Google LLC patent applications published on October 12th, 2023

Summary of the patent applications from Google LLC on October 12th, 2023

Google LLC has filed several recent patents related to various technologies. These patents cover areas such as power-saving modes for Wi-Fi chips, presenting information about events based on metadata, calculating differentially private histograms, dynamic voltage control for laser diodes in wearable display devices, composite lithium-ion batteries, automated speech recognition models, ensuring uniform brightness across AMOLED displays, instant motion tracking for augmented reality, verifying embedded digital content, and creating personalized references for generative models.

Notable applications:

• Wi-Fi chip power-saving mode: A Wi-Fi chip that operates in a power-saving mode by checking for periodic beacons at longer intervals and entering standby mode for a specific duration.
• Metadata-based event information presentation: Presenting information about an event based on user device information and content items obtained from connected sensors.
• (ε, δ)-differentially private histograms: Calculating private histograms over data from multiple users using a two-server model of secure multi-party computation.
• Dynamic voltage control for laser diodes: A wearable display device that adjusts the voltage level for laser diodes based on the frame to be displayed.
• Composite lithium-ion batteries: Batteries with special materials in the electrodes for improved performance, including faster charging, increased energy storage capacity, and higher energy density.
• Automated speech recognition model: A model consisting of encoders and decoders to generate speech recognition hypotheses.
• Uniform brightness across AMOLED displays: Techniques to maintain consistent voltage distribution across the display, even with a hole in the active area.
• Instant motion tracking for augmented reality: Analyzing sequential image frames to estimate rotation and translation for rendering virtual content.
• Verifying embedded digital content: An image processing system that analyzes pixels to determine if specific visual elements meet predetermined criteria.
• Personalized references for generative models: Using a set of images to create a reference for generating accurate images of a specific person.

These patents demonstrate Google LLC's focus on developing technologies related to power-saving, data privacy, display quality, augmented reality, and image processing. These innovations have the potential to enhance user experiences and improve the performance of various devices and systems.

Patent applications for Google LLC on October 12th, 2023

Modular and Collapsible Server Lift Assist for Immersion Cooling System (17714582)

Main Inventor

Jerry Chiu

Brief explanation

The abstract describes a gantry crane designed to be used with an immersion cooling system. The crane consists of rails that can be attached to the cooling tank. It also has a collapsible frame that can move along the rails. The crane includes a connector that can be connected to a computer hardware component that needs to be cooled in the immersion cooling system. Additionally, the crane has a winch mounted on the frame, which can be used to raise and lower the connector as needed.

Abstract

A gantry crane for attachment to an immersion cooling system includes rails attachable to a tank of the immersion cooling system. The gantry also includes a collapsible frame configured to travel along the rails. A connector is adapted for connecting to a computer hardware component intended to be cooled in the immersion cooling system, and the gantry includes a winch mounted to the frame and configured to raise and lower the connector.

Client-Server Approach for Indoor-Outdoor Detection of an Electronic Device, and Associated Systems and Methods (18040743)

Main Inventor

Mohammed Khider

Brief explanation

This document describes a method for detecting whether an electronic device is indoors or outdoors. The method involves a client-server approach where a server collects information from devices that detect access points. The electronic device scans for access points and collects sensor data from other sensors. This information is then transmitted to the server. The server uses crowdsourced information to determine the percentage of detections of each access point that are accompanied by a GPS signal and an RSS value below a certain threshold. This percentage and RSS value help determine the probability of whether the electronic device is located outdoors, lightly indoors, or deep indoors. This information can then be used to trigger specific functions on the electronic device.

Abstract

This document describes a client-server approach for indoor-outdoor detection of an electronic device, and associated systems and methods. A server () collects crowdsourced information () from devices that detected a plurality of access points. An electronic device (), performing a wire-less-network scan, detects access points () within range and detects sensor data () from other sensors (). The electronic device () transmits such information to the server (). The server accesses the crowdsourced information () to determine, per access point () detected in the scan, a percentage of total detections of the access point that are accompanied by a GPS signal of a device that detected the access point and an RSS value below which no such GPS signal accompanies the detections. The percentage and the RSS value enable a determination of a probability indicating whether the electronic device () is located outdoors, lightly indoors, or deep indoors, which enables the electronic device to trigger a corresponding function.

Systems And Methods For Detecting Mechanical Disturbances Using Underwater Optical Cables (18208498)

Main Inventor

Valey Kamalov

Brief explanation

This abstract describes a system and method for creating a model to detect seismic events. The system uses data from underwater optical stations, which record the polarization states of light signals over a period of time. The data is then transformed into frequency spectrums. The system also receives seismic activity data from the same region as the optical stations. Using this data, the system generates a model that can detect seismic events based on the spectrums and the seismic activity data.

Abstract

Systems and methods are provided for generating a model for detection of seismic events. In this regard, one or more processors may receive from one or more stations located along an underwater optical route, one or more time series of polarization states of a detected light signal during a time period. The one or more processors may transform the one or more time series of polarization states into one or more spectrums in a frequency domain. Seismic activity data for the time period may be received by the one or more processors, where the seismic activity data include one or more seismic events detected in a region at least partially overlapping the underwater optical route. The one or more processors then generate a model for detecting seismic events based on the one or more spectrums and the seismic activity data.

