Digital Commons Network^™

Spatially And Temporally Directed Noise Cancellation Using Federated Learning, Shantanu Goel, Piyush T. Itankar

Defensive Publications Series

Machine learning models can be trained to cancel noise of diverse types or spectral characteristics, e.g. traffic noise, background chatter, etc. Such models are trained by feeding training data that includes labeled noise waveforms, which is an expensive and time-consuming procedure. Further, the effectiveness of such machine learning models is limited in canceling types of noise absent from training data. Trained models occupy significant amounts of memory which limits their use in consumer devices. This disclosure describes the use of federated learning techniques to train noise canceling models locally at diverse device locations and times. With user permission, the trained …

Cooking System, Lukasz Wasinkiewicz, Stefan Stefan Brenner, Walter Frost, Constantin Krauße

Defensive Publications Series

The invention refers to a cooking system, preferably cooking hob, more preferably induction hob.

Cooking System, Ab Electrolux - Group Patents

Defensive Publications Series

This disclosure describes a cooking hob, particularly an induction cooking hob.

Slicing Notes Based On Context Data And Meeting Activity, Sandro Feuz, Thomas Deselaers, Philippe N. Gervais, Julie Zhu, Robin Dua, Andrew Bunner, Prithi Pichai, Trond Thomas Wuellner, Jake Walker, Jin Kim, Ming Guang Yong

Defensive Publications Series

Notes taken by users are often unstructured, e.g., users may mix notes from different meetings into one document or divide notes from one meeting-series across different documents. This increases the difficulty in the retrieval and clustering of notes relevant to a particular topic. Currently, in a tedious process, users that want to organize their jottings from several meetings manually select and collate relevant sections into a single document. This disclosure describes techniques to automatically cluster and assemble relevant meeting notes that are present across one or more documents to generate a single, theme-based view, based, e.g., on topic, co-attendees, location, …

Obtaining Query Examples For Virtual Assistant Actions Using Search Logs, Asaf Zomet, Tomer Shmiel, Dvir Keysar, Amitai Cohen

Defensive Publications Series

A user’s interaction with a virtual assistant typically involves the issuance of commands to perform certain actions, e.g., “play Mozart,” “turn on the lights every night,” etc. To execute actions accurately, a virtual assistant is trained to understand or detect action intent. This disclosure describes techniques to obtain a wide variety of query examples, with user permission, to train a virtual assistant to understand action intents. Information-seeking queries submitted to a search engine are transformed into action intents, e.g., commands, based on the grammar of the language of the search-engine query. The techniques are especially useful when launching new features, …

Secure And Fully Transparent Roaming For Long Range, Marcelo Yannuzzi, Carlos M. Pignataro, Bart Brinckman

Defensive Publications Series

Roaming with Long Range wide-area network (LoRaWAN) requires connections to home Network Servers (hNS), serving Network Servers (sNS) and forwarding Network Servers (fNS) across LoRa domains. Today, this entails a tedious manual process for generating the keys, certificates and configuration files to connect these servers on a peer-to-peer basis. This means that manual configurations grow quadratically with the number of LoRa networks involved. In addition, today there is no way to enable roaming across LoRa networks dynamically (i.e., not if these pre-configurations among the visited and the home/serving LoRa networks aren't in place). Accordingly, presented herein are techniques to scale …

Detecting Missing Capacitors By Load Sensing, Puneet Gupta, Chintan Trehan

Defensive Publications Series

Factory testing on a production line can uncover several types of device failures. However, certain subtle failures can be masked from test procedures only to be discovered too late, e.g., by the customer. An example of such a failure is a missing bulk capacitor in circuits such as amplifiers, regulators, etc. Such a failure can go undetected at a factory due to the presence of smaller capacitors that have a marginally compensating effect. This disclosure describes techniques to detect missing bulk capacitors on circuit boards by inserting a simulated load resistance and measuring the decay time of an injected test …

Providing Responses To Local-Intent Queries From Web Results, N/A

Defensive Publications Series

Local search engines use stored structured data to answer user queries relating to businesses and other places of interest in specific locations. When such structured data is unavailable, a local search engine cannot provide useful responses to user queries. This disclosure describes techniques to provide responses to local-intent queries based on snippets from web results. Responses are provided that include attributes such as phone number, address, etc. of local businesses, and businesses within specified categories (e.g., restaurants, repair shops, etc.).

