Digital Commons Network^™

Ssd Admission Control For Content Delivery Networks, Dragos Ionescu, Richard Schooler, Kenneth Barr

Defensive Publications Series

Content delivery networks (CDNs) are utilized to deliver content with low latency. Solid State Drive (SSD) devices are a cost-effective option for intermediate tier caches in servers used in CDNs. However, excessive writes of content to SSD devices can degrade useful life of such devices. This disclosure describes admission control techniques to manage content in an SSD cache. In particular, a popularity metric for content items is determined based on an inter-arrival time. The popularity metric takes into account past admission history and is dynamically to account for variations in content popularity over different time periods. These techniques balance the …

Non-Deterministic Video Frame Sampling To Thwart Frame Insertion Attacks, Juhyun Lee

Defensive Publications Series

Accurate machine understanding of videos is important, e.g., to maintain the integrity and policy compliance of videos at a video hosting website. For example, a video-hosting website with a policy of hosting only family-friendly videos deploys video understanding systems to automatically exclude non-compliant videos. Techniques of image analysis are applied to individual video frames to understand the video. Due to the computational cost of processing individual frames, videos are sampled, e.g., at a rate of one frame per second, and only the sampled frames are subjected to image analysis.

Sampling-based video understanding is susceptible to attack. For example, such systems …

Increase The Power Loss Of The Electric Machine To Heat The Hv-Battery, Daniel Hoppe

Defensive Publications Series

INCREASE THE POWER LOSS OF THE ELECTRIC MACHINE TO HEAT THE HV-BATTERY

Automated Capturing And Reporting Of Real World Data From Autonomous/Automated Vehicles, Matt Ruhstaller, Phillip Ames

Defensive Publications Series

A system and method are disclosed that automatically captures and reports real world data from autonomous/automated vehicles. The system includes sensors such as camera, GPS, microphone, accelerometer etc. that provide real-world tracking features in an automobile. The system may report the data to a reporting station to act on the reported data for a wide variety of applications. Cameras may detect billboards on a given route, which may provide information on ad impressions in the real world to ad buyers. The system might detect a pothole and relay its location, dimensions etc. to local authority who may take action to …

Volumetric Slider On User Interface, John N.A. Brown

Defensive Publications Series

A device and a method are disclosed for providing an improved volumetric slider on a touch user interface (UI). The device includes a UI, such as a touch screen, displaying a volumetric slider or controller. The controller is designed as a blob for adjusting the volume by a natural pinch or un-pinch motion using fingertips. The un-pinch motion stretches the blob to create two smaller blobs and a visually-compelling thread, which indicates a distance between the two smaller blobs. The thread allows the user to judge the adjustment they have made to the volume. Further, the thread can be brought …

An Eye Tracking System For Smart Devices, Justin Lewis, Thomas Price

Defensive Publications Series

A system and method is disclosed to track user’s eye gaze on a lockscreen and infer whether a user has read each message on the screen. An incoming message is initially marked as unread. The system presents the message to the user on a lockscreen. Eye tracking software uses the camera's video stream to detect that the user's gaze is directed at the message on screen. When the user's gaze is detected on the message for a sufficient duration, the system may tentatively mark that message as read. The system may notify and request confirmation from the user before marking …

Allocating Rf Spectrum Partitions In A Cbrs Network, Yi Hsuan

Defensive Publications Series

Methods for allocating portions of an available radio frequency spectrum to wireless access points and systems for carrying out such methods are disclosed herein. An example method can include receiving location and transmission information for a plurality of wireless access points associated with respective networks. A coexistence graph for the plurality of wireless access points is generated based on the received information. The coexistence graph is divided into coexistence areas. For each coexistence area, a coloring algorithm is executed to determine a number of colors needed to color the nodes in the coexistence area such that no two connected nodes …

Pause Viewing Button For A Multiple-Person Household To Facilitate Resuming Later, Rochelle Edwards