PASSIVE WORLD-REFERENCED EYE TRACKING FOR SMARTGLASSES (18297372)

Main Inventor

Daniel Adema

Brief explanation

The abstract describes improved systems for accurately measuring eye gaze in an augmented reality head-mounted wearable device. These systems use a radiation direction rerouter, such as a partial retroreflector, to adjust the angle of radiation in the device's waveguide. This adjustment ensures that the radiation is directed towards an outcoupler at an angle that is parallel to the user's gaze angle. By placing a partial retroreflector on the waveguide's surface opposite the outcoupler, an additional reflection is created. This reflection helps match the gaze angle of the user's eye to the angle of an image of the eye pupil captured by a world-facing radiation detector, such as a camera.

Abstract

Improved systems ensuring an accuracy of eye gaze measurements in an augmented reality head-mounted wearable device include a radiation direction rerouter (e.g., a partial retroreflector) configured to adjust an angle of incidence of the radiation in the waveguide at a first surface of a waveguide to produce radiation directed at an adjusted angle of incidence at an outcoupler such that the output direction is essentially parallel to the gaze angle of the user's eye. For example, a partial retroreflector may be disposed on a surface of the waveguide opposite the outcoupler provides an additional reflection so that the gaze angle of the user's eye is matched to an angle of an image of the eye pupil onto a world-facing radiation detector (e.g., a camera).

PROPORTIONAL FRAME DEFLECTION TO MAINTAIN GLASSES DISPLAY ALIGNMENT (18297394)

Main Inventor

Daniel Adema

Brief explanation

The abstract describes a technology that improves augmented reality smartglasses by ensuring that the displays are properly aligned. This is achieved by determining a relationship between the material properties of the frame at the shoulder and nose bridge. For example, this relationship could be related to rigidity or stiffness. The specified relationship ensures that the difference in bending stiffness between the two areas is within a certain tolerance, such as 5%, 1%, or even less than 1%.

Abstract

Improved augmented reality smartglasses ensure alignment of displays in a head-mounted wearable device such as AR smartglasses by specifying a relationship between material properties of the frame at the shoulder and nose bridge. For example, such a material relationship may be a rigidity and/or stiffness characteristic. In some implementations, the specified relationship is that a difference between the first bending stiffness and the second bending stiffness is less than a tolerance (e.g., 5%, 1%, or less than 1%).

TORQUE CONTROL MECHANISM FOR FRAME OF HEAD MOUNTED WEARABLE DEVICE (18192904)

Main Inventor

Daniel Adema

Brief explanation

The abstract describes systems and methods for controlling the amount of torque transmitted from the temple arm portions to the front frame portion of a head-mounted wearable device. The hinge mechanisms that connect the temple arm portions to the front frame portion include torque control devices that respond to rotation and deflection of the temple arm portions. These torque control devices help regulate the amount of torque transmitted to the front frame portion. Two types of torque control devices are mentioned: a magnetic torque control device and an elastomer torque control device. The magnetic torque control device consists of two arrays of magnets within the hinge mechanism that can be engaged or disengaged to provide compliance in multiple axes. The elastomer torque control device includes an engagement pin that selectively engages or disengages recesses in an elastomer layer within the hinge mechanism to provide compliance in multiple axes.

Abstract

Systems and methods for controlling an amount of torque transmitted from the temple arm portions to a front frame portion of a frame of a head mounted wearable device are provided. Hinge mechanisms that rotatably couple the temple arm portions to the front frame portion may include torque control devices that provide for torque control in response to rotation and/or deflection of the temple arm portions, to control an amount of torque that is transmitted to the front frame portion. A magnetic torque control device may include first and second arrays of magnets arranged within the hinge mechanism, and that are selectively engaged and disengaged to provide compliance in multiple axes. An elastomer torque control device may include an engagement pin that selectively engages and disengages recesses in an elastomer layer provided within the hinge mechanism to provide compliance in multiple axes.

Contextual Assistant Using Mouse Pointing or Touch Cues (18331643)

Main Inventor

Dongeek Shin

Brief explanation

This abstract describes a method for a contextual assistant to understand and respond to user queries based on audio input and spatial input on a screen. The assistant receives audio data of the user's query and a user input indication showing where the user interacted with the screen. It processes the audio to transcribe the query and interprets it to understand that it refers to an object on the screen without knowing which specific object. The assistant then requests information about the object and uses the user's spatial input to identify the object uniquely. It obtains the requested information and provides a response to the user's query.

Abstract

A method for a contextual assistant to use mouse pointing or touch cues includes receiving audio data corresponding to a query spoken by a user, receiving, in a graphical user interface displayed on a screen, a user input indication indicating a spatial input applied at a first location on the screen, and processing the audio data to determine a transcription of the query. The method also includes performing query interpretation on the transcription to determine that the query is referring to an object displayed on the screen without uniquely identifying the object, and requesting information about the object. The method further includes disambiguating, using the user input indication indicating the spatial input applied at the first location on the screen, the query to uniquely identify the object that the query is referring to, obtaining the information about the object requested by the query, and providing a response to the query.

TRANSLATING BETWEEN PROGRAMMING LANGUAGES INDEPENDENTLY OF SEQUENCE-TO-SEQUENCE DECODERS (17717609)

Main Inventor

Rishabh Singh

Brief explanation

This abstract describes techniques for translating a source code snippet from one programming language to another. Instead of using traditional sequence-to-sequence decoding methods, the approach utilizes an encoder portion of a transformer network to generate an embedding of the source code. This embedding is then processed using an all-pair attention layer to create an attended embedding. Finally, an output layer transforms the attended embedding into data that represents the translation of the source code into the second programming language.

