Abstract: Systems, methods, and computer program products for performing semi-supervised contrastive learning of visual representations are provided. For example, the present disclosure provides systems and methods that leverage particular data augmentation schemes and a learnable nonlinear transformation between the representation and the contrastive loss to provide improved visual representations. Further, the present disclosure also provides improvements for semi-supervised contrastive learning.

Abstract: Systems and methods of the present disclosure relate generally to allocation of content resources in a voice activated packet-based computer network. A request for a digital component, by a client computing device, can be fulfilled by multiple content sources. Fulfillment by the different content sources can consume different amounts of computational and network resources. The selection of the best content provider can improve the efficiency and effectiveness of data packet transmission over one or more computer networks.

Abstract: Analyzing or debugging applications is provided. The system identifies an action for an application provided by a developer. The system determines a first classification score based on historical execution of the action. The system generates a machine generated action for the application based on metadata associated with the application. The system determines a second classification score based on a comparison of the action with the machine generated action. The system selects, via a matching program, a second application that matches the application. The system determines a third classification score based on a comparison of an action approved for the second application with the action provided by the application developer. The system updates a delivery control parameter based on the first classification score, the second classification score and the third classification score. The system controls delivery of the application based on the delivery control parameter.

Abstract: Transform-level partitioning of a prediction residual block is performed to improve compression efficiency of video data. During encoding, a prediction residual block is generated responsive to prediction-level partitioning performed against a video block, a transform block partition type to use is determined based on the prediction residual block, a non-recursive transform-level partitioning is performed against the prediction residual block according to the transform block partition type, and transform blocks generated as a result of the transform-level partitioning are encoded to a bitstream.

Abstract: Systems and methods of restricting storage of data received in content requests include a data processing system obtaining a public encryption key for a resource provider. The data processing system can receive a content request from an information resource of the resource provider running on a client device. The request can include one or more key values corresponding to one or more data keys. The data processing system can identify a data key using an encryption policy specific to the resource provider, and encrypt the key value using the public encryption key. The data processing system can store the key value in encrypted form. The data processing system can generate, using the key value in encrypted form, a data report, and provide access to the data report.

Abstract: In a graphics processing unit (GPU), receiving an input image comprising an array of pixels. Each pixel having a gray scale value from a range of N grayscale values. For each particular input patch of pixels of a set of input patches partitioning the input image and in parallel for each particular grayscale value the range, counting the number of pixels in the particular input patch having the particular grayscale value. In parallel for each particular input patch of pixels of a set of input patches partitioning the input image, creating an output image patch as an ordered sequence of N pixels, with the color value of the nth pixel in each corresponding output patch representing the count of pixels in the particular input patch having the nth grayscale value. Combining the output image patches into a single composite output image of N pixels, the pixel value of the nth pixel in the single composite output image corresponding to the count of pixels in the input image having the nth grayscale value.

Abstract: Provided are methods, systems, and apparatus for hierarchical decorrelation of multichannel audio. A hierarchical decorrelation algorithm is designed to adapt to possibly changing characteristics of an input signal, and also preserves the energy of the original signal. The algorithm is invertible in that the original signal can be retrieved if needed. Furthermore, the proposed algorithm decomposes the decorrelation process into multiple low-complexity steps. The contribution of these steps is generally in a decreasing order, and thus the complexity of the algorithm can be scaled.

Abstract: In one example, a method includes method comprising: receiving audio data generated by a microphone of a current computing device; identifying, based on the audio data, one or more computing devices that each emitted a respective audio signal in response to speech reception being activated at the current computing device; and selecting either the current computing device or a particular computing device from the identified one or more computing devices to satisfy a spoken utterance determined based on the audio data.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining a plurality of sibling search queries for an input search query. In one aspect, a method comprises: receiving an input search query that satisfies a context template comprising a sequence of one or more words and a wildcard, wherein a wildcard represents variable data, wherein the input search query satisfies the context template and comprises a target word sequence that corresponds to the wildcard in the context template; and determining a plurality of sibling search queries for the input search query, wherein each sibling search query satisfies the context template and comprises a sibling word sequence that corresponds to the wildcard in the context template.