Using Eeg To Predict Search Intent And To Control Iot Devices, N/A

Defensive Publications Series

It has been demonstrated that electroencephalogram (EEG) readings can be mapped to images, audio, or language perceived by a user, such that a machine-learning model trained on EEG readings can infer the user’s thoughts. This disclosure describes techniques that use a head-mounted device, e.g., smart glasses, to measure, with user permission, a user’s EEG, and based on the measurement, infer the user’s intent or reconstruct the user’s thoughts, e.g., images or audio therein. In situations where a user can only partially recall a search phrase, the techniques enable determination of the user’s search intent from the user’s EEG. In other …

Error Correction In Automatic Speech Recognition, Ágoston Weisz, Ragnar Groot Koerkamp

Defensive Publications Series

This disclosure describes techniques to correct errors in automatic speech recognition, e.g., as performed to recognize spoken queries from a user to a virtual assistant or other application. A machine learning model detects potentially misrecognized n-grams within transcribed text which are then underlined in a user interface. A user can tap on the underlined n-gram, or another portion of the transcribed text to activate a dropdown menu that presents alternatives to the transcribed text. The alternatives can be based on speech hypothesis scores. To correct the error in transcribed text, the user picks an alternative from the dropdown menu, or, …

Mixed Language Speech Recognition, Ágoston Weisz, Herbert Jordan, Konrad Jan Miller

Defensive Publications Series

Mixed language speech, e.g., speech in two or more languages, is common in many geographies and application domains. A mixed language voice command to a virtual assistant is likely to be misinterpreted by conventional speech recognition techniques that are based on a single language model. This disclosure describes techniques to combine multiple language models to perform multilingual speech recognition of mixed language speech and commands.

Single-Click Ad-Catching On Tv And At The Movies, Venu Annamraju

Defensive Publications Series

It is today possible for a user with an appropriate mobile device application to catch a television ad, e.g., to make a purchase taking advantage of an offer in an ad that just aired. However, such ad-catching is still generally a multistep procedure with some friction. It is often the case that by the time a user catches a short-formatted ad (of a duration between 15-30s) with their ad-catching application, the time window for the offer has expired. This disclosure describes techniques that enable a user to catch an ad aired on television or in a movie theater with a …

Graph Visualization By Organizing Connections In Collapsible Hierarchical Graphs, David Souther, Alan Samanta, Mateusz Jedruch, Matthew Tschiegg, Szymon Barglowski

Defensive Publications Series

Network graphs in certain applications, e.g., cloud-network graphs, have connections in multiple dimensions. At present, it is difficult or inconvenient for a user to visualize such graphs at varying levels of granularity or hierarchy. Per the techniques of this disclosure, a hull is defined as a node with descendants, and a segment is defined as a bundle of edges between descendants below a pair of nodes. By enabling a user to expand or collapse a hull, and by routing edges via segments connecting parent nodes, the described techniques enable a high-level visualization of large graph networks that can be quickly …

Sharing Photos And Associated Text Messages Within Messaging Apps In The Intended Order, Victor Cărbune, Thomas Deselaers

Defensive Publications Series

The order in which a photo and messages are delivered and displayed in a chat conversation is affected by processing and/or network delays as well as specific implementations of messaging applications. Receiving a photo out-of-sequence from text messages related to the photo can create confusion and/or misunderstanding and can ruin the intended effect of the sender’s communication. This disclosure describes the use of machine learning techniques to identify text messages that are associated with a shared photo and bundling those together as a joint set. Descriptive text for the photo is shown as a placeholder until the photo is fully …

Contextual Error Correction In Automatic Speech Recognition, Manaal Faruqui, Janara Christensen

Defensive Publications Series

This disclosure describes techniques that leverage the context of a conversation between a user and a virtual assistant to correct errors in automatic speech recognition (ASR). Once confirmed by the user, the correction event is used to augment the training data for ASR.