Abstract

Techniques are described herein for translating a source code snippet from a first programming language to a second programming language independently of sequence-to-sequence decoding. In various implementations, the source code snippet written in the first programming language may be processed using an encoder portion of a transformer network to generate an embedding of the source code snippet. The embedding of the source code snippet may be processed using an all-pair attention layer to generate an attended embedding of the source code snippet. The attended embedding of the source code snippet may be processed using an output layer to generate, by way of a single transformation of the attended embedding of the source code snippet, data indicative of a translation of the source code snippet in the second programming language.

Methods for Optics Module Firmware In-Field Upgrade (18201365)

Main Inventor

Jianqiao Liu

Brief explanation

The abstract describes a method for upgrading the firmware of optics modules in a uniform and unified way. This ensures compatibility, security, code quality, and scalability. The method involves defining, receiving, and parsing an intermediate representation that includes vendor firmware upgrade operations and control logic. Read/write operations are then communicated to the optics modules based on this control logic. Additionally, a unified firmware upgrade framework is proposed, which includes multiple software layers. These layers translate, abstract, and make uniform the read/write operations, provide libraries of intermediate representations, package the representations into executables/scripts, monitor module status, determine when new firmware is released, and gradually upgrade the firmware of the optics modules.

Abstract

A uniform and unified firmware in-field upgrade capability for the optics modules may ensure compatibility, security and code quality, and scalability. In some examples, an intermediate representation, which includes vendor firmware upgrade operations and control logic, may be defined, received, and parsed. Read/write operations may be communicated to optical module(s) based on the control logic. In some examples, a unified optics module firmware in-field upgrade framework, which has multiple defined software layers, may ensure a uniform and unified approach to managing optics module(s) from different vendors and used by different projects. The software layers that may properly translate optics module read/write operations, abstract and make uniform the read/write operations, provide libraries of intermediate representations, package the intermediate representations into executables/scripts, monitor optics module status, determine when a new firmware is released, and gradually upgrade the optics module firmware.

DATA TRANSFERS FROM MEMORY TO MANAGE GRAPHICAL OUTPUT LATENCY (18208705)

Main Inventor

Jian Wei Leong

Brief explanation

This abstract describes a system and method for reducing latency in graphical output on a device with a display. The device receives an acoustic signal carrying a query and determines the state of a wireless controller. Based on this information, the device establishes an interaction mode for a graphical user interface. It sets a prefetch parameter to a certain value and prefetches electronic content items accordingly. Later, it may switch to a different interaction mode and adjust the prefetch parameter to a different value, prefetching a new set of electronic content items.

Abstract

Systems and methods of transferring data from memory to manage graphical output latency are provided. A device having a display receives an acoustic signal that carries a query. The device determines that a wireless controller is in a first state. The device establishes, based on receipt of the acoustic signal and the determination that the wireless controller device is in the first state, a first interaction mode for a graphical user interface rendered by the computing device for display via the display device. The device sets a prefetch parameter to a first value and prefetches the corresponding amount of electronic content items. The device establishes a second interaction mode and overrides the first value of the prefetch parameter to a second value, and prefetches a second amount of electronic content items corresponding to the second value.

Interface and Mode Selection for Digital Action Execution (18194854)

Main Inventor

Prithvi Balaram

Brief explanation

This abstract describes a system that provides an interface and mode selection for executing digital actions. The system analyzes the level of engagement between a client device and digital assistants to determine the appropriate interface. It then generates a digital interface with a call-to-action based on this analysis. When instructed to execute the call-to-action, the system determines the mode of execution and selects a digital assistant and another client device to carry out the action. The call-to-action is then transmitted to the second client device for execution.

Abstract

Interface and mode selection for digital action execution is provided. For example, a system loads a script library embedded in an electronic resource. The system determines a historic level of engagement between a client computing device and one or more digital assistants. The system selects, based on a first property of the client computing device and the historic level of engagement, a type of digital interface. The system generates, based on the type of digital interface, a digital interface with the call-to-action. The system determines, responsive to an instruction to execute the call-to-action, a mode of execution. The system selects a digital assistant and a second client device to execute the call-to-action. The system transmits the call-to-action to the second client device for execution.

Hot Growing A Cloud Hosted Block Device (18328751)

Main Inventor

Igor Belianski

Brief explanation

This abstract describes a method for resizing a block storage volume for a virtual machine. The method involves attaching a virtual storage device to the virtual machine, which exposes the block storage volume on memory hardware. The block storage volume initially has a certain storage capacity. The method then mounts the block storage volume to the virtual machine and resizes it while the virtual machine is still running. Importantly, this resizing is done without attaching any additional virtual storage devices or mounting additional block storage volumes to the virtual machine.

Abstract

A method of resizing a block storage volume for a virtual machine includes executing the virtual machine and attaching a virtual storage device to the virtual machine. The virtual storage device exposes the block storage volume on memory hardware to the virtual machine. The block storage volume includes a first storage capacity. The method also includes mounting the block storage volume to the virtual machine and resizing the block storage volume while the virtual machine continues to execute. The block storage volume is resized without attaching an additional virtual storage device to the virtual machine or mounting an additional block storage volume to the virtual machine.

