Abstract: Systems and methods are provided for determining an order of execution of a neural network. For instance, data indicative of a neural network and data indicative of an amount of available memory in a constrained memory space can be obtained. The neural network can include a plurality of operators. An order of execution associated with the neural network can then be determined. The order of execution specifies an order in which to execute each of the plurality of operators. The order of execution is determined based at least in part on the available memory in the constrained memory space. In particular, one or more graph search algorithms can be performed on a graph that is representative of the neural network to obtain the order of execution.

Abstract: The present disclosure describes methods and apparatuses for satellite-based narrow-band communication. In some aspects, the systems include a user device (102) that establishes a Narrow-Band Internet of Things (NB-IoT) wireless connection (106) with one or more orbital satellites (104) in order to transmit (108) and receive (110) data when other wireless connections are not available. The data may be relayed by the orbital satellite (104) to and from a base station (202) on Earth.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for context-adaptive scanning of digital components. In one aspect, a method comprises: selecting a given digital component from among a plurality of digital components based on a current scanning priority of the given digital component; scanning the given digital component, comprising determining a current state of the given digital component; determining a current context of the given digital component based on one or more of: (i) the current state of the given digital component, or (ii) a current scan index of the given digital component that specifies a number of times the given digital component has been scanned; determining an updated scanning priority of the given digital component based on the current context of the given digital component; and re-scanning the given digital component according to the updated scanning priority.

Abstract: This specification describes methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a larger neural network from a smaller neural network. One of the described methods includes obtaining data specifying an original neural network and generating a larger neural network from the original neural network The larger neural network has a larger neural network structure than the original neural network structure. The values of the parameters of the original neural network units and the additional neural network units are initialized so that the larger neural network generates the same outputs from the same inputs as the original neural network and the larger neural network is trained to determine trained values of the parameters of the original neural network units and the additional neural network units from the initialized values.

Abstract: A content provider can provide content, such as advertisements or other promotional material, for display by a mobile application. As contextual information is typically limited for mobile applications, the content provider can use other information in addition to or in place of contextual information to select advertisements. This information can include location information, user activity with regards to the mobile application, information regarding a recipient of the advertisement, such as demographic information, information regarding the mobile application, such as information regarding the type of mobile application, and product inventory information. A user profile also may be used to select advertisements. The user profile can include a set of merchants and merchant types, for example clothing retailers, to which a user may be more responsive. The user profiles can be generated and updated by logging merchants that the user visits or interacts with via advertisements.

Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes receiving an audio signal and initiating speech recognition tasks by a plurality of speech recognition systems (SRS's). Each SRS is configured to generate a recognition result specifying possible speech included in the audio signal and a confidence value indicating a confidence in a correctness of the speech result. The method also includes completing a portion of the speech recognition tasks including generating one or more recognition results and one or more confidence values for the one or more recognition results, determining whether the one or more confidence values meets a confidence threshold, aborting a remaining portion of the speech recognition tasks for SRS's that have not generated a recognition result, and outputting a final recognition result based on at least one of the generated one or more speech results.

Abstract: A method includes generating a utility portal interface in response to a request from a utility computer system that receives parameters that specify a demand response event; providing a display of groups of energy-consuming locations that are available to be selected to participate in the demand response event; providing a display of an energy demand profile for the utility during the demand response event; receiving a selection of a subset of the groups of energy-consuming locations to participate in the demand response event; causing the display of the energy demand profile for the utility during the demand response event to be dynamically updated as the subset of the groups of energy-consuming locations are selected or deselected by the utility computer system to participate; and sending transmissions to thermostats associated with the subset of the groups of energy-consuming locations to execute the demand response event.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for obtaining, for each of multiple words or sub-words, audio data corresponding to multiple users speaking the word or sub-word; training, for each of the multiple words or sub-words, a pre-computed hotword model for the word or sub-word based on the audio data for the word or sub-word; receiving a candidate hotword from a computing device; identifying one or more pre-computed hotword models that correspond to the candidate hotword; and providing the identified, pre-computed hotword models to the computing device.

Abstract: In general, the subject matter described in this specification can be embodied in methods, systems, and program products for receiving user input that defines a search query, and providing the search query to a server system. Information that a search engine system determined was responsive to the search query is received at a computing device. The computing device is identified as in a first state, and a first output mode for audibly outputting at least a portion of the information is selected. The first output mode is selected from a collection of the first output mode and a second output mode. The second output mode is selected in response to the computing device being in a second state and is for visually outputting at least the portion of the information and not audibly outputting the at least portion of the information. At least the portion of information is audibly output.