Automatic Detection And Use Of Content Available In Multiple Languages, Matthew Sharifi

Defensive Publications Series

Content producers often create and serve versions of their pages and websites in multiple languages. The versions in each language produced by the content provider typically incorporate the provider’s in-depth knowledge of the content and associated nuances. When available, users are likely to prefer content in their preferred language that originates from the content producer to that generated by automated translation of the content from another language, e.g., as provided by a web browser or translation service. This disclosure describes techniques that, with permission from the user, detect whether the content provider of the web page or website has made …

An Algorithm That Uses Live Updates Provided By Clustered Users Attending A Live Event To Show Updates For The Live Event On Google Search And Assistants, N/A

Defensive Publications Series

As technology improves, users increasingly turn to electronic devices to obtain various information in which they are interested. Information services can be provided to deliver information that users are interested in upon request. Particularly, information describing the current state of live sporting events can be of particular interest to users who are not in attendance at the live sporting event. However, accessing and disseminating information for live events while those events are occurring can be difficult.

Automatic Activation Of Mobile Device Security Upon Detection Of Theft, Peter Lewis

Defensive Publications Series

This disclosure describes techniques to reduce the risk of a data breach when a mobile device is stolen. In certain modes, e.g., augmented reality (AR) mode, selfie mode, etc., optimal use of a mobile device requires that the user hold the device out at an arm’s length in a public location, leading to an increased likelihood of theft. With user permission, a potential risk state is activated upon detecting such use of the device. If sudden movement of the device, e.g., as detected by an on-board accelerometer, location sensor, or other sensor is detected, the device is automatically locked, requiring …

Automatic Input Mode Selection For Text Entry, Vilem Ries

Defensive Publications Series

Users can provide input to a computing device via various modalities such as keyboard, mouse, via voice, etc. When interacting with a virtual assistant, users currently need to switch manually between keyboard and voice input. Such cumbersome interaction can lead to reduced usage of a virtual assistant or other software. This disclosure describes a user interface in which a virtual keyboard and voice input mode are both activated simultaneously. When the user starts providing input via one modality, the other modality is automatically turned off. The user interface enables fast interaction with a virtual assistant or other software.

System For Faulty Battery Identification And Control, James Robert Lim, Jesse Liang, Wilson Tang, Rob Oliver, Aaron Chen, Ravi Soma

Defensive Publications Series

This publication describes systems and techniques to identify a faulty battery pack in a portable electronic device. Portable electronic devices, such as video game controllers and smartphones, are often powered by rechargeable lithium-ion or lithium-ion polymer battery packs. These battery packs must comply with various safety standards and regulations. As one example, one industry standard requires identification of a faulty battery pack and termination or non-initiation of battery charging. In some portable electronic devices, compliance with this standard (and other standards and regulations) requires additional or specialized hardware components that increase the cost of the electronic device and take up …

Recording And Automated Execution Of Driving Cycles By Means Of Functional Expansions In The Existing Control Unit Network, Verena Blunder

Defensive Publications Series

No abstract provided.

Machine-Learning Based Smoke Detection, Dongeek Shin, Andrii Korchak, Andy Goldenson

Defensive Publications Series

A machine learning based arrangement can be used to more accurately detect whether a smoke alarm should be sounded based on a determined rate of change in the measured amount of smoke. The machine learning model may be pre-trained based on training data then executed by a smoke detector to accurately distinguish between likely emergencies and nuisance conditions.

Enhanced Whiteboard Capture For Chat, Anon Ymous

Defensive Publications Series

A system may provide a mechanism for chat and collaboration services, whereby:

1) The system may allow a user of a chat or collaboration service to activate a specialized flow for uploading a whiteboard image, which uploaded image may then be subject to further processes of the system pertaining to whiteboard images.