General-Purpose Systolic Array (17703479)

Main Inventor

Reginald Clifford Young

Brief explanation

The abstract describes a systolic array cell, which is a component that includes two arithmetic logic units (ALUs) and a register-file. Multiple cells can be arranged in a matrix or array. The output of the first ALU in a cell is connected to the second cell on the right, and the output of the second ALU in the cell is connected to the third cell below it. This configuration allows for different instructions to be processed simultaneously in each cycle, thanks to the two ALUs in each cell.

Abstract

A systolic array cell is described, the cell including two general-purpose arithmetic logic units (ALUs) and register-file. A plurality of the cells may be configured in a matrix or array, such that the output of the first ALU in a first cell is provided to a second cell to the right of the first cell, and the output of the second ALU in the first cell is provided to a third cell below the first cell. The two ALUs in each cell of the array allow for processing of a different instruction in each cycle.

Online Migration From An Eventually Consistent System To A Strongly Consistent System (17716093)

Main Inventor

Lavina Jain

Brief explanation

The abstract describes a method for transferring data from one distributed system to another without locking the data. This is done by setting up a replication process between the two systems, where the first system initially acts as the primary system and receives write requests. Each write request is then asynchronously updated in the second system. Once the second system is caught up and has the same data as the first system, it can take over as the primary system and handle all future read and write requests directly.

Abstract

Generally disclosed herein is an approach to migrate data from a first type of distributed system to a second type of distributed system without locking data, where transactional dual writes are not available across the two systems. The approach starts by setting up a bi-directional replication between the first system and the second system. The first system will initially operate as a primary system, where the primary system receives and serves write requests from clients or other devices. For each write to the first system, the second system is updated with an asynchronous write. When the second system is caught up to the first system, such that both the first and second systems reflect approximately the same data, the second system can be switched over to serve as the primary system. The second system can now directly receive and serve all future read and write requests.

Selecting from Arrays of Multilingual Content (18022260)

Main Inventor

Prajakta Suresh Kalekar

Brief explanation

This abstract describes a system and method for selecting content to provide in networked environments. The system receives input from a client device, which includes keywords in a specific language. It determines the language based on these keywords and also identifies the location of the client device. Based on this location, it identifies a second language associated with it. Using the input, the system then identifies a set of content items in both the first and second languages. Finally, it provides a content item from either set of content items to the client device.

Abstract

Systems and methods of selecting content to provide in networked environments are provided herein. A data processing system can receive an input from a client device, the input including keywords in a first language. The data processing system can determine the first language based on the keywords of the input. The data processing system can determine, using the input, a location identifier identifying a location of the client device. The data processing system can identify a second language associated with the location identifier. The data processing system can identify a first plurality of content items in the first language and a second plurality of content items in the second language based on the input. The data processing system can provide, to the client device, a content item from one of the first plurality of content items and the second plurality of content items.

SYSTEM AND METHOD PROVIDING FOR SELF-HEALING VISUAL CODES (18044174)

Main Inventor

Alexander James Faaborg

Brief explanation

This abstract describes a system and method that can be used to identify visual codes, even if they are compromised or not fully scanned. The system uses supplemental information, such as location, images, and audio, along with a readable portion of the visual code, to determine the identity of the code and provide related information. This allows the system to accurately identify visual codes even if they are partially scanned or compromised.

Abstract

A system and method is provide which allows supplemental information to be used, in combination with a portion of a visual code that is readable, to identify the visual code, and to provide information related to the visual code, even in the event of a scan of a compromised visual code and/or an inadequate scan of the visual code which yields only a portion of the data payload associated with the visual code. The supplemental information may include, for example, location based information, image based information, audio based information, and other types of information which may allow the system to discriminate a location of the scanned visual code, and to identify the scanned visual code visual code based on the portion of the data payload and the supplemental information.

USING EMBEDDING FUNCTIONS WITH A DEEP NETWORK (17972466)

Main Inventor

Gregory S. Corrado

Brief explanation

The abstract describes a method for using embedded functions with a deep network. The method involves receiving input with different types of features, processing each feature using a specific embedding function to generate numeric values, and then using a deep network to generate an alternative representation of the input. Finally, a logistic regression classifier is used to predict a label for the input based on the alternative representation.

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using embedded function with a deep network. One of the methods includes receiving an input comprising a plurality of features, wherein each of the features is of a different feature type; processing each of the features using a respective embedding function to generate one or more numeric values, wherein each of the embedding functions operates independently of each other embedding function, and wherein each of the embedding functions is used for features of a respective feature type; processing the numeric values using a deep network to generate a first alternative representation of the input, wherein the deep network is a machine learning model composed of a plurality of levels of non-linear operations; and processing the first alternative representation of the input using a logistic regression classifier to predict a label for the input.

CONDITIONAL OUTPUT GENERATION THROUGH DATA DENSITY GRADIENT ESTIMATION (18010426)

Main Inventor

Nanxin Chen

Brief explanation

This abstract describes a method for generating outputs based on inputs from a network using neural networks. The method involves obtaining the network input, initializing a current network output, and generating the final network output through a series of iterations. Each iteration corresponds to a different noise level, and the current network output is updated at each iteration. This updating process involves using a noise estimation neural network to process the current network output and the network input, generating a noise output that provides a noise estimate for each value in the current network output. The current network output is then updated using the noise estimate and the noise level for that iteration.