Defensive Publications Series

A method for resuming playback of a media item after user co-participation in playback of the media item is described. A processing device provides playback of a media item to a first user and a second user. The processing device determines that the first user is not participating in the playback at a first point in time corresponding to a first temporal location in the media item and that the second user is not participating in the playback at a second point in time corresponding to a second temporal location in the media item. The processing device receives a request …

Low-Bandwidth, Client-Based, Rendering For Gaming Videos, Ian Fischer, Shumeet Baluja

Defensive Publications Series

A system for low-bandwidth, client-based, rendering for gaming videos is described. The system may include a gaming device, server device, and user devices. The gaming device may include a processing device and graphics processing unit (GPU). The processing device receives user input and generates rendering commands from the user input. A first rendering unit of the GPU generates gaming video from the rendering commands. The server device receives the gaming video and the rendering commands from the gaming device. The server device determines the first user device is not compatible with the rendering commands, compresses the gaming video, and transmits …

Virtual Reality Video Compression, Andrew Ian Russell, Matthew Milton Pharr, Evan Rapoport

Defensive Publications Series

Uncompressed light field files tend to be big, and may take gigabytes of space for storing a small portion of a scene. However, there can be redundancy in the data. Therefore, compressing the data can result in significant reduction of data size for transmission and/or storage. Further, interpolating uncaptured light-fields can be based on two blurry images generated from a frame of streaming video.

Neural Network Based Regression Test Selection, Komateswar Dhanadevan, Jayasankar Nallasamy, Sundaravadivel Murugavel

Defensive Publications Series

Need for regression testing has been ever growing demand for software development organizations so does the cost of executing regression test suites. The intent of this idea is to propose a unique method in regression test for selecting a subset of test cases as oppose to select all test cases for a given software change using Neural Network. In general, a software consist of multiple features and it interacts with each other in the order of different magnitude. Each feature interaction is assessed based on five relative dimensional complexity. These complexities for each feature interaction is determined and subjected to …

Elastic Multi-Resolution Model-Serving To Compute Inferences, Christopher Olston, Noah Fiedel, Ed H. Chi, Alexander Beutel

Defensive Publications Series

Machine-learning models are consuming an increasing fraction of the world's computing resources. The cost of computing inferences with some machine-learning models is extremely high. Provisioning computing resources for peak performance, e.g., high availability and quality of service, entails the creation of headroom for traffic spikes (increases in demand) and preparing for the possibility of outages (decreases in capacity). Executing computer applications that utilize machine-learning models, also known as machine-learned models, can require significant capital and operational expenses.

This disclosure describes techniques to optimize use of computing resources for a machine-learning model. Multi-resolution models and/or models with recurrence are utilized. These …

Context-Sensitive Audio Notifications, Sandro Feuz, Victor Carbune, Thomas Deselaers, Google Inc.

Defensive Publications Series

A method is disclosed for triggering context-sensitive audio notifications. Systems and devices according to the method also predict a least disruptive moment to trigger a notification within a given period of time. The method comprises two parts. In the first part, a machine learning (“ML”) subsystem determines a current context of a user (based on, e.g., surrounding audio environment, current location, user calendar, etc.) and assesses the potential disruptiveness of an audible notification in the current context. The ML subsystem predicts a time-series of scores denoting the disruptiveness of the audible notification if emitted at particular times in the future. …

A/B Experiment Framework For Native Ads In Ad Servers, Bo Lin, Chetna Mcdonald, Roman Apostol, Sonny Thai, Woo Lee, Mikhail Shalai, Zutao Zhu, Olawale Abiri

Defensive Publications Series

Techniques are described for experiment frameworks around server-side rendering of native advertising content for presentation within an application environment. As described, a server receives a request, from a client device, for content that includes a content slot identifier. The server, using the content slot identifier, accesses a database to identify a presentation schema used to arrange content components and generates rendered content files according to a specific configuration for display within the environment of the resource. The server renders the content components based on the identified presentation schema and transmits the rendered content to the client device for display within …

Self Configuration Of Wireless Devices For Cloud Deployments, Frédéric Brunner, Jean-Michel Lauriol

Defensive Publications Series

In a context of Cloud deployments, we need to find a way for “out of the box” devices with no MMI (Man to Machine Interface) capabilities to easily discover the Cloud Service Provider and Application Server instance they are attached to.

