Abstract: A computer-implemented technique identifies terms in an original reference transcription and original ASR output results that are considered valid variants of each other, even though these terms have different textual forms. Based on this finding, the technique produces a normalized reference transcription and normalized ASR output results in which valid variants are assigned the same textual form. In some implementations, the technique uses the normalized text to develop a model for an ASR system. For example, the technique may generate a word error rate (WER) measure by comparing the normalized reference transcription with the normalized ASR output results, and use the WER measure as guidance in developing the model. Some aspects of the technique involve identifying occasions in which a term can be properly split into component parts. Other aspects can identify other ways in which two terms may vary in spelling, but nonetheless remain valid variants.

Abstract: A training-stage technique trains a language model for use in an ASR system. The technique includes: obtaining a training corpus that includes a sequence of terms; determining that an original term in the training corpus is not present in a dictionary resource; segmenting the original term into two or more sub-terms using a segmentation resource; determining that the segmentation of the original term into the two or more sub-terms is a valid segmentation, based on two or more validity tests; and training the language model based on the terms that have been identified. A computer-implemented inference-stage technique applies the language model to produce ASR output results. The inference-stage technique merges a sub-term with a preceding term if these two terms are separated by no more than a prescribed interval of time.

Abstract: Device cohort management techniques are described that are usable to control resource utilization by the devices. This is performable by managing usage together through grouping the devices through membership in a cohort. As a result, interaction with resources by the various devices is coordinated across the cohort, thereby improving device operation and user efficiency in resource usage by the devices.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing miscalibration detection. One of the methods includes receiving sensor data from each of multiple sensors of a device in a system configured to provide augmented reality or mixed reality output to a user. Feature values are determined based on the sensor data for a predetermined set of features. The determined feature values are processed using a miscalibration detection model that has been trained, based on examples of captured sensor data from one or more devices, to predict whether a miscalibration condition of one or more of the multiple sensors has occurred. Based on the output of the miscalibration detection model, the system determines whether to initiate recalibration of extrinsic parameters for at least one of the multiple sensors or to bypass recalibration.

Abstract: Methods, systems, and apparatus for performing bundling adjustment using epipolar constraints. A method includes receiving image data from a headset for a particular pose. The image data includes a first image from a first camera of the headset and a second image from a second camera of the headset. The method includes identifying at least one key point in a three-dimensional model of an environment at least partly represented in the first image and the second image and performing bundle adjustment. Bundle adjustment is performed by jointly optimizing a reprojection error for the at least one key point and an epipolar error for the at least one key point. Results of the bundle adjustment are used to perform at least one of (i) updating the three-dimensional model, (ii) determining a position of the headset at the particular pose, or (iii) determining extrinsic parameters of the first camera and second camera.

Abstract: Embodiments herein disclose an OTP low power circuit and methods for providing bias voltages using a single regulator. The circuit includes a Bitcell, a diode drop, a charge pump, a combinational logic controller, a program current sink load, and a read current sink load. The Bitcell includes programmable word lines and read lines, and is configured to operate in either a programmable mode or a read mode. The diode drop is configured to provide a second bias voltage to drive the read lines and the single regulator is configured to provide a first bias voltage to drive the WP in the read mode. The charge pump is configured to provide a third bias voltage to drive the WP in the program mode.

Abstract: For electronic devices that have an electrically conductive surface across a gap from a battery enclosure, the described circuit may be used to detect an anomalous event. The circuit may connect the battery enclosure to ground and may include a switch that operably connects the electrically conductive surface to ground or, upon actuation of the switch, to a portion of the circuit including a resistor. In an anomaly detection mode, the switch is controlled to disconnect the surface from ground and to connect the surface to a resistance measurement channel and a resistor that is connected to the sub-circuit and ground. The system may determine an impedance across the resistor and may use the measured impedance to detect and/or classify an anomalous event.

Abstract: System and method for anonymizing logs generated in applications running in a computing environment detects log data being generated in an application and compares the log data to a set of predefined search pattern policies to find sensitive information contained in the log data. The sensitive information contained in the log data is converted into anonymous information to produce anonymized log data within the application. The anonymized log data is then written to a destination.

Abstract: Systems, methods, and non-transitory computer-readable media can determine a set of candidate values for a field in a page. The set of candidate values can be evaluated for accuracy based at least in part on a machine learning model, wherein the machine learning model outputs a respective score for each candidate value that measures an accuracy of the candidate value for the field in the page. A best scoring candidate value can be determined from the set of candidate values. The field in the page can be associated with the best scoring candidate value.

Abstract: Techniques for addressing deformations in a virtual or augmented headset described. In some implementations, cameras in a headset can obtain image data at different times as the headset moves through a series of poses of the headset. One or more miscalibration conditions for the headset that have occurred as the headset moved through the series of poses can be detected. The series of poses can be divided into groups of poses based on the one or more miscalibration conditions, and bundle adjustment for the groups of poses can be performed using a separate set of camera calibration data. The bundle adjustment for the poses in each group is performed using a same set of calibration data for the group. The camera calibration data for each group is estimated jointly with bundle adjustment estimation for the poses in the group.