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating outputs conditioned on network inputs using neural networks. In one aspect, a method comprises obtaining the network input; initializing a current network output; and generating the final network output by updating the current network output at each of a plurality of iterations, wherein each iteration corresponds to a respective noise level, and wherein the updating comprises, at each iteration: processing a model input for the iteration comprising (i) the current network output and (ii) the network input using a noise estimation neural network that is configured to process the model input to generate a noise output, wherein the noise output comprises a respective noise estimate for each value in the current network output; and updating the current network output using the noise estimate and the noise level for the iteration.

Video Anchors (18334648)

Main Inventor

Gabe Culbertson

Brief explanation

The abstract describes a method for analyzing videos by extracting anchor text and identifying entities mentioned in the audio. It also involves determining the importance of each entity and training a model using human rater data to predict entity labels for video anchors.

Abstract

In one aspect, a method includes obtaining videos and for each video: obtaining a set of anchors for the video, each anchor beginning at the playback time and including anchor text; identifying, from text generated from audio of the video, a set of entities specified in the text, wherein each entity in the set of entities is associated with a times stamp at which the entity is mentioned; determining, by a language model and from the text generated from the audio of the video, an importance value for each entity; for a subset of the videos, receiving rater data that describes, for each anchor, the accuracy of the anchor text in describing subject matter of the video; and training, using the human rater data, the importance values, the text, and the set of entities, an anchor model that predicts an entity label for an anchor for a video.

DATA VALUATION USING REINFORCEMENT LEARNING (18333301)

Main Inventor

Sercan Omer Arik

Brief explanation

This abstract describes a method for training a machine learning model using a batch of training samples. The method involves using a data value estimator model to generate predicted values for each training sample in the batch. Based on these predicted values, a subset of the training samples is selected. For each sample in the subset, the machine learning model is used to determine a prediction performance measurement. Finally, the estimator parameter values of the data value estimator model are adjusted based on these performance measurements.

Abstract

A method includes obtaining a batch of training samples. For each particular training sample in the batch of training samples, the method includes generating, using a data value estimator model and the particular training sample, a corresponding predicted value of the particular training sample when used to train a machine learning model. The method includes selecting, based on the corresponding predicted values, a subset of the batch of training samples. For each particular training sample in the subset of the batch of training samples, the method includes determining, using the machine learning model and the particular training sample, a corresponding prediction performance measurement. The method includes adjusting one or more estimator parameter values of the data value estimator model based on the corresponding prediction performance measurements.

ACTIVE LEARNING VIA A SAMPLE CONSISTENCY ASSESSMENT (18333998)

Main Inventor

Zizhao Zhang

Brief explanation

This method involves using a machine learning model to generate predictions for a set of unlabeled training samples. The predictions are compared to modified versions of the training samples to determine the difference. Based on these differences, a subset of the training samples is selected. Ground truth labels are obtained for the samples in the subset, and labeled training samples are generated using these labels. Finally, the machine learning model is trained using these labeled training samples.

Abstract

A method includes obtaining a set of unlabeled training samples. For each training sample in the set of unlabeled training samples generating, the method includes using a machine learning model and the training sample, a corresponding first prediction, generating, using the machine learning model and a modified unlabeled training sample, a second prediction, the modified unlabeled training sample based on the training sample, and determining a difference between the first prediction and the second prediction. The method includes selecting, based on the differences, a subset of the set of unlabeled training samples. For each training sample in the subset of the set of unlabeled training samples, the method includes obtaining a ground truth label for the training sample, and generating a corresponding labeled training sample based on the training sample paired with the ground truth label. The method includes training the machine learning model using the corresponding labeled training samples.

User Equipment-Coordination Set Federated for Deep Neural Networks (18027059)

Main Inventor

Jibing Wang

Brief explanation

The abstract describes a method and system for coordinating user equipment (UE) in a user equipment-coordination set (UECS) to perform federated learning for deep neural networks (DNNs). The coordinating UE communicates update conditions to a subset of UEs in the UECS, indicating when to generate updated machine learning (ML) configuration information for a DNN. The coordinating UE then receives reports containing the updated ML configuration information from the subset of UEs. Each UE generates the updated ML configuration information using a training procedure and local input data. The coordinating UE applies federated learning techniques to determine a common UECS ML configuration and directs at least one UE in the subset to update its respective DNN using the common UECS ML configuration.

Abstract

Techniques and apparatuses are described for user equipment-coordination set (UECS) federated learning for deep neural networks (DNNs). A coordinating user equipment (UE) in a UECS communicates (), to at least a subset of UEs in the UECS, one or more update conditions that indicate when to generate updated ML configuration information for a respective DNN that processes UECS communications. The coordinating UE then receives () one or more reports that include the updated ML configuration information from respective UEs of at least the subset of UEs. In aspects, the respective UE generates the updated ML configuration information using a training procedure and local input data. The coordinating UE determines () a common UECS ML configuration by applying federated learning techniques to the updated ML configuration information and directs () at least one UE in the subset to update the respective DNN using the at least one common UECS ML configuration.

CALIBRATION OF QUANTUM PROCESSOR OPERATOR PARAMETERS (18311178)

Main Inventor

Paul Klimov

Brief explanation

This abstract describes methods, systems, and apparatus for determining operating parameters for a quantum processor with multiple interacting qubits. The method involves creating a graph with nodes representing qubits and edges representing interactions between qubits. Each node and edge is associated with an operating parameter. An algorithm is selected to traverse the graph based on a traversal rule. Disjoint subsets of nodes or edges are identified based on their relationship via the traversal rule. Calibrated values for the nodes or edges in each subset are determined using a stepwise constrained optimization process, where constraints are determined using previously calibrated operating parameters.