Abstract: Systems and methods for analysis of execution patterns for applications executing on remote devices. In some implementations of the system, a knowledge base stores successful traces from a plurality of instances of an application and one or more computing processors in the system receive, via a network interface, call-stack information from an instance of the application executing on a remote device, call-stack information including periodic captures of an execution status for the instance of the application, and determine whether there is a similarity between the call-stack information received from the instance of the application and the stored plurality of successful traces. Responsive to determining a similarity, the computing processors add the remote device to a population of devices likely to execute the object and facilitate further actions specific to the device population.

Abstract: Techniques are described herein for automated assistants that proactively provide content to participant(s) of multi-participant message exchange threads (e.g., group chats, audio and/or video calls in which oral messages are transcribed for analysis, etc.) based on signals such as individual participant profiles associated with participant(s). In various implementations, automated assistant(s) that may not be explicitly invoked may analyze content of a message exchange thread involving multiple human participants and/or document(s) associated with the message exchange thread. Based on the analyzing, the automated assistant(s) may identify topic(s) pertinent to the message exchange thread. Based on individual participant profiles associated with the participants, the automated assistant(s) may identify shared interest(s) of the participants.

Abstract: Systems and methods of a gain adjustment component for content selection infrastructure are provided. The system can receive a selection of multiple topics identified by a semantic analysis technique, and identify one or more keywords. The system can determine relevance scores between each of the keywords and the multiple topics. The system can generate an aggregated relevance score for each keyword based on the relevance score for the keyword and each of the multiple topics. The system can determine a threshold based on a metric associated with the multiple topics. The system can determine to input or exclude each of the one or more keywords based on a comparison of the corresponding aggregated relevance score with the threshold.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for instantiating a dialog process at a particular child node state. In one aspect, a method includes receiving a request specifying a particular child node state of a dialog process, instantiating the dialog process at the root node state, providing input responses to the dialog process that cause the dialog process to traverse from the root node state to the particular child node state, where each input response is provided in response to an input request, and providing data that causes a user device to present the dialog process at the particular child node state.

Abstract: A system of multi-modal transmission of packetized data in a voice activated data packet based computer network environment is provided. A natural language processor component can parse an input audio signal to identify a request and a trigger keyword. Based on the input audio signal, a direct action application programming interface can generate a first action data structure, and a content selector component can select a content item. An interface management component can identify first and second candidate interfaces, and respective resource utilization values. The interface management component can select, based on the resource utilization values, the first candidate interface to present the content item. The interface management component can provide the first action data structure to the client computing device for rendering as audio output, and can transmit the content item converted for a first modality to deliver the content item for rendering from the selected interface.

Abstract: To control vehicular congestion at intersections, a method includes detecting a group of two or more vehicles stopped at an intersection regulated by a traffic light, where for each of the vehicles in the group, movement through the intersection in response to a change in a state of the traffic light is affected by movement of at least one other vehicle in the group. The method further includes determining a time when the state of the traffic light changes to allow the group of vehicles to start moving through the intersection, determining acceleration parameters for a certain vehicle in the group in view of potential and/or actual movement of the other vehicles in the group through the intersection, and providing guidance to a device operated by a driver of the certain vehicle regarding the determined acceleration parameters, in accordance with the determined time.

Abstract: Systems, methods, and computer readable media related to determining one or more responses to provide that are responsive to an electronic communication that is generated through interaction with a client computing device. For example, determining one or more responses to provide for presentation to a user as suggestions for inclusion in a reply to an electronic communication sent to the user. Some implementations are related to training and/or using separate input and response neural network models for determining responses for electronic communications. The input neural network model and the response neural network model can be separate, but trained and/or used cooperatively.

Abstract: A bundle of public counters and a corresponding bundle of private counters are created and transmitted to a user device. The user device receives a request and processes the request without accessing a secure element processor on the user device. The user device calculates a security code using the private counter and a number. The user device transmits the calculated security code and one of the bundle of public counters in response to the request. A receiver of the response to the request determines the validity of the public counter and looks up the corresponding private counter using the public counter. The receiver determines the validity of the security code by recomputing it using the private counter and the number.