Abstract: In a computing resource environment including at least one resource capable of being allocated to at least one of a plurality of tasks, techniques are disclosed for applying a taint to a resource in a computing resource environment, the taint being configured to prevent the resource from being claimed for a resource request without a toleration to that taint. Variations include receiving, at a resource scheduler in the resource environment, a request to allocate the resource to perform a particular task and determining whether the resource is subject to a taint. If the resource is subject to a taint, analyzing the request to determine if it includes a toleration for the taint. If the request includes a toleration for the taint, allocating the resource to the task. If the request does not include a toleration for the taint, not allocating the resource due to the taint.

Abstract: A method for determining a placement for machine learning model operations across multiple hardware devices is described.

Abstract: Methods and systems for providing a third party application with access to files stored on a server are disclosed. A method may include receiving, from a browser at a client device, a request for a file stored on the server, wherein the request is received via a web page provided by the third party application and rendered by the browser, the web page comprising an embedded user interface (UI) component associated with the server to access the file stored on the server, wherein the request includes a document identifier associated with the file, an application identifier of the third-party application, and an origin identifier, wherein the origin identifier is associated with the web page provided by the third party application and rendered by the browser.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for suppressing hotwords are disclosed. In one aspect, a method includes the actions of receiving audio data corresponding to playback of an utterance. The actions further include providing the audio data as an input to a model (i) that is configured to determine whether a given audio data sample includes an audio watermark and (ii) that was trained using watermarked audio data samples that each include an audio watermark sample and non-watermarked audio data samples that do not each include an audio watermark sample. The actions further include receiving, from the model, data indicating whether the audio data includes the audio watermark. The actions further include, based on the data indicating whether the audio data includes the audio watermark, determining to continue or cease processing of the audio data.

Abstract: Implementations relate to relate to interface elements for directed display of content data items. In some implementations, a method includes causing a subset of content elements to be presented in a display view of a graphical user interface on a device, the set of content elements corresponding to a collection of content data items. Search topics are identified based on one or more content characteristics associated with content elements that are within a threshold distance of the subset of content elements in the display view. Topic elements are displayed in the interface, corresponding to the search topics. In response to a selection of a particular topic element based on user input, the collection of content data items is searched to identify resulting content data items that have content characteristics corresponding to the selected search topic. The display view is updated to present resulting content elements.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for presenting content items. In one aspect, a method includes identifying an opportunity to present a content item at a device and in a manner that precludes a user from accessing content of one or more applications executing on the device. Data relating to a status of each of the one or more applications are identified. A level of urgency for a user to access content of the one or more applications is determined based on the data. A response to the opportunity to present a content item is selected based on the level of urgency for the user to access content of the one or more applications. The response is selected from a first response that causes presentation of the content item and a second response that precludes presentation of the content item.

Abstract: A device receives a base video that includes an area in frames of the base video reserved for being overlaid with metadata. The device determines a first set of videos to be played next for a particular user. A selection of the first set of videos is based on a second set of videos associated with the particular user. The device receives metadata for the first set of videos and populates an executable presentation template for the base video with a set of extracted metadata. The device plays the base video and synchronizes execution of the populated presentation template with the playing of the base video to overlay the reserved area of the frames with the set of metadata to create a custom interstitial transition video that informs the particular user about the videos to be played next.

Abstract: Systems and methods relating to communication within a fabric network are presented. The fabric network includes one or more logical networks that enables devices connected to the fabric to communicate with each other using various profiles known to the devices. A device sending a message may follow a general message format to encode the message so that other devices in the fabric may understand the message regardless of which logical networks the devices are connected to. Within the message format, a payload of data may be included for the receiving device to forward, store, or process the message. The format and the contents of the payload may vary according to a header within the payload that indicates a profile and a message type within the profile. Using the profile and message type, the receiving devices may decode the message to process the message.

Abstract: An example of a system and method implementing a live migration of a guest on a virtual machine of a host server to a target server is provided. For example, a host server may utilize a flow key to encrypt and decrypt communications with a target server. This flow key may be encrypted using a receive master key, which may result in a receive token. The receive token may be sent to the Network Interface Controller of the host server, which will then encrypt the data packet and forward the information to the target server. Multiple sender schemes may be employed on the host server, and various updates may take place on the target server as a result of the new location of the migrating guest from the host server to the target server.

Abstract: A device may be configured to receive a query; receive information regarding documents that are relevant to the query; identify one or more entities associated with the documents; determine a category for the query based on: the query, a topic of the documents, and the one or more entities; determine, based on the query and the category, that an entity list should be presented in response to the query; and present a search result document based on determining that the entity list should be presented in response to the query. The search result document may include a list with information identifying the one or more entities.

Abstract: Methods and systems for identifying video clips in a video clip repository that are of the same event are disclosed. Video clips are identified as being of the same event based on timestamps and geolocation stamps associated with the video clips. In one embodiment, authors of video clips receive messages that notify them of the existence of other video clips of the same event.