Abstract

Methods, systems and apparatus for determining operating parameters for a quantum processor including multiple interacting qubits. In one aspect, a method includes generating a graph of nodes and edges, wherein each node represents a respective qubit and is associated with an operating parameter of the respective qubit, and wherein each edge represents a respective interaction between two qubits and is associated with an operating parameter of the respective interaction; selecting an algorithm that traverses the graph based on a traversal rule; identifying one or multiple disjoint subsets of nodes or one or multiple disjoint subsets of edges, wherein nodes in a subset of nodes and edges in a subset of edges are related via the traversal rule; and determining calibrated values for the nodes or edges in each subset using a stepwise constrained optimization process where constraints are determined using previously calibrated operating parameters.

Parameter Efficient Prompt Tuning for Efficient Models at Scale (17718738)

Main Inventor

Brian David Lester

Brief explanation

This abstract describes a method for natural language processing that uses trained prompts to guide a pre-trained machine learning model to generate specific outputs for a given task. The approach involves training a subset of parameters for the task and then inputting them along with input data into the pre-trained model. The pre-trained model's parameters are frozen during the prompt training, which helps reduce computational resources while still benefiting from the knowledge learned by the pre-trained model.

Abstract

Systems and methods for natural language processing can leverage trained prompts to condition a large pre-trained machine-learned model to generate an output for a specific task. For example, a subset of parameters may be trained for the particular task to then be input with a set of input data into the pre-trained machine-learned model to generate the task-specific output. During the training of the prompt, the parameters of the pre-trained machine-learned model can be frozen, which can reduce the computational resources used during training while still leveraging the previously learned data from the pre-trained machine-learned model.

ZOOM AGNOSTIC WATERMARK EXTRACTION (17926213)

Main Inventor

Dake He

Brief explanation

This abstract describes methods, systems, and computer programs for detecting and decoding a watermark in an image. The process involves using a machine learning model to determine if the image contains a visually imperceptible or perceptible watermark. If a watermark is detected, the image is sent to a watermark decoder for further processing. If no watermark is detected, the image is filtered out. The decoder then decodes the watermark, and the extracted data is used to validate the item depicted in the image.

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for detecting and decoding a visually imperceptible or perceptible watermark. A watermark detection apparatus determines whether the particular image includes a visually imperceptible or perceptible watermark using detector a machine learning model. If the watermark detection apparatus detects a watermark, the particular image is routed to a watermark decoder. If the watermark detection apparatus cannot detect a watermark in the particular image, the particular image is filtered from further processing. The watermark decoder decodes the visually imperceptible or perceptible watermark detected in the particular image. After decoding, an item depicted in the particular image is validated based data extracted from the decoded visually imperceptible or perceptible watermark.

ZOOM AGNOSTIC WATERMARK EXTRACTION (18008544)

Main Inventor

Dake He

Brief explanation

This abstract describes methods, systems, and computer programs for determining and decoding watermarks in images. A watermark decoder uses a machine learning model to decode watermarks at different levels of zoom. It determines whether a watermark was successfully decoded and outputs a result based on this determination. This includes outputting a zoomed output of the decoded watermark.

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining a visually imperceptible or a visually perceptible watermark and outputting a result based on the determination. A watermark decoder receives an input image. The watermark decoder applies a decoder machine learning model to decode a watermarks at different levels of zoom. The water mark decoder determines whether a watermark was decoded to obtain a decoded watermark. The watermark decoder outputs a result based on the determination whether the watermark was decoded through application of the decoder machine learning model to the input image that includes outputting a zoomed output decoded through application of the decoder machine learning model to the input image.

CREATING SECURE PIPELINE CACHE OBJECTS USING DIFFERENTIAL PRIVACY TECHNIQUES (18210974)

Main Inventor

Robert Fraser

Brief explanation

The abstract describes a method for reconstructing graphics pipeline caches in a gaming session. These caches store information about graphical objects rendered in previous gaming sessions. By using pipeline structures that represent these objects, the method can recreate the graphics pipeline caches for the current gaming session. These caches are then used to initialize graphics pipelines that render the graphical objects in the gaming application.

Abstract

A graphics pipeline cache reconstruction operation is implemented to reconstruct one or more graphics pipeline caches for a current client gaming session based on one or more pipeline structures. The pipeline structures each represent a graphical object rendered during a respective previous client gaming session and are used to reconstruct one or more graphics pipeline caches that include graphics pipeline cache objects related to the graphical objects of the pipeline structures. These graphics pipeline cache objects are used to initialize one or more graphics pipelines used to render the graphical objects in a gaming application for a current client gaming session.

SYSTEMS AND METHODS FOR INPAINTING IMAGES AT INCREASED RESOLUTION (18022396)

Main Inventor

Soo Ye KIM

Brief explanation

This abstract describes a method for inpainting masked regions in an input image using neural networks. The method involves using a first neural network to inpaint the masked regions at a lower resolution, and then increasing the resolution to create a second inpainted image with higher resolution. This second inpainted image is then provided to a second neural network, which produces a refined version of the inpainted image at the higher resolution.