Abstract: An apparatus controlling peak-to-peak ripple in an output voltage may include; a charge pump unit configured to generate the output voltage in response to an input supply voltage, and a low voltage detector configured to generate a control signal, wherein the control signal defines the level of the input supply voltage.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for camera calibration during bundle adjustment. One of the methods includes maintaining a three-dimensional model of an environment and a plurality of image data clusters that each include data generated from images captured by two or more cameras included in a device. The method includes jointly determining, for a three-dimensional point represented by an image data cluster (i) the newly estimated coordinates for the three-dimensional point for an update to the three-dimensional model or a trajectory of the device, and (ii) the newly estimated calibration data that represents the spatial relationship between the two or more cameras.

Abstract: Some embodiments provide a method for establishing secure connections between several services operating in an on-premises network and external devices operating in an external network. The method configures each service to communicate with a reverse proxy operating in the on-premises network. The reverse proxy establishes a secure connection with each service. The method programs the reverse proxy to communicate with a forward proxy that establishes communications with the external devices. In some embodiments, the forward proxy is part of the on-premises network, while in other embodiments the forward proxy is a cloud-based service. Through the reverse and forward proxies, the on-premises services communicate with the external devices securely. For instance, in some embodiments, the forward proxy hides internal network addresses (e.g., IP addresses) and domain names of on-premises services and allows administrators to configure network monitoring to monitor and block malicious activities.

Abstract: Multimodal disentanglement can include generating a set of silhouette images corresponding to a human face, the generating undoing a correlation between an upper portion and a lower portion of the human face depicted by each silhouette image. A unimodal machine learning model can be trained with the set of silhouette images. As trained, the unimodal machine learning model can generate synthetic images of the human face. The synthetic images generated by the unimodal machine learning model once trained can be used to train a multimodal rendering network. The multimodal rendering network can be trained to generate a voice-animated digital human. Training the multimodal rendering network can be based on minimizing differences between the synthetic images and images generated by the multimodal rendering network.

Abstract: A system and method for sharing secrets with virtual computing instances in a distributed system uses a time-to-live (TTL) address written in a virtual computing instance using a cluster management center that manages the virtual computing instance as part of a logical cluster of virtual computing instances. The secret information is retrieved when the TLL address is invoked. The secret information is used to execute an operation that requires the secret information.

Abstract: Image generation using a hierarchical, model-based system includes generating a first region of an image using a first neural network model. The first region of the image is provided to a second neural network model as input. A second region of the image is generated using the second neural network model. The second region of the image shares a boundary with at least a portion of the first region of the image.

Abstract: A system for managing power distribution among devices based on one or more applicable power policies. The system can receive state data, including power condition data representing power conditions associated with a plurality of devices including a first device. The system can determine, based on the power condition data, a second device to wirelessly provide power to the first device. The system can generate a power distribution command configured to cause wireless power transmission from the second device to the first device and send the power distribution command to at least one of the first device or the second device to initiate wireless power transmission from the second device to the first device.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for calibrating an augmented reality device using camera and inertial measurement unit data. In some implementations, a bundle adjustment process jointly optimizes or estimates states of the augmented reality device. The process can use, as input, visual and inertial measurements as well as factory-calibrated sensor extrinsic parameters. The process performs bundle adjustment and uses non-linear optimization of estimated states constrained by the measurements and the factory calibrated extrinsic parameters. The process can jointly optimize inertial constraints, IMU calibration, and camera calibrations. Output of the process can include most likely estimated states, such as data for a 3D map of an environment, a trajectory of the device, and/or updated extrinsic parameters of the visual and inertial sensors (e.g., cameras and IMUs).

Abstract: Systems and methods for contextual natural language censoring are disclosed. For example, configuration data indicating details associated with content provided by a client device and/or about the client device may be received and may be utilized to determine impermissible and permissible exceptions for a given client. One or more queries may be generated utilizing the impermissible and permissible exceptions, and when input data is received from the client device and/or in association with the client identifier, the queries may be utilized to evaluate the input data for impermissible and permissible exceptions. The results may be filtered based on user preferences, the input data may be censored, the input data may be prevented from being exposed to a user device, the application associated with the input data may be removed from availability, and/or a maturity setting may be changed for the application, for example.

Abstract: The present invention discloses a wordline driver circuit for a random-access memory (RAM), which can reduce leakage during power down mode. The circuit includes a pre-driver stage on header and footer. The pre-driver stage includes a strap buffer defining a header and comprising a first switch connecting a first set of wordlines to a first voltage. The pre-driver stage includes an input-output buffer defining a footer and comprising a second switch connecting a second set of wordlines to a second voltage. In the pre-driver stage, the strap buffer further includes a third switch connecting the second set of wordlines to the first voltage and a fourth switch connecting the first set of wordlines to the second voltage.