Abstract: Systems and methods for training a machine-learned model are provided. A method can include can include obtaining an unlabeled audio signal, sampling the unlabeled audio signal to select one or more sampled slices, inputting the one or more sampled slices into a machine-learned model, receiving, as an output of the machine-learned model, one or more determined characteristics associated with the audio signal, determining a loss function for the machine-learned model based at least in part on a difference between the one or more determined characteristics and one or more corresponding ground truth characteristics of the audio signal, and training the machine-learned model from end to end based at least in part on the loss function. The one or more determined characteristics can include one or more reconstructed portions of the audio signal temporally adjacent to the one or more sampled slices or an estimated distance between two sampled slices.

Abstract: Techniques are described herein for providing smart suggestions for image zoom regions. A method includes: receiving a search query; performing a search using the search query to identify search results that include image search results including a plurality of images that are responsive to the search query; for a given image of the plurality of images included in the image search results, determining at least one zoom region in the given image; and providing the search results including the image search results, including providing the given image and an indication of the at least one zoom region in the given image.

Abstract: Implementations can identify a given assistant device from among a plurality of assistant devices in an ecosystem, obtain device-specific signal(s) that are generated by the given assistant device, process the device-specific signal(s) to generate candidate semantic label(s) for the given assistant device, select a given semantic label for the given semantic device from among the candidate semantic label(s), and assigning, in a device topology representation of the ecosystem, the given semantic label to the given assistant device. Implementations can optionally receive a spoken utterance that includes a query or command at the assistant device(s), determine a semantic property of the query or command matches the given semantic label to the given assistant device, and cause the given assistant device to satisfy the query or command.

Abstract: Techniques are disclosed that enable determining and/or utilizing a misrecognition of a spoken utterance, where the misrecognition is generated using an automatic speech recognition (ASR) model. Various implementations include determining a recognition based on the spoken utterance and a previous utterance spoken prior to the spoken utterance. Additionally or alternatively, implementations include personalizing an ASR engine for a user based on the spoken utterance and the previous utterance spoken prior to the spoken utterance (e.g., based on audio data capturing the previous utterance and a text representation of the spoken utterance).

Abstract: An example method, apparatus, and computer-readable storage medium are provided to predict high-dynamic range (HDR) lighting from low-dynamic range (LDR) background images. In an example implementation, a method may include receiving low-dynamic range (LDR) background images of scenes, each LDR background image captured with appearance of one or more reference objects with different reflectance properties; and training a lighting estimation model based at least on the received LDR background images to predict high-dynamic range (HDR) lighting based at least on the trained model. In another example implementation, a method may include capturing a low-dynamic range (LDR) background image of a scene from an LDR video captured by a camera of the electronic computing device; predicting high-dynamic range (HDR) lighting for the image, the predicting, using a trained model, based at least on the LDR background image; and rendering a virtual object based at least on the predicted HDR lighting.

Abstract: Alpha channel post processing in image coding can include decoding, from multiple color channels of a bitstream, color channel values for an encoded image, decoding, from an alpha channel of the bitstream, alpha channel values for the encoded image, determining a bilateral filter based on a level of compression for encoding the alpha channel, post processing the alpha channel values by filtering the alpha channel values using the bilateral filter to obtain filtered alpha channel values, and generating at least a portion of a reconstructed image corresponding to the encoded image using the filtered alpha channel values and the color channel values.

Abstract: A method, in a user device configured to communicate with a base station, for managing communication of a segmented radio resource control (RRC) message that includes N segments includes transmitting (222) a first M segments of the segmented RRC message to the base station, M being an integer greater than zero and less than N, detecting (230 or 330), by processing hardware of the user device and before transmitting an (M+1)-th segment of the segmented RRC message, an intervening event, that triggers an RRC procedure, and, after detecting the intervening event, transmitting (260) the (M+1)-th segment through an N-th segment of the segmented RRC message to the base station before the RRC procedure has completed.

Abstract: Methods and apparatus related to associating a task with a user based on the user selecting a task suggestion that is provided to the user in response to a user query. In some implementations, the task may be identified based on similarities between the words and/or phrases of the user query and a task suggestion that is associated with a task. In some implementations, the task may be identified based on user data associated with the user. In some implementations, the task may be associated with additional information related to completing the task.

Abstract: A head mounted display includes a combiner configured to receive display light from a micro-display. The world-facing surface of the combiner has a curvature that corresponds to a user's vision correction prescription. The head mounted display also includes a corrective layer having a second curvature that corresponds to the user's vision correction prescription. The corrective layer is disposed on the eye-facing surface of the combiner such that the focal point of the display light is adjusted for the specific user as the display light exits the combiner towards the user's eye.

Abstract: 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.

Abstract: One or more servers receive a natural language query from a client device associated with a user. The one or more servers classify the natural language query as a query that seeks information previously accessed by the user. The one or more servers then obtain a response to the natural language query from one or more collections of documents, wherein each document in the one or more collections of documents was previously accessed by the user. The one or more servers generate search results based on the response. Then, the one or more servers communicate the search results to the client device.