Abstract

A method includes receiving an input image. The input image corresponds to one or more masked regions to be inpainted. The method includes providing the input image to a first neural network. The first neural network outputs a first inpainted image at a first resolution, and the one or more masked regions are inpainted in the first inpainted image. The method includes creating a second inpainted image by increasing a resolution of the first inpainted image from the first resolution to a second resolution. The second resolution is greater than the first resolution such that the one or more inpainted masked regions have an increased resolution. The method includes providing the second inpainted image to a second neural network. The second neural network outputs a first refined inpainted image at the second resolution, and the first refined inpainted image is a refined version of the second inpainted image.

FACIAL IMAGE EDITING AND ENHANCEMENT USING A PERSONALIZED PRIOR (18334700)

Main Inventor

Kfir Aberman

Brief explanation

This abstract describes a technology that uses a set of images of a person to create a personalized model. This model can then be used to edit images of that person by filling in missing areas, improving resolution, or reducing blurriness. The technology ensures that the edited images still retain the person's unique features.

Abstract

Systems and methods for identifying a personalized prior within a generative model's latent vector space based on a set of images of a given subject. In some examples, the present technology may further include using the personalized prior to confine the inputs of a generative model to a latent vector space associated with the given subject, such that when the model is tasked with editing an image of the subject (e.g., to perform inpainting to fill in masked areas, improve resolution, or deblur the image), the subject's identifying features will be reflected in the images the model produces.

IMAGE PROCESSING SYSTEM FOR VERIFICATION OF RENDERED DATA (18335431)

Main Inventor

Tuna Toksoz

Brief explanation

The abstract describes an image processing system that verifies if embedded digital content meets a specific requirement for display. The system includes a content embedding engine that adds content to a resource, a rendering engine that displays the embedded content, an application interface engine that generates a visualization of the resource and embedded content, and an image processing engine that analyzes the visualization to determine if it meets the predetermined criterion. The system then sends verification data indicating whether the criterion is satisfied or not.

Abstract

An image processing system for verifying that embedded digital content satisfies a predetermined criterion associated with display of the content, the image processing system a content embedding engine that embeds content in a resource provided by a content provider and that configures the resource for rendering, a rendering engine that renders the content embedded in the resource; an application interface engine that interfaces with the rendering engine and that generates a visualization of the resource and of the embedded content rendered in the resource; and an image processing engine that processes one or more pixels of the generated visualization of the resource and of the embedded content and the resource to verify that the specified visual element satisfies the predetermined criterion; and transmits verification data comprising an indication of whether the predetermined criterion is satisfied.

Calibration-Free Instant Motion Tracking for Augmented Reality (18335614)

Main Inventor

Jianing Wei

Brief explanation

This abstract describes a system and method for instant motion tracking without the need for calibration. It is specifically designed for rendering virtual content in augmented reality settings. The system uses a computing system to analyze sequential image frames and estimate the rotation of the image capture system as well as the translation of an anchor region within the scene. This information allows the system to accurately determine where and how to render virtual content within the image frames.

Abstract

The present disclosure provides systems and methods for calibration-free instant motion tracking useful, for example, for rending virtual content in augmented reality settings. In particular, a computing system can iteratively augment image frames that depict a scene to insert virtual content at an anchor region within the scene, including situations in which the anchor region moves relative to the scene. To do so, the computing system can estimate, for each of a number of sequential image frames: a rotation of an image capture system that captures the image frames; and a translation of the anchor region relative to an image capture system, thereby providing sufficient information to determine where and at what orientation to render the virtual content within the image frame.

POWER LINE DESIGN MODIFICATION TO MITIGATE VERTICAL BAND CROSSTALK (17754830)

Main Inventor

Sangmoo Choi

Brief explanation

This abstract discusses techniques to ensure uniform luminance across a computing device display, specifically an active matrix organic light emitting diode (AMOLED) display. The display may have a hole within its active area for a camera, button, or other function, which can cause a non-uniform voltage drop in the power supply mesh near the hole. The power supply mesh supplies electrical energy to the display elements. The techniques described in this abstract involve using portions of the power supply mesh that are not connected to the voltage supply bus to maintain a uniform voltage drop across the entire active area of the display.

Abstract

Techniques to provide uniform luminance across a computing device display, such as an active matrix organic light emitting diode (AMOLED) display. In some examples, a computing device display may include a hole within the active area of the display that may be used for a camera, a button or some other function. The hole may result in a non-uniform voltage drop in a power supply mesh in regions of the active area near the hole. The power supply mesh may provide electrical energy to elements of the display. The techniques of this disclosure include portions of the power supply mesh that are not connected to the voltage supply bus to ensure a uniform voltage drop across the active area of the display.

Unified Cascaded Encoder ASR model for Dynamic Model Sizes (18182925)

Main Inventor

Shaojin Ding

Brief explanation

The abstract describes an automated speech recognition (ASR) model that consists of several components. The first encoder takes a sequence of acoustic frames as input and generates a higher order feature representation for each frame. The first decoder uses this representation to generate a probability distribution for different speech recognition hypotheses. The second encoder takes the same higher order feature representation and generates another representation for each frame. The second decoder then uses this representation to generate another probability distribution for speech recognition hypotheses.

Abstract

An automated speech recognition (ASR) model includes a first encoder, a first encoder, a second encoder, and a second decoder. The first encoder receives, as input, a sequence of acoustic frames, and generates, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The first decoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a first probability distribution over possible speech recognition hypotheses. The second encoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a second higher order feature representation for a corresponding first higher order feature frame. The second decoder receives, as input, the second higher order feature representation generated by the second encoder, and generates a second probability distribution over possible speech recognition hypotheses.

