Abstract: To detect a change in lane marker geometry, a computing device receives map patch data from one or more vehicles within a geographic area indicative of measured lane marker positions. For each data point within the map patch data, the computing device determines an error metric based on one or more differences between one or more measured lane marker positions for the data point and one or more reference lane marker positions from a database. The computing device identifies relationships between the data points according to the error metric for each data point, a time when each data point was collected, or a location of each data point, and identifies a lane marker change event occurring within the geographic area based on the relationships between the data points. Then the computing device discards reference lane marker positions from the database corresponding to the lane marker change event.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for classifying data objects. One of the methods includes obtaining data that associates each term in a vocabulary of terms with a respective high-dimensional representation of the term; obtaining classification data for a data object, wherein the classification data includes a respective score for each of a plurality of categories, and wherein each of the categories is associated with a respective category label; computing an aggregate high-dimensional representation for the data object from high-dimensional representations for the category labels associated with the categories and the respective scores; identifying a first term in the vocabulary of terms having a high-dimensional representation that is closest to the aggregate high-dimensional representation; and selecting the first term as a category label for the data object.

Abstract: The present disclosure provides methods, systems, and devices for non-invasively measuring a level of an analyte in a user using Stimulated Raman Scattering. A device can include a Raman pump source that emits pump light toward a tissue at a pump wavelength and a Stokes source that emits Stokes light toward the tissue at one or more Stokes wavelengths. The device can further include one or more mirrors and lenses. The device can further include a beam controller that controls the one or more mirrors and the one or more lenses to target the pump light emitted by the Raman pump source and the Stokes light emitted by the Stokes source at an optimal target location. The device can further include a photodetector that measures light that emanates from the tissue. The device can further include a processor that estimates the level of the analyte in the user.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for detecting hotwords using a server. One of the methods includes receiving an audio signal encoding one or more utterances including a first utterance; determining whether at least a portion of the first utterance satisfies a first threshold of being at least a portion of a key phrase; in response to determining that at least the portion of the first utterance satisfies the first threshold of being at least a portion of a key phrase, sending the audio signal to a server system that determines whether the first utterance satisfies a second threshold of being the key phrase, the second threshold being more restrictive than the first threshold; and receiving tagged text data representing the one or more utterances encoded in the audio signal when the server system determines that the first utterance satisfies the second threshold.

Abstract: A method includes receiving a request indication indicating a GUI interaction by a user on a user device, and in response, providing to the device a response configured to cause the device to display, within a GUI, a structured prompt including a plurality of user input fields, each user input field representing a corresponding training sample and including a first corresponding text input field for capturing input text to be provided to an ML model, and a second corresponding text input field for capturing ground-truth output text. The method also includes receiving, from the device, the training samples, and, in response, adjusting the ML model using the training samples. The method further includes receiving, from the device, test input text, generating, using the adjusted ML model, test output text based on the test input text, and providing, to the user device, the test output text for display within the GUI.

Abstract: A method for training a machine learning model included generating training data for the machine learning model. Generating the training data includes generating first training input that includes candidate text portions of one or more electronic documents and generating a first target output for the first training input. The first target output identifies a formatting type for each of the candidate text portions. The training data is provided to train the machine learning model on (i) a set of training inputs including the first training input, and (ii) a set of target outputs including the first target output.

Abstract: The technology is generally directed to determining whether a consumer being presented with a digital component is a disengaged and/or qualifying disengaged consumer at the time the digital component is being selected. An artificial intelligence model may be trained and used to predict whether the consumer is a disengaged and/or qualifying disengaged consumer. The prediction may be a probability representing the likelihood that the user is a disengaged and/or qualifying disengaged consumer for a given merchant. The probability may be used to dynamically adjust the merchant's bid at the time of auction to have the merchant's digital component selected. The probability may, in some examples, may be used to dynamically adjust the conversion values after the merchant's digital component is provided for output.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for the selection, provision and display of one or more digital components during display of content. Methods can include identifying a plurality of digital components that can be presented on the client device. A maximum number of digital components that can be presented in a slot of a content and the time duration of the slot is determined. For each digital component a score is generated based on the duration, a position requirement and the number of times the digital component is available for provision within the slot is generated. A first set of digital component is selected based on the scores and provided to the client device.