These devices “MMI less” are devices without input (e.g. keyboard ) or output (e.g. Display) allowing user to type any characters or read information on the screen, typically wireless DECT (Digital Enhanced Cordless Telecommunications) Base Station or WLAN (Wireless Local Area Networks) Access Points.

Delivery Box With Perforated Folding Edges, Roger Ryan Quinlan, Daniel Lee

Defensive Publications Series

A delivery box is provided that is perforated at one or more of the folding edges to allow the box to be easily ripped apart for placement in a recycling bin or for storage.

Mechanism For Fine-Grained Software Controlled Core Dumps, Matthew D. Fleming, Robert Lee, Boris Feigin

Defensive Publications Series

We present a mechanism to generate a minidump file that will only include the exact memory an application wants when the application receives a fatal signal – the application itself can write a file in valid ELF CORE file format.

Real Estate Project Delivery System, Roshan Mehdizadeh Corsiglia

Defensive Publications Series

Organizations that have real estate holdings often spend substantial amounts on development of real estate for office spaces, production facilities, data centers, etc. Such organizations typically rely on third-party software solutions and consultants to manually track real-estate project costs and schedules. The present techniques provide a cloud based project delivery system that integrates the organization’s real-estate management applications with cloud based applications and building information models. With such integration, the techniques enable monitoring of project costs and schedules using a flexible user interface. The techniques can extract insights by integrating project cost, schedule, and attribute data with building information models …

Virtual Assistant Service For Patients And Caregivers

Defensive Publications Series

A major healthcare cost is the lack of strict adherence to, or outright non-compliance with, medicines and medical schedule by patients. The techniques disclosed herein assist patients by automatically performing routine healthcare-related tasks, e.g., reminding patients to take medications at appropriate times; ordering prescription refills at pharmacies; scheduling and managing medical appointments; tracking and charting their vital statistics, e.g., blood glucose, blood pressure, heart rate, etc.; asking for and logging patient symptoms, etc. The virtual personal assistant uses a machine learning model to adapt itself to the particulars of the patient. The virtual personal assistant may interface with the patient …

Assistant Enabled Entertainment System, Google Inc.

Defensive Publications Series

A virtual, intelligent, or computational assistant (e.g., also referred to simply as an “assistant”) is described that relies on supplemental data (e.g., contextual information, user information, etc.) to determine media, including movies, television shows, music, etc., that a user may want to consume and provide recommendations or automatically perform actions based on to assist the user in consuming the media. With explicit permission from a user, the assistant may access a user’s location history, calendar, e-mail, messages, past assistant interactions, contacts, photos, search history, sensor data, social network accounts, and other contextual or user information develop recommendations. The supplemental data …

Proactive Assistance For A Predicted Destination, Google Inc.

Defensive Publications Series

A virtual, intelligent, or computational assistant (e.g., also referred to simply as an “assistant”) is described that relies on supplemental data (e.g., contextual information, user information, etc.) to predict a user’s destination and offer to assist the user with actions the user will likely want to take at the predicted destination. With explicit permission from a user, the assistant may access a user’s location history, calendar, e-mail, messages, past assistant interactions, contacts, photos, search history, sensor data, and other contextual or user information to predict a destination of a user as well as actions the user will likely want to …

Assistant Generated Answers Using Supplemental Data, Google Inc.

