding data may, upon receiving a request for data, determine whether the data needs to be encoded for the purpose of tracking access to the data. If the system determines that the data needs to be encoded, the system may further determine the information which is to be encoded into the data, such as a client ID/user ID, timestamp, etc. The system may further determine an encoding scheme for the data such as a least significant bit algorithm. The system may further determine encoding data based on the encoding scheme and the information which is to be encoded into the data. The system may also encode the encoding data into the data based on the encoding scheme.

Accessing and encoding data using a least significant bit encoding algorithm

Techniques for generating a personalization value that measures how tailored certain system interactions are for a user are described. A dialog exchange between a user and a skill may be determined, with the dialog exchange including user input data and system output data. It may be determined that the system output data was generated without respect to at least one previous user input or system output of the dialog exchanges. Based on this, a personalization value may be generated and sent to the skill.

Natural language dialog scoring

A method includes obtaining time series data for a usage or performance metric for computing resources in a service provider network comprising a plurality of observations recorded in a plurality of respective time steps. A prediction error is determined for a previous prediction of an observation in the time series data. The prediction error is used to update a standard deviation of a set of predication errors for the usage or performance metric. The standard deviation and the prediction error are then used to update a confidence coefficient. A prediction limit for the usage or performance metric is then determined based on an expected value, the confidence coefficient, and the standard deviation. One or more events may be generated based on the prediction limit, which may be used to trigger a reconfiguration or auto-scaling of the computing resources.

Metrics prediction using dynamic confidence coefficients

The present disclosure provides various metrics that may be used to generate a score representing a conversational quality of a user/computing system dialog. User input data, corresponding to a dialog with a natural language processing system, may be determined and associated with a dialog identifier. System output data, corresponding to the dialog, may also be determined and associated with the dialog identifier. After the system output data is output to a user, the user input data and system output data may be processed to determine a score corresponding to the dialog exchange.

Techniques for generating a conformance score for a system/user dialog are described. A conformance score may represent a degree to which output data, provided by a skill, conforms to various policies (e.g., the data includes content appropriate for the age of the user, the data does not include profanity, etc.). User input data and system output data, corresponding to a dialog exchange between a user and a skill, ma be determined. A user type associated with the dialog exchange may also be determined. Based on the user type and the system output data, it may be determined that one or more filtering resources are to be assigned to process future data, received from the skill, prior to the future data being presented to a user.

Techniques for generating a prompt coverage score, which measures an extent to which data output to a user during a dialog is repetitive and monotonous, are described. User input data and system output, corresponding to a dialog exchange between a user and a skill, may be determined. A portion of the system output data, corresponding to a system prompt representing default output data, may be determined. A first number, representing possible variants of the prompt, may be determined along with a second number, representing variants of the prompt output during the dialog exchange. A prompt coverage score may be determined based on the first and second numbers.

Systems and methods are described for the collection and transmission of virtual machine resource operations information. Individual agents on virtual machine resources collect and store operations information in accordance with a current operations information collection configuration. The individual agents will initiate a transmission of the collected operations information. Responsive to the receipt of the transmission of the collected operations information, the monitoring processing service calculates a hierarchy of anomaly scores utilizing machine learning techniques. The monitoring processing service can generate a processing result.

Dynamically configurable operation information collection

Growing community of inventors

Seattle, WA, United States of America

Wei Huang