Abstract: The present document describes techniques for a magnetically self-shielded speaker box for a mobile device. The speaker box described herein includes a thin metal enclosure, thereby reducing or eliminating plastic components from the exterior enclosure of the speaker box. The metal of the enclosure has a high magnetic permeability to prevent penetration of a magnetic field generated by the speaker transducer within the exterior enclosure. Also, the exterior enclosure includes at least two enclosure portions that are joined together with a continuous welded seam. Accordingly, the speaker box described herein is a thin-walled, fully metal, high-permeability enclosure resulting in a smaller, self-shielded solution that reduces both (i) the total physical volume of the speaker box for a given internal speaker back volume and (ii) the spacing requirements to adjacent voice-coil-motor auto-focus camera modules.

Abstract: Generally disclosed herein is a mechanism to manage the thermal control of servers equipped in a data center based on the measurements of the power consumption of each server and each thermally critical component in the server system. A controller may monitor the power consumption of one or more components or one or more servers and trigger responsive cooling action by one or more cooling fans based on changes in the power consumption. In this regard, the cooling fans are proactively activated prior to temperatures reaching high levels, thereby efficiently maintaining safe operating temperatures in the data center for the computing devices and/or for occupational safety.

Abstract: A method includes obtaining a video having image frames, and determining, for each respective image frame, a corresponding frame content score based on a visual content thereof. The method also includes selecting, from the image frames, a plurality of video clips, where each respective video clip includes a plurality of consecutive image frames and is selected based on the corresponding frame content scores of the plurality of consecutive image frames thereof. The method additionally includes determining, for each respective video clip, a corresponding clip content score based on the corresponding frame content scores of the plurality of consecutive image frames thereof. The method further includes generating a highlight video that includes the plurality of video clips and a representation of a ranking of the plurality of video clips, where the ranking is based on the corresponding clip content score of each respective video clip.

Abstract: A system for repair of an accelerator architecture including a plurality of accelerator host having a plurality of accelerator elements in each host and high-speed interconnects between accelerator elements and including a link health monitor service in the host determining a status of each interconnect internal and external to the host. The link health monitor service detects link status of the interconnects and communicates this to a pod manager responsible for managing a number of interconnected hosts. The pod manager communicates with a scheduler for scheduling processing tasks to each accelerator element or group of elements. Upon detection of an interconnect problem, the pod manager flags a least common ancestor in a binary tree of elements containing elements adjacent to the failed interconnect and flags the ancestor and its parent nodes as unavailable. Pod manager further generates a corrective action repairing the problem interconnect communicating the action to a technician.

Abstract: A method includes obtaining a pre-trained machine learning model and a training embedding snapshot from a remote system, and obtaining one or more input data samples captured by a user device. The method includes, for each particular input data sample of the one or more input data samples, processing, using an on-device machine learning model corresponding to the pre-trained machine learning model, the particular input data sample to generate a corresponding on-device embedding and one or more corresponding predicted outputs, and generating, using the training embedding snapshot and the corresponding on-device embedding, corresponding performance data. The method includes aggregating the corresponding performance data for the one or more input data samples to determine one or more performance metrics for the on-device machine learning model, and transmitting the one or more performance metrics to the remote system.

Abstract: A method executed on data processing hardware and including displaying, on a display in communication with the data processing hardware, a user interface for a chat session between a user and a chatbot, the user interface including a plurality of entries input by the user and a plurality of corresponding responses from the chatbot. The method also includes receiving, in the user interface, a user input indication indicating selection of a particular entry of the plurality of entries, the particular entry being other than a last entry of the plurality of entries. The method further includes receiving, in the user interface, a new entry input by the user for an additional entry associated with the particular entry. The method additionally includes based on the additional entry associated with the particular entry, causing the chatbot to update one or more of the plurality of corresponding responses from the chatbot.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for speeding up document loading. In some implementations, a resource of a document is requested from a first source, and metadata for the document is requested from a second source that is different from the first source. The requested metadata is received from the second source, and the requested resource is received from the first source. A first representation of the document based on the received metadata is provided for display. After providing the first representation, a second representation of the document that combines portions of the first representation with additional portions of the document is generated, and the second representation is provided for display.