Defensive Publications Series

A virtual, intelligent, or computational assistant (e.g., also referred to simply as an “assistant”) is described that relies on supplemental data (e.g., contextual information, user information, etc.) to decipher, and determine answers to, user queries. With explicit permission from a user, the assistant may access a user’s location history, calendar, e-mail, messages, past assistant interactions, contacts, photos, search history, sensor data, and other contextual or user information to answer a user query. The supplemental data can be stored locally on a device that is executing the assistant or in a cloud computing environment that is accessible to the assistant from …

Cable Vault Lids With Solar Panels, Eric Street

Defensive Publications Series

This disclosure describes techniques fiber optic cable vault lids with built-in solar panels. Vault lids are built with suitable materials based on deployment location. Power generated by the solar panels is used to power equipment such as Wi-Fi access points, power monitors, etc. within the fiber optic cable vault. Surplus power is supplied to the grid. By having power generation built-in, the fiber optic cable vault lid enable vaults to be deployed in locations where external power is unavailable. This improves the ability to communicate with fiber mid points, maintain health of the fiber network, and ease isolation of faults. …

Applying Machine Learning Techniques To Determine Product Risks

Defensive Publications Series

Generally, the present disclosure is directed to techniques to automatically determine risks associated with a product. In some implementations, the techniques of the present disclosure can include or otherwise leverage one or more machine-learned models to determine if the release and continued sales of product has legal, privacy and/or business vulnerabilities based on identifying sensitive keywords in product-related documentation and product areas.

This disclosure applies machine learning techniques to automate product audits and improve product review quality. Machine learning techniques are applied to proactively identify risks associated with a product, e.g., for a software product or service, the techniques are …

Detection Of Perceptually Dominant Colors In Images, Zachary Krejci

Defensive Publications Series

Detection of colors in an image is an important task in many contexts. For example, providers of website hosting services can recommend theme colors for a website based on a logo that is featured on the website. Prior color detection techniques have various shortcomings. For example, prior techniques often detect colors that lack chroma that are unsuitable as website theme colors; identify multiple perceptually similar colors; fail to detect a dominant color or detect colors that do not exist in the image. Further, the techniques make detection errors in the presence of image features such as shadows, anti-aliasing gradients, or …

Smart Voicemail With Action Triggers Based On Semantic Understanding, Robin Dua, Sujith Ravi

Defensive Publications Series

Techniques for responding to voicemails based on language processing of voicemails to identify actions that can be performed in response to the voicemails. The actions can be presented as selectable suggestion elements that, when selected, cause certain actions to be performed in reply to the voicemail. For example, a voicemail that includes a question for the recipient can be processed in order to generate selectable suggestion elements for answering the question via text message. Some techniques identify requests (e.g., request for a document, contact information, payment, etc.) from the recipient and, in response, queries can be performed in order to …

Intelligent Floating System, Daniel Hoppe

Defensive Publications Series

intelligent floating system

Use Of A High-Voltage Refrigeration Circuit For The Operation Of A Refrigerator / Freezer, Daniel Hoppe

Defensive Publications Series

USE OF A HIGH-VOLTAGE REFRIGERATION CIRCUIT FOR THE OPERATION OF A REFRIGERATOR / FREEZER

System For The Automatic Recognition Of Trailer Features For The Automatic Take-Over In Vehicle On-Board Computers, Daniel Hoppe

Defensive Publications Series

SYSTEM FOR THE AUTOMATIC RECOGNITION OF TRAILER FEATURES FOR THE AUTOMATIC TAKE-OVER INVEHICLE ON-BOARD COMPUTERS

Adaptive Ui Optimization Based On Automatic Detection Of Usage And Orientation Of A Mobile Phone, Sandro Feuz, Thomas Deselaers

Defensive Publications Series

A system and method are disclosed that enable adaptive user interface (UI) optimization by automatically detecting usage and orientation of a mobile phone. The method includes detecting how the phone is oriented (portrait/landscape orientation) and by which hand (right or left) the phone is held and by which hand it is operated by the user. The input data such as raw touch input data, raw gyro sensor data, context of the phone (active application) or context of the user (time of day) are collected from the phone sensors and fed to a machine learning (ML) model. The ML model infers …