Abstract: A method includes executing an instruction to execute a query for a data block, the data block associated with a corresponding memory level of a logarithmic number of memory levels (li) of memory, each memory level (li) including physical memory (RAMi) residing on memory hardware of a distributed system. The method also includes retrieving a value associated with the data block from an oblivious hash table using a corresponding key, and extracting un-queried key value pairs from the oblivious hash table associated with un-queried data blocks after executing a threshold number of queries for data blocks. The method also includes a multi-array shuffle routine on the extracted key value pairs from the oblivious hash table to generate an output array containing the un-queried key value pairs.

Abstract: Methods, systems, and apparatus include computer programs encoded on a computer-readable storage medium, including a method for providing content. Co-located devices are identified, including first and second devices associated with first and second users. A determination is made that a device is configured to be viewable by multiple users based on device characteristics. Use data is gathered that is associated with current sessions associated with users related to their use of their respective devices, including first data related to the use of the first device and second data related to the use of the second device. The first and second data are aggregated and evaluated to identify a first content item to present on the first device. Second different content items are selected for display on other co-located devices based at least in part on the first content item.

Abstract: An example method includes obtaining user consent to collect and make use of personal information for providing behavioral coaching; obtaining contextual and fitness related information of a user; determining, by inputting the contextual and fitness related information into a model that defines a motivational state, a current motivational state of the user; determining, based at least in part on the current motivational state of the user, a type of information to output as part of the behavioral coaching, wherein the type of information is selected from a group comprising education information, inspirational information, and achievement information; determining, based on the type of information to output, a channel for outputting the type of information as part of the behavioral coaching; and outputting, by the computing device, via the channel, a notification including content of the type of information.

Abstract: Audiovisual presentations, such as film recordings, may have been originally created having an audio soundtrack with multiple audio tracks mixed for a surround sound system that includes a set of speakers physically surrounding a user. The present disclosure presents systems and methods to remix these soundtracks into 3D audio that when presented to the ears of a user can be perceived as a virtual surround sound system that mimics the physical system. What is more, the disclosed systems and methods can enhance the virtual surround sound system by adjusting virtual speakers of the virtual surround sound system according to video content of the audiovisual presentation. Further enhancement may be possible by adjusting the virtual speakers of the virtual surround sound system according to a sensed position of a user.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media for classification using neural networks. One method includes receiving audio data corresponding to an utterance. Obtaining a transcription of the utterance. Generating a representation of the audio data. Generating a representation of the transcription of the utterance. Providing (i) the representation of the audio data and (ii) the representation of the transcription of the utterance to a classifier that, based on a given representation of the audio data and a given representation of the transcription of the utterance, is trained to output an indication of whether the utterance associated with the given representation is likely directed to an automated assistance or is likely not directed to an automated assistant.

Abstract: Techniques and apparatuses are described for reducing a flicker effect of multiple light sources in an image captured with an imaging device. A lighting frequency associated with each of the multiple light sources is detected and prioritized relative to a flicker effect upon the image to identify at least a first-prioritized lighting frequency and a second-prioritized lighting frequency. A first exposure-time factorization set is determined for the first-prioritized lighting frequency, and a second exposure-time factorization set is determined for the second-prioritized lighting frequency. An exposure time of the imaging device is adjusted to an exposure time identified in the first exposure-time factorization set that matches, or aligns near-to-matching, an exposure time identified in the second exposure-time factorization set.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network using consistency measures. One of the methods includes processing a particular training example from a mediator training data set using a first neural network to generate a first output for a first machine learning task; processing the particular training example in the mediator training data set using each of one or more second neural networks, wherein each second neural network is configured to generate a second output for a respective second machine learning task; determining, for each second machine learning task, a consistency target output for the first machine learning task; determining, for each second machine learning task, an error between the first output and the consistency target output corresponding to the second machine learning task; and generating a parameter update for the first neural network from the determined errors.

Abstract: The various implementations described herein include a camera device that includes: (1) a housing; (2) an image sensor positioned within the housing and having a field of view corresponding to a scene in the smart home environment; (3) at least one infrared (IR) illuminator positioned with the housing, the IR illuminator configured to selectively illuminate the scene; and (4) a front face that is at least partially concave-shaped and coupled to the housing, the front face positioned in front of the image sensor such that light from the scene passes through the front face prior to entering the image sensor, and the front face positioned in front of the IR illuminator such that IR light from the IR illuminator is directed through the front face.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for providing conversion path performance measures and reports. In one aspect, user interaction data are obtained, were the user interaction data specify user interactions for a plurality of conversions. User interactions that are associated with each conversion are selected from the user interaction data, where the associated user interactions for each conversion are user interactions with a converting user during the conversion cycle for the conversion. Using the user interaction data for the selected user interactions, a quantity of user interactions that are associated with each conversion and occurred during the conversion cycle for the conversion are determined. In turn, conversion path performance measures are computed and reports specifying the conversion path performance measures are generated.

Abstract: A system and method for frame accurate editing of high resolution live video streams is disclosed. A method includes providing a low resolution video stream displayed in a first viewing portion of an edit user interface, wherein the low resolution video stream corresponds to a high resolution video stream of the live video content, receiving, via the viewing portion of the user interface, user input identifying a first start point and a first end point within the low resolution video stream, and transmitting the first start point and the first end point to a server to obtain a respective intermediate video segment of an intermediate resolution video stream corresponding to the high resolution video stream.