Abstract: Timestamp synchronization may be performed by determining a time offset between a first virtual machine having a first virtual clock counter of a computing system and a second virtual machine having a second virtual clock counter of the computing system; and adjusting timestamps of traces with timestamps based on the second virtual clock counter according to the time offset.

Abstract: Implementations relate to ascertaining to what extent predictions, generated using a machine learning model, can be effectively reconstructed from model updates, where the model updates are generated based on those predictions and based on applying a particular loss technique (e.g., a particular cross-entropy loss technique). Some implementations disclosed generate measures that each indicate a degree of conformity between a corresponding reconstruction, generated using a corresponding model update, and a corresponding prediction. In some of those implementations, the measures are utilized in determining whether to utilize the particular loss technique (utilized in generating the model updates) in federated learning of the machine learning model and/or of additional machine learning model(s).

Abstract: A first base station communicates with a UE via the first base station and a second base station. The first base station transmits, via a radio interface to the second base station, a configuration for concurrent communication between the UE and a group of base stations including the first base station and the second base station. The first base station communicates, by processing hardware, data over the radio interface directly with the UE, and via the radio interface and the second base station.

Abstract: Implementations process, using a large language model, a free-form natural language (NL) instruction to generate to generate LLM output. Those implementations generate, based on the LLM output and a NL skill description of a robotic skill, a task-grounding measure that reflects a probability of the skill description in the probability distribution of the LLM output. Those implementations further generate, based on the robotic skill and current environmental state data, a world-grounding measure that reflects a probability of the robotic skill being successful based on the current environmental state data. Those implementations further determine, based on both the task-grounding measure and the world-grounding measure, whether to implement the robotic skill.

Abstract: The present disclosure provides for Contextual Alerting Functions (CAFs). CAFs are an interface paradigm for synchronized device frameworks across multi-device ecosystems and user data models. CAFs replace existing concepts such as apps and/or notifications.

Abstract: Techniques and apparatuses are described for implementing multifaceted thermal-response shaping. In example aspects, a thermal control system manages heat that is generated by a subsystem of a system-on-chip to protect the subsystem from being damaged due to overheating, to maintain reliability of the system-on-chip, and to avoid creating a potentially unsafe situation for the user to operate a computing device with the system-on-chip. The thermal control system triggers the subsystem to operate at different operation points to shape a thermal response of the system-on-chip. The thermal control system appropriately shapes the thermal response based on an evaluation of two or more metrics, such as temperature and power, which can be associated with similar or different time scales. With this flexible, multifaceted approach, the thermal control system can determine an operation point that can improve the user experience and does not compromise a safety or reliability of the subsystem.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for time-limited key derivation. In some implementations, a module receives a request that provides a key identifier and obtain a control value indicating a future time. The module obtains a counter value indicating a first time, where the counter value is based on a current state of a counter, and where the counter is configured to update the state of the counter at a predetermined frequency. The module generates a comparison result based on comparing the control value indicating the future time with the counter value indicating the first time. The module generates a key based on the key identifier, the control value, the comparison result, and a stored random number, and the module provides the key in response to the request.

Abstract: The present document describes techniques for use in troubleshooting and debugging complex circuits, devices, and systems. These techniques may be implemented to provide hardware-assisted instruction-level debugging. By way of an example, techniques are provided that may be implemented in an apparatus having a plurality of state sequencers. The techniques may include receiving signals from a plurality of state sequencers, and storing an instruction history indicated, at least in part, by one or more of the signals received from a selected one of the state sequencers. The stored instruction history may indicate at least a state of the selected state sequencer during a specific cycle of instruction execution.

Abstract: A method includes receiving image data characterizing a face of an individual and receiving a voice-based command captured by a microphone of a wearable device worn by a user. The voice-based command includes a natural language command spoken by the user that requests the wearable device to identify the individual. The method also includes performing person identification to identify the individual by extracting an evaluation vector from the image data characterizing the face of the individual, determining the evaluation vector matches a reference vector for the individual, and determining an identify of the individual. The method also includes providing, for output from the wearable device, an identification cue that conveys an identity of the induvial to the user.