2) The system may allow a user to tag or otherwise indicate images uploaded to a chat or collaboration service as being whiteboard images, subsequent to those images having been uploaded, which images may then be subject to further processes of the system pertaining to whiteboard images.

3) …

Personalized Augmented Reality Content Delivery Using Computer Vision, Aiko Nakano, Diane Wang

Defensive Publications Series

This disclosure describes techniques that personalize, with user permission, a user’s real surroundings with augmented reality content that is matched to user preferences and history. With user permission, an AR device, e.g., AR glasses, a head-mounted device, etc., discovers physical objects in the user’s surroundings. The device downloads AR content relevant to the physical objects that match user preferences and prior user history. The downloaded AR content is delivered to the user via text-to-speech, audio, visual, or audiovisual mediums. A next point-of-interest in the user’s journey is suggested.

Providing Informative Nutrition Facts For Food Ordering, Xinru Yang, Mingtong Wu, Xingyue Chen

Defensive Publications Series

Online food ordering services provide little to no information regarding nutritional facts. This makes it difficult for customers to make decisions based on their nutrition and health needs. This disclosure describes an online food ordering service that provides users the nutritional facts of a food item prior to order placement. The information is provided via a suitable user interface of a website or app that enables the user to drill down further to obtain more granular information. If the user permits, prior orders of the users are used to provide periodic feedback regarding eating behavior and nutritional intake.

Using Clustering Algorithm And Fog Computing For En-Route Filtering In Low Power And Loss Networks, Lele Zhang, Akram Sheriff, Chuanwei Li

Defensive Publications Series

Due to trustless link quality in Wireless Mesh Networks (WMNs), Power Outage Notification (PON) and Power Restoration Notification (PRN) messages are often dropped or delayed en-route, which may fail to satisfy customer requirements in practice. Therefore, proposed herein are techniques that use machine learning and Fog computing to efficiently deduce missing PON/PRN messages.

Surface Positioning Sensor For Contact Between Unrestrained Objects, Ryan Ehrenreich, Stuart Scollay, Allison Juhasz

Defensive Publications Series

The device of this invention is a flexible, adaptable, accurate system for detecting surface positioning data during physical contact between multiple objects. For any pair of objects, the technology records where a fixed location on the surface of the first object (striker object or Striker) makes contact within the perimeter of a positioning sensor on the surface of the second object (target object or Target).

Limiting Application Access To Sensitive Data And Resources Through Use Of A Sensitive Data Sandbox, Russell Quong

Defensive Publications Series

This publication describes techniques and apparatuses to limit access by a software application to sensitive user data (e.g., contacts, location, biometric data, photos). When a user installs or initially uses an application, the application may request permission to access user data, along with system resources (e.g., global positioning satellite (GPS) data) and particular hardware (e.g., microphones, cameras). Depending on the permission request, the user and the operating system (OS) may be unclear whether the data, resources, and/or hardware is needed by the application most of the time, some of the time, not needed at all, or whether the request is …

Safeguarding Against Improper Configurations For Lorawan Gateways, Su Xia, Christine Hwang, Bob Lo, Xiaochen Cao, Simarpreet Kaur

Defensive Publications Series

Presented herein are safeguard features for Long Range wide-area network (LoRaWAN) deployments on a LoRa gateway. The techniques presented herein provide a solution to prevent end users from wrongly or accidentally deploying LoRa gateways in unlawful Industrial, Scientific and Medical (ISM) radio bands. The techniques presented herein require little computational and storage resources to be integrated into any LoRa gateway embedded system, operate automatically based on the standard LNS protocol, and are sufficiently flexible to suit different use cases.

Hypervisor-Agnostic System And Method For Modification Of Snapshot Files For Recovery, Mayuresh Vartak, Geetha Srikantan, Vishwas Srinivasan, Harpreet Gandhi, Swapnil Daingade

Defensive Publications Series

This proposal provides a general-purpose technique to reconfigure the execution environment for virtual machines to facilitate a file patch service for modifying snapshot files during virtual machine recovery. The technique is hypervisor-agnostic and may be implemented within many different types of environments.