Abstract: A method (400) for accessing one or more service processes (222) of service (250) includes executing at least one service enclave (220) and executing an enclave sandbox (200) that wraps the at least one service enclave. The at least one service enclave provides an interface to the one or more service processes. The enclave sandbox is configured to establish an encrypted communication tunnel (210) to the at least one service enclave interfacing with the one or more service processes, and communicate program calls (302) to/from the one or more service processes as encrypted communications through the encrypted communication tunnel.

Abstract: A method includes receiving audio data encoding an utterance, processing, using a speech recognition model, the audio data to generate speech recognition scores for speech elements, and determining context scores for the speech elements based on context data indicating a context for the utterance. The method also includes executing, using the speech recognition scores and the context scores, a beam search decoding process to determine one or more candidate transcriptions for the utterance. The method also includes selecting a transcription for the utterance from the one or more candidate transcriptions.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for scheduling operations represented as a computational graph on a distributed computing network. A method includes: receiving data representing operations to be executed in order to perform a job on a plurality of hardware accelerators of a plurality of different accelerator types; generating, for the job and from at least the data representing the operations, features that represent a predicted performance for the job on hardware accelerators of the plurality of different accelerator types; generating, from the features, a respective predicted performance metric for the job for each of the plurality of different accelerator types according to a performance objective function; and providing, to a scheduling system, one or more recommendations for scheduling the job on one or more recommended types of hardware accelerators.

Abstract: A method (500) for demarcating carrier-mediated and user-mediated networks includes, for a mobile device (110) simultaneously connected to a first wireless network (120) and a second wireless network (120), determining whether the first wireless network is associated with a same network operator (70) as the second wireless network. When the operator is the same, the method includes establishing a virtual private network (VPN) (30) between the mobile device and a remote server (130). The VPN includes a first VPN connection (122) that connects the mobile device to the remote server over the first wireless network and a second VPN connection (122) that connects the mobile device to the remote server over the second wireless network. The method includes transmitting data (50) from the mobile device to the remote server over the VPN via one of the VPN connections. The remote server routes the data to a destination server (60).

Abstract: This document describes techniques and systems for radar-based gesture-recognition with context-sensitive gating and other context-sensitive controls. Sensor data from a proximity sensor and/or a movement sensor produces a context of a user equipment. The techniques and systems enable the user equipment to recognize contexts when a radar system can be unreliable and should not be used for gesture-recognition, enabling the user equipment to automatically disable or “gate” the output from the radar system according to context. The user equipment prevents the radar system from transitioning to a high-power state to perform gesture-recognition in contexts where radar data detected by the radar system is likely due to unintentional input. By so doing, the techniques conserve power, improve accuracy, or reduce latency relative to many common techniques and systems for radar-based gesture-recognition.

Abstract: Methods, and systems, including computer programs encoded on computer storage media for neural network architecture search.

Abstract: A method for provisioning private-cloud server nodes by receiving a request to provision a specified number of server nodes for a private cloud, wherein the request is associated with a user, identifying a plurality of server nodes including (a) the specified number of hypervisor server nodes from a first pool that includes prepared hypervisor server nodes, each of which includes a previously-installed hypervisor, and (b) a management server node from a second pool that includes prepared management server nodes, each of which includes a previously-installed hypervisor and one or more previously-installed management components, configuring the identified server nodes to use a network associated with the user, creating a private cloud that includes the identified server nodes, and providing, to the user, permission to access the identified server nodes.

Abstract: A method (400) for representing an intended prosody in synthesized speech includes receiving a text utterance (310) having at least one word (240), and selecting an utterance embedding (204) for the text utterance. Each word in the text utterance has at least one syllable (230) and each syllable has at least one phoneme (220). The utterance embedding represents an intended prosody. For each syllable, using the selected utterance embedding, the method also includes: predicting a duration (238) of the syllable by decoding a prosodic syllable embedding (232, 234) for the syllable based on attention by an attention mechanism (340) to linguistic features (222) of each phoneme of the syllable and generating a plurality of fixed-length predicted frames (260) based on the predicted duration for the syllable.

Abstract: Various arrangements for sonar-based vital sign monitoring are presented herein. A mobile device may be determined to be stationary. In response, sonar-based movement sensing can be activated. Sonar data can then be captured in response to activating the sonar-based movement sensing. A breathing pattern can be detected in the sonar data and used to collect respiration data about a user.

Abstract: An auxiliary device charging case is used to facilitate translation features of a mobile computing device or auxiliary device. A first user, who may be a foreign language speaker, holds the charging case and speaks into the charging case. The charging case communicates the received speech to the mobile computing device, either directly or through the auxiliary device, which translates the received speech into a second language for a second user, who is the owner of the mobile computing device and auxiliary device. The second user may provide input in the second language, such as by speaking or typing into the auxiliary or mobile computing device. The mobile computing device may translate this second input to the first language, and transmit the translated input to the charging case either directly or through the auxiliary device. The charging case may output the translated second input to the first user, such as through a speaker or display screen.