Abstract: Techniques and apparatuses are described for implementing dynamic thermal control for a system-on-chip. In example aspects, a thermal control system of a system-on-chip is a passive system capable of managing heat that is generated by a subsystem of the system-on-chip. The thermal control system dynamically triggers the subsystem to operate at different operation points to address thermal control requirements of the system-on-chip. In more detail, the thermal control system provides flexible and adaptive thermal control policy enforcement by adjusting a temperature threshold and/or feedback signal dynamics (e.g., a feedback time scale). The temperature threshold can be set based on a predicted operation of the subsystem or based on a current operation of the subsystem relative to one or more safety or performance limitations. Through dynamic thermal control, the thermal control system can enforce different throttling limits while the subsystem is operating under different workloads.

Abstract: A method includes capturing, by a content generation component of a computing device, initial content comprising video, and audio associated with the video; identifying one or more audio clips in the audio associated with the video based on one or more transient points in the audio; extracting, for each audio clip, a corresponding video clip from the video of the initial content; providing a control interface to enable a user-generated sequence of audio clips, wherein each audio clip in the sequence of audio clips is selected from the one or more identified audio clips; generating new audiovisual content comprising a sequence of video clips to correspond to the user-generated sequence of audio clips, wherein each video clip in the sequence of video clips is the extracted corresponding video clip for each audio clip in the user-generated sequence of audio clips; and providing, by the control interface, the new audiovisual content.

Abstract: Implementations set forth herein relate to conditionally delaying fulfillment of client update requests in order to preserve network bandwidth and other resources that may be consumed when an ecosystem of linked assistant devices are repeatedly pinging servers for updates. In some implementations, a server device can delay and/or bypass fulfillment of a client request based on one or more indications that certain requested data is currently, or is expected to be, expired. For example, a user that is modifying assistant settings via a cellular device can cause an update notification to be pushed to several other assistant devices before the user finishes editing the assistant settings. Implementations herein can limit fulfillment of update requests from the client devices according to certain criteria-such as whether the user is continuing to modify the assistant settings from their cellular device.

Abstract: An optical component includes a waveguide combiner that has one or more optical components composed of a polymer and at least partially formed via a first casting process, a transparent body formed via a second casting process, the transparent body composed of a polymer and having a first working surface and an opposing second working surface. One or more optical components are disposed at a first surface of the transparent body. A thin film layer is disposed at least partially on one of the one or more optical components. A first layer of resin material overlays the one or more optical components and the first working surface of the transparent body.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining watch time loss regions in media content items. In one aspect, features for a video are input into a trained model that is trained to output watch time loss regions. The trained model is trained using labels corresponding to known watch time loss regions in training videos and features of training videos that correspond to the known watch time loss regions. A watch time loss region defines a time window of a video during which a likelihood of a user stopping playback of the video is more than a threshold likelihood. In response to inputting the feature for the first video into the trained model, data regarding watch time loss regions for the video is obtained from the model and provided to an entity involved in providing the video to a user.

Abstract: Systems and methods of the present disclosure are directed to a computer-implemented method for multimodal search refinement. The method includes obtaining a visual search query from a user comprising one or more query images. The method includes providing a search interface for display to the user, the search interface comprising one or more result images responsive to the one or more query images and an interface element indicative of a request to the user to refine the visual search query. The method includes obtaining, from the user, textual data comprising a refinement to the visual search query. The method includes appending, by the computing system, the textual data to the visual search query to obtain a multimodal search query.

Abstract: A web browsing application can implement a vector database to store session data. A web browsing application can automatically embed loaded web content into an embedded data store that maintains session data for a number of web sessions. A user can query the web browser using simple instructions. The web browser can interpret the instructions and use the embedded data store to quickly search across multiple modalities of embedded data to retrieve relevant results. The web browser can use machine-learned models to answer queries or perform other tasks by performing vector-based queries over the content of visited web data.

