Google Inc.

X Development LLC

Loon LLC

Keith Allen Bonawitz

Daniel Ramage

Johan Mathe

Adrien Treuille

David Petrou

Bradley J Rhodes

Joshua L Weaver

Eric Teller

Richard Wayne Devaul

Richard Wayne DeVaul

C O Lee Boyce, Jr

Dan P Piponi

Blaise Hilary Aguera-Arcas

Adam Sadilek

The present disclosure provides systems and methods for distributed training of machine learning models. In one example, a computer-implemented method is provided for training machine-learned models. The method includes obtaining, by one or more computing devices, a plurality of regions based at least in part on temporal availability of user devices; selecting a plurality of available user devices within a region; and providing a current version of a machine-learned model associated with the region to the plurality of selected user devices within the region. The method includes obtaining, from the plurality of selected user devices, updated machine-learned model data generated by the plurality of selected user devices through training of the current version of the machine-learned model associated with the region using data local to each of the plurality of selected user devices and generating an updated machine-learned model associated with the region based on the updated machine-learned model data.

Systems and methods for distributed on-device learning with data-correlated availability

The present disclosure provides systems and methods that leverage machine-learned models in conjunction with user-associated data and disease prevalence mapping to predict disease infections with improved user privacy. In one example, a computer-implemented method can include obtaining, by a user computing device associated with a user, a machine-learned prediction model configured to predict a probability that the user may be infected with a disease based at least in part on user-associated data associated with the user. The method can further include receiving, by the user computing device, the user-associated data associated with the user. The method can further include providing, by the user computing device, the user-associated data as input to the machine-learned prediction model, the machine-learned prediction model being implemented on the user computing device. The method can further include receiving, by the user computing device, a current disease prediction for the user as an output of the machine-learned prediction model. The method can further include providing, by the user computing device, data indicative of the current disease prediction for the user to a central computing system for use in updating a prevalence map that models prevalence of the disease over a plurality of geographic locations.

Privacy-first on-device federated health modeling and intervention

Systems and methods are shown for providing private local sponsored content selection and improving intelligence models through distribution among mobile devices. This allows greater data gathering capabilities through the use of the sensors of the mobile devices as well as data stored on data storage components of the mobile devices to create predicted models while offering better opportunities to preserve privacy. Locally stored profiles comprising machine intelligence models may also be used to determine the relevance of the data gathered and in improving an aggregated model for identifying the relevance of data and the selection of sponsored content items. Distributed optimization is used in conjunction with privacy techniques to create the improved machine intelligence models. Publishers may also benefit from the improved privacy by protecting the statistics of type or volume of sponsored content items shown with publisher content.

Systems and methods for private local sponsored content

The present disclosure provides systems and methods for the management and/or evaluation of machine-learned models based on locally logged data. In one example, a user computing device can obtain a machine-learned model (e.g., from a server computing device) and can evaluate at least one performance metric for the machine-learned model. In particular, the at least one performance metric for the machine-learned model can be evaluated relative to data that is stored locally at the user computing device. The user computing device and/or the server computing device can determine whether to activate the machine-learned model on the user computing device based at least in part on the at least one performance metric. In another example, the user computing device can evaluate a plurality of machine-learned models against locally stored data. At least one of the models can be selected based on the evaluated performance metrics.

Management and evaluation of machine-learned models based on locally logged data

Example implementations may relate to selection between a first mode and a second mode. The first mode may involve (i) directing an aerial vehicle (e.g., in an aerial network including a plurality of aerial vehicles) to navigate to each of a plurality of altitudes and (ii) determining respective wind-related data at each respective altitude. Whereas, the second mode may involve (i) selecting at least one altitude based on the determined wind-related data and (ii) directing the aerial vehicle to reposition to the at least one selected altitude. As such, a control system may determine flight data for the aerial vehicle. Based on the flight data, the control system may make a selection between the first mode and the second mode. And based on the selection, the control system may then operate the aerial vehicle according to the first mode or may operate the aerial vehicle according to the second mode.

Selection between explore mode and control mode for aerial vehicle

Example methods and systems for determining when to launch vehicles into a fleet of autonomous vehicles are described. A method comprises receiving a sequence of coverage requirements for a region over a period of time. The region may be characterized by landmarks and the period of time can be divided into time intervals. The method also includes defining a landmark as a launch site representative of a landmark at which a given vehicle can be added to a plurality of operating vehicles, and determining for a respective landmark, estimated landmarks that can be reached by a vehicle starting from the respective landmark by an end of a time interval. The method also includes based on the sequence of coverage requirements and the estimated landmarks, determining a given launch site and corresponding time interval at which to add the given vehicle to the plurality of operating vehicles.

Methods and systems for determining when to launch vehicles into a fleet of autonomous vehicles

Disclosed herein are embodiments of a balloon-based positioning system and method. In one example embodiment, a system includes a group of at least three balloons deployed in the stratosphere and a control system configured for: determining a first set of spatial relationships relating to the group; determining a second set of spatial relationships relating to at least a portion of the group and to a reference point; determining a position of the reference point relative to the earth; using the determined first set, the determined second set, and the determined position of the reference point relative to the earth as a basis for determining a position of a target balloon in the group relative to the earth; and transmitting the determined position of the target balloon relative to the earth.

Balloon-based positioning system and method

Example methods and systems for determining when to decommission vehicles from a fleet of autonomous vehicles are described. One method includes receiving information indicating a sequence of coverage requirements for a region over a period of time, and the region is characterized by landmarks and the period of time is divided into time intervals. Landmarks may be defined as a decommissioning site representative of a location at which a given vehicle can be taken out of service, and the method includes determining estimated landmarks that can be reached by one or more vehicles starting from a respective landmark by an end of a respective time interval, and based on the sequence of coverage requirements, determining which vehicles from among the vehicles that can reach the one or more landmarks defined as the decommissioning site to take out of service.

Methods and systems for determining when to decommission vehicles from a fleet of autonomous vehicles

An example computer-implemented method involves determining that an aerostat is located within a first atmospheric layer. A first wind-velocity measure is indicative of wind velocity in the first atmospheric layer. The method also involves determining an altitude of an atmospheric layer boundary between the first atmospheric layer and a second atmospheric layer. A second wind-velocity measure is indicative of wind velocity in the second atmospheric layer, and the second wind-velocity measure differs from the first wind-velocity measure by at least a threshold amount. The method further involves determining a desired location for the aerostat, and during at least a portion of a flight towards the desired location, causing the aerostat to traverse back and forth across the atmospheric layer boundary while remaining within a predetermined vertical distance from the atmospheric layer boundary in order to achieve a desired horizontal trajectory.

Growing community of inventors

Berkeley, CA, United States of America

Keith Allen Bonawitz