LITHIUM ION BATTERY WITH COMPOSITE ELECTRODES (18008847)

Main Inventor

Taisup Hwang

Brief explanation

The abstract describes composite lithium ion batteries that have an anode, a cathode, a separator, and an electrolyte. These batteries have several advantages compared to regular lithium ion batteries. They have lower DC impedance, which means they can deliver power more efficiently. They also charge faster and have more reserve capacity between 3.0V and 2.5V. Additionally, these batteries have a higher volumetric energy density, meaning they can store more energy in a smaller space.

Abstract

Composite lithium ion batteries having an anode with an anode collector and a composite anode material, a cathode with a cathode collector and a composite cathode material, a separator positioned between the anode and the cathode, and an electrolyte in contact with the anode and the cathode. Advantages of the composite lithium ion batteries include lower DC impedance, faster charging times, more reserve capacity between 3.0V and 2.5V, and an increased volumetric energy density relative to lithium ion batteries that do not include the described composite material electrodes.

TECHNIQUES FOR DYNAMIC VOLTAGE CONTROL FOR LASER DIODES IN HEAD-WEARABLE DISPLAY DEVICES (18209793)

Main Inventor

Kai Xu

Brief explanation

This abstract describes a device, specifically a wearable display device, that uses a laser diode driver with dynamic voltage control for laser diodes. The device can have different colored laser diodes, such as red, blue, and green. The device analyzes the frame to be displayed and determines the load condition, such as the image signal processor frame buffer load. Based on this load condition, the device calculates a target voltage level for the laser diode. This target voltage level is generated by adding it to a base voltage level using a voltage booster.

Abstract

The present disclosure relates generally to a device, such as a wearable display device configured with a laser diode driver implementing dynamic voltage control for laser diodes. The laser diodes may include one or more of a red laser diode, a blue laser diode, and a green laser diode. The device may determine a load condition based on a frame to be displayed at the device, and determine a target voltage level for a laser diode operably coupled to the laser diode driver of the device based on the load condition (e.g., an image signal processor (ISP) frame buffer load). The device may generate the target voltage level for the laser diode based on a base voltage level. For example, the device may be configured with a voltage booster operably coupled to the laser diode driver to provide the target voltage level in addition to the base voltage level.

Distributed, Private, Sparse Histograms in the Two-Server Model (18297084)

Main Inventor

Badih Ghazi

Brief explanation

The abstract describes systems and methods for calculating sparse, (ε, δ)-differentially private histograms in a secure multi-party computation (MPC) model with two servers. These protocols allow two servers to compute histograms over data from multiple users, while only gaining access to a private view of the data.

Abstract

Provided are systems and methods for the computation of sparse, (ε, δ)-differentially private (DP) histograms in the two-server model of secure multi-party computation (MPC). Example protocols enable two semi-honest non-colluding servers to compute histograms over the data held by multiple users, while only learning a private view of the data.

METHODS, SYSTEMS, AND MEDIA FOR PRESENTING INFORMATION RELATED TO AN EVENT BASED ON METADATA (18205050)

Main Inventor

Eric HC Liu

Brief explanation

This abstract describes methods, systems, and media for presenting information about an event based on metadata. It involves detecting a user-initiated event on a device and using sensors connected to the device to obtain content items related to the event. The content items are then associated with the event based on device information from various data sources. Information related to the content items is retrieved and additional content items related to the initial ones are detected. Finally, information about the user-initiated event is presented on the device based on these determinations.

Abstract

Methods, systems and media for presenting information related to an event based on metadata are provided. In some implementations, the method comprises: detecting that a user-initiated event has been performed on a user device; detecting a plurality of sensors connected to the user device; causing a plurality of content items related to the user-initiated event to be obtained using the plurality of sensors connected to the user device; determining whether each of the plurality of content items is to be associated with the user-initiated event based on user device information from a plurality of data sources; associating a portion of the plurality of content items with the user-initiated event based on the determination; retrieving information related to the portion of the plurality of content items; detecting, using the plurality of sensors connected to the user device, a second plurality of content items; determining that at least one of the second plurality of content items is related to at least one of the first plurality of content items; and causing information related to the user-initiated event to be presented by the user device based at least in part on the determination.

Wi-Fi Adaptive Beacon Skipping for Battery-Powered Devices (18334944)

Main Inventor

Howard M. Harte

Brief explanation

The abstract describes a Wi-Fi chip that has a power-saving mode. In this mode, the chip checks for periodic beacons transmitted by a wireless access point at a longer interval than the access point transmits them. When the chip receives a standby message, it enters a standby mode for a specific duration. In this mode, the chip checks for the beacons at a shorter interval than before.

Abstract

A Wi-Fi chip is configured to operate in a power saving Wi-Fi mode in which the Wi-Fi chip repeatedly checks, at a first time interval, periodic DTIM beacons transmitted at a second time interval by a wireless access point of a Wi-Fi network, wherein the first time interval is longer than the second time interval. Upon receiving a standby message, the Wi-Fi chip, during a predefined time duration, operates in a standby mode in which the Wi-Fi chip checks at a third time interval periodic DTIM beacons transmitted by the wireless access point, wherein the third time interval is shorter than the first time interval.