Abstract: Example embodiments of the present disclosure provide for an example method including maintaining a current version info list including version info tuples for software layers. The example method includes, upon receipt of a request for a registered version key, performing a comparison algorithm to authenticate a requested version info list including a number of version info tuples associated with software layers. The tuples can include a security version number (SVN) and a security context string for each software layer. The requested version info list can be authenticated using the comparison algorithm to determine that the requested version info list includes version info tuples with higher SVNs than the current version info list. Responsive to authenticating the requested version info list, the method include providing a portion of the requested version info list as input into a key derivation function (KDF) and obtaining a device requested version key as output.

Abstract: A computing system and method that can be used to surface a selectable action element for at least one ongoing task. In particular, the present disclosure provides a general pipeline to identify potential tasks that a user has an ongoing interest in or has not yet completed so that a suggestion of a content item can be made to further advance an identified user's task. This pipeline can incorporate probabilistic transition graphs, machine-learned models, and/or historical data to determine the relevance and completion of tasks that a user may desire to continue acting upon.

Abstract: An example method is provided. Unverified interaction data descriptive of interactions of a user with the computing device can be obtained. Feature values that embed characteristics of the unverified interaction data can be generated using a machine-learned embedding network of a machine-learned verification pipeline. A user account associated with the unverified interaction data can be determined using a verification model of the machine-learned verification pipeline.

Abstract: An object localization network (OLN) can be used to localize object(s) (e.g., known and/or unknown object(s)) in an instance of vision data. Various implementations include detecting the localized object(s) based on the localization. Many implementations include processing the instance of vision data using the OLN to generate a objectness score (e.g., a centerness score) as well as an intersection of union (IoU) score for one or more proposed object locations in the instance of vision data. Object(s) can be localized in the instance of vision data based on the objectness scores and the IoU scores.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for detecting utterances of a key phrase in an audio signal. One of the methods includes receiving, by a key phrase spotting system, an audio signal encoding one or more utterances; while continuing to receive the audio signal, generating, by the key phrase spotting system, an attention output using an attention mechanism that is configured to compute the attention output based on a series of encodings generated by an encoder comprising one or more neural network layers; generating, by the key phrase spotting system and using attention output, output that indicates whether the audio signal likely encodes the key phrase; and providing, by the key phrase spotting system, the output that indicates whether the audio signal likely encodes the key phrase.

Abstract: A method includes receiving an automated speech recognition (ASR) request from a user device that includes a speech input captured by the user device and content metadata associated with the speech input. The content metadata is generated by the user device. The method also includes determining a priority score for the ASR request based on the content metadata associated with the speech input and caching the ASR request in a pre-processing backlog of pending ASR requests each having a corresponding priority score. The pending ASR requests in the pre-processing backlog are ranked in order of the priority scores. The method also includes providing, from the pre-processing backlog, one or more of the pending ASR requests to a backend-side ASR module, wherein pending ASR requests associated with higher priority scores are processed before pending ASR requests associated with lower priority scores.

Abstract: The present disclosure is directed to object and/or character recognition for use in applications such as computer vision. Advantages of the present disclosure include lightweight functionality that can be used on devices such as smart phones. Aspects of the present disclosure include a sequential architecture where a lightweight machine-learned model can receive an image, detect whether an object is present in one or more regions of the image, and generate an output based on the detection. This output can be applied as a filter to remove image data that can be neglected for more memory intensive machine-learned models applied downstream.

Abstract: A method for searchable encryption with a public key includes receiving an operation request from a user device associated with a user requesting that encryption of data associated with the user. The data includes a corpus of documents stored on a remote storage device. The method also includes receiving a public key associated with the user. The public key includes an asymmetric cryptographic public key. The method also includes generating a random data key. The data key includes a symmetric cryptographic key. The method also includes encrypting, using the data key, a search index for the corpus of documents based on keywords within the corpus of documents. The method also includes encrypting, using the public key, the data key and sending the encrypted data key to a user device associated with the user.