Abstract: A computer-implemented method for expanding a set of matched nodes in a partially-matched graph can include obtaining, by a computing system, a partially-matched graph having a matching set, the partially-matched graph including one or more edges and a plurality of nodes, the one or more edges having a matching label. The method can include obtaining at least two unmatched nodes. The method can include determining an alternating path from a first unmatched node of the at least two unmatched nodes to a second unmatched node of the at least two unmatched nodes, the alternating path including at least one edge of the one or more edges. The method can include inverting the matching label of the at least one edge of the alternating path such that the at least two unmatched nodes are included in the matching set of the partially-matched graph.

Abstract: Text independent speaker recognition models can be utilized by an automated assistant to verify a particular user spoke a spoken utterance and/or to identify the user who spoke a spoken utterance. Implementations can include automatically updating a speaker embedding for a particular user based on previous utterances by the particular user. Additionally or alternatively, implementations can include verifying a particular user spoke a spoken utterance using output generated by both a text independent speaker recognition model as well as a text dependent speaker recognition model. Furthermore, implementations can additionally or alternatively include prefetching content for several users associated with a spoken utterance prior to determining which user spoke the spoken utterance.

Abstract: Methods, apparatus, and computer-readable media for determining and utilizing corrections to robot actions. Some implementations are directed to updating a local features model of a robot in response to determining a human correction of an action performed by the robot. The local features model is used to determine, based on an embedding generated over a corresponding neural network model, one or more features that are most similar to the generated embedding. Updating the local features model in response to a human correction can include updating a feature embedding, of the local features model, that corresponds to the human correction. Adjustment(s) to the features model can immediately improve robot performance without necessitating retraining of the corresponding neural network model.

Abstract: Systems and methods are disclosed herein for improved per-frequency counting systems that record interactions between individuals and a group of providers while maintaining differential privacy. A protocol may be defined that specifies frequency bins to categorize identifiers corresponding to individuals. A provider may generate a plurality of private sketches, each corresponding to a plurality of frequencies defined in the protocol. Frequency data is determined for each identifier. Identifiers are encoded into the private sketches corresponding to the identifiers' associated frequency. The plurality of private sketches from each provider in the group of providers are combined to generate a deduplicated distribution across the group. In one implementation, the private sketches of each provider are sequentially merged until all sketches have been combined, from which the total distribution can be estimated.

Abstract: Systems and techniques for persistent calibration of an electronic device configured to implement an extended reality (XR) system involve estimating and validating visual-inertial odometry (VIO) calibration parameters during an active XR session of the electronic device. Validating the estimated VIO calibration parameters involves performing a strict calibration qualification of the estimated VIO calibration parameters using a thresholding module, machine learning module, or both. An initial calibration qualification is performed concurrently with the strict calibration qualification based on VIO performance. If the estimated VIO calibration parameters pass the strict calibration qualification and initial calibration qualification, they are stored for use to calibrate the device in future XR sessions. Persistent calibration of time alignment between the inertial management unit and the image sensor of the electronic device is also performed during active XR sessions upon detection of time alignment issues.

Abstract: Systems and methods are provided for receiving at least one image and a reference image, and performing a plurality of downscaling operations having separable convolutions on the received at least one image. A plurality of residual blocks may be formed, with each residual block containing two separable convolutions of the kernel and two instance normalizations. A plurality of upscaling operations may be performed on the plurality of residual blocks, and a stylized image may be displayed based on at least the performed plurality of upscaling operations and the reference image.

Abstract: A user equipment (UE) employing different radio access technologies (RATs) concurrently provides the UE the opportunity to connect with different RAT-based base stations and concurrently transmit data thereto. A power-sharing control mechanism provides for sharing and allocating transmit power to multiple active RATs at the UE based on a priority designation of the data type associated with transmissions scheduled for each of the multiple active RATs. The power-sharing control mechanism provides efficient transmit power sharing between multiple transmit active RATs such that allocation of power to one RAT does not adversely affect the performance or coverage of the remaining RATs.

Abstract: Implementations described herein relate to determining how to fulfill a spoken utterance based on a user that provided the spoken utterance. For example, implementations can receive a spoken utterance from a user, determine a set of fulfillment actions for the spoken utterance, and determine whether the user that provided the spoken utterance corresponds to a first user or a second user. Further, and in response to determining that the user corresponds to the first user, implementations can select a subset of first fulfillment action(s) from the set, and cause the subset of first fulfillment action(s) to be implemented to satisfy the spoken utterance. Moreover, and in response to determining that the user corresponds to the second user, implementations can select a subset of distinct, second fulfillment action(s) from the set, and cause the subset of second fulfillment action(s) to be implemented to satisfy the spoken utterance.

Abstract: 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.