Apple Inc.

Google Inc.

Deepmind Technologies Limited

Ashish Agarwal

Vivek Raghunathan

Arash Baratloo

Chiu Wah Kelvin So

Qicheng Ma

Vojislav Samsalovic

Max Muller

Alice S Tull

Nicholas Chamandy

Fei Peng

Diane L Tang

Diane Lambert

Matthew H Thorn

Wenyu Zhang

Michael M Meyer

Daryl Pregibon

Mithilesh Kumar

Ilia Mirkin

Sean Michael Harold

James Angus Morrison

Eric Bauer Arbanovella

Susan Shannon

Douglas Isaac Friedman

Leticia Marie Alarcón

Ian Aric Mondragon

Rahul Ravikumar

Calvin Lin

Philip Stanger

Robert William Mayor

David Benjamin Millman

Ricardo D Cortes

Vincent P Arroyo

Justin M Strawn

Vitali Lovich

Darin Tay

Joseph Ding-Jiu Huang

Austin Marusco

Abhinav R Patel

Ashish Sharma

Michael H Walker

Jessica Noel Tsoong

David Benjamin Pfau

David Barrett

Zhui Deng

Stig Petersen

Kimberly Stachenfeld

Trystan Peter Kosmynka

Alexander Ashley Miyamura

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network having a plurality of network parameters and being configured to process an input data item to generate a feature representation comprising a values for each of a plurality of features of the input data item.

Training spectral inference neural networks using bilevel optimization

The disclosed techniques generally relate to devices and methods for generating cropped images in which one or more features of interest are preserved or emphasized. In one implementation, such features of interest may include facial features. In accordance with certain implementations, location and extent of such features may be identified and used in conjunction with information regarding display characteristics so as to generate cropped images preserving or emphasizing the feature of interest when displayed on a given device.

Dynamic image analysis and cropping

In some implementations, a computing device can determine the similarity of binary executables. For example, the computing device can receive an application, including a binary executable. The computing device can generate function signatures for the functions called within the binary executable. The computing device can generate a locality sensitive hash value for the application based on the function signatures. The computing device can group applications based on the locality sensitive hash value generated for each application. The computing device can compare the function signatures of the binary executables of the applications within a group to determine the similarity of the applications. If two applications have binary executables that are over a threshold percentage of similarity, the two applications can be identified as clones of each other.

Determining the similarity of binary executables

The present technology consists of an automatic application testing tool that can initiate execution of an application submitted for testing by an application developer. The application testing tool can analyze views of the application while the application is executing to determine navigation objects and possible interactions with the navigation objects. The application testing tool can automatically apply possible interactions to the navigation objects and can record application execution data during the testing.

Testing tool for testing applications while executing without human interaction

Techniques for storing information representing a venue are described. A venue data provider may upload venue data to a venue data service. The venue data can include locations, geometries, and business categories of the buildings and floors. The venue data service can validate the venue data, and request a survey service to survey the venue using the validated venue data. A survey can include recording measurements of an environment variable at various locations of the venue. The location service can generate location fingerprint data from the survey. The location fingerprint data can include expected measurements of the environment variable at various locations of the building. The venue data service can provide the location fingerprint data to a mobile device such that the mobile device can determine a location of the mobile device in the venue using sensor readings of the environment variable.

Venue data framework

Methods, systems, and apparatus, including computer programs encoded on computer storage media are provided. An example method includes receiving user input in a control, for example, one associated with a search engine. The user input includes a series of input characters that when completed would form a completed user request. The method further includes, while receiving the user input, evaluating the characters to identify completions for the completed request. The method further includes identifying, by one or more processors, a content item associated with the one or more completions and identifying display data associated with the content item for transmission to the user along with one or more completions in response to receiving the user input.

Selecting and presenting content relevant to user input

Techniques for generating maps of venues including buildings and floors

Methods, systems, and computer program products for polygonal routing are described. A computer system can provide turn-by-turn navigation in a venue for a mobile device using a navigation graph. The navigation graph can include nodes representing a series of navigation areas leading from a start point to an end point in a venue including indoor space. Each navigation area can be a polygon occupying a non-zero geographic area. The computer system updates the turn-by-turn instructions when the mobile device enters or exits a navigation area in the series of navigation areas, until the device reaches the end point.

Polygonal routing

Methods, systems, and computer program product for generating location fingerprint data for a venue are described. A sampling device surveying a venue can move inside the venue. While the sampling device moves, the sampling device can record environmental readings including, for example, strengths of signals from multiple radio signal sources. The sampling device can take the recording at fixed or various time intervals. Meanwhile, the sampling device can determine, based on a starting point and motion sensor readings, an estimated location of the mobile device for each time the sampling device takes the recordings. The sampling device can display a motion path of the estimated locations and a user interface item for receiving a user input for correcting the estimated locations. The sample device can tag the recorded environmental readings with the corrected locations, and submit the tagged readings to a server for determining a location fingerprint for the venue.

Indoor location survey assisted by a motion path on a venue map

A system and method for integrating media productions imported from a plurality of content provider systems are described. The system and method may communicate with a plurality of content provider systems to import a set of media productions from each content provider system and may normalize each media production into a normalized media document format. The media productions may be matched to each other based on the normalized media documents. The system and method may also create and store a plurality of canonical objects based on the matches determined for the media productions. For example, if a given media production matches one or more other media productions then all of the matching media productions may be linked together by a canonical object, e.g., so that the canonical object functions as an abstract representation of all of the matching media productions.

Media system with canonical architecture for integrating media productions from different content providers

User queries are received, with each query requesting a service from a server. Overlapping experiments are performed on at least a portion of the queries, with each experiment modifying one or more parameters associated with the queries or parameters associated with processing of the queries, and with the experiments organized into layers. Two or more experiments in different layers are allowed to be performed on the same query, and for any given layer, at most one experiment is allowed to be performed on the same query.

Overlapping experiments

Methods, systems, and apparatus, including computer programs encoded on computer storage media are provided. One example method includes receiving a textual input that forms a part of a user request, the textual input in the form of a prefix that when complete constitutes the request. The method further includes identifying completions based on the prefix, determining one or more content items associated with each completion, computing initial scores including a score for each content item-completion pair including determining a likelihood that a given completion represents the request and a quality of the content item as compared to the completion, computing final scores including scores for each prefix-content item pair using the initial scores, computing a bid for each content item, running an auction to select a content item based on the bids and final scores, and identifying display data associated with the selected content item.

Dynamically selecting and presenting content relevant to user input

Growing community of inventors

Sunnyvale, CA, United States of America

Ashish Agarwal