Abstract: Features described herein pertain to extending the functionality of an application executing on an electronic device. A device trait specification associated with a smart device can be used to generate an application extension template for defining a layout of a graphical user interface page of the application and binding at least one device trait of the set of device traits to the application. An application extension can be generated based on the application extension template being populated with layout defining values and binding values. The application extension can be provided to the application. The application extension can cause the graphical user interface page of the application to be displayed on a display of the electronic device with the defined layout and elements bound to the layout and configured to cause the smart device to be controlled based on the device traits.

Abstract: Implementations relate to improving instruction following capabilities of large language models (LLMs) using instruction decomposition, self-evaluation, and optionally progressive refinement. Processor(s) of a system can: obtain natural language (NL) based input, generate a plurality of candidate responses and evaluate the candidate responses based on instructions included in the NL based input, using an LLM, and progressively refine the candidate responses until it is determined that one or more termination criteria are satisfied. In some implementations, the NL based input can be received from a client device. In these implementations, a given candidate response that is progressively refined can be rendered for presentation at the client device and responsive to the NL base input. In additional or alternative implementations, the NL based input can be obtained from database(s). In these implementations, a given candidate response that is progressively refined can be utilized in fine-tuning of the LLM.

Abstract: Implementations relate to leveraging large language model(s) (LLMs) and vision language model(s) (VLMs) to facilitate human-to-computer dialogs. In various implementations, one or more digital images may be processed using one or more VLMs to generate VLM output indicative of a state of an environment. An LLM prompt may be assembled based on the VLM output and a natural language input. The LLM prompt may be processed using one or more LLMs to generate content that is responsive to the natural language input. The content that is responsive to the natural language input may subsequently be rendered at one or more output devices.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing a classification task on a data item. In particular, a system classifies an input data item using key and value embeddings of memory data items.

Abstract: Implementations relate to reducing latency in generating and/or rendering a given stream of natural language (NL) based output generated using a large language model (LLM). Processor(s) of a system can: receive NL based input associated with a client device, generate the stream of NL based output utilizing the LLM that is responsive to the NL based input and that is for a given dialog context of an ongoing dialog, and cause the stream of NL based output to be rendered at the client device. Notably, the processor(s) can employ attribute classifier(s) and a multi-objective scorer to implement a blockwise controlled decoding technique in generating the stream of NL based output utilizing the LLM. By implementing the blockwise controlled decoding technique in generating the stream of NL based output utilizing the LLM, the processor(s) can reduce latency in generating and/or of the stream of NL based output generated utilizing the LLM.

Abstract: A system for dynamically controlling point-of-coherency or a point-of-serialization of shared data includes a plurality of processing engines grouped into a plurality of separate clusters and a shared communications path communicatively connecting each of the plurality of clusters to one another. Each respective cluster includes memory shared by the processing engines of the respective cluster, each unit of data in the memory being assigned to a single owner cluster responsible for maintaining an authoritative copy and a single manager cluster permanently responsible for assigning the owner cluster responsibility. Each respective cluster also includes a controller configured to receive data requests, track each of a manager status and an ownership status of the respective cluster, and control ownership status changes with respect to respective units of data based at least in part on the tracked ownership and manager statuses of the respective cluster.

Abstract: A method for recognizing persons in video streams includes obtaining a live video stream, detecting a first person in the live video stream, determining from analysis of the live video stream first information that identifies an attribute of the first person, determining based on at least some of the first information that the first person is not identifiable to the computing system, storing at least some of the first information, receiving a user classification of the first person as being a stranger, and deleting the stored first information.

Abstract: A method includes receiving a time series anomaly detection query from a user and training one or more models using a set of time series data values. For each respective time series data value in the set, the method includes determining, using the trained models, an expected data value for the respective time series data value and determining a difference between the expected data value and the respective time series data value. The method also includes determining that the difference between the expected data value and the respective time series data value satisfies a threshold. In response to determining that the difference between the expected data value and the respective time series data value satisfies the threshold, the method includes determining that the respective time series data value is anomalous and reporting the anomalous respective time series data value to the user.

Abstract: Systems and methods for a tunable digitally controlled oscillator. In one aspect, an apparatus includes one or more lattices of delay stages. Multiple, different paths may be selected through the lattices. Each path begins at a first delay stage with an input connected to a signal input node and ends at a second delay stage with an output connected to a signal output node.