Digital Commons Network^™

Method For Filtering Displayed Content Based On Occupants, Scott Randolph, Bradley Stenning, Steve Paik

Defensive Publications Series

This publication describes an operating system (OS) that re-defines a user as a sum of all occupants in a shared space, and not merely as a single individual who has signed into a shared device. To identify the occupants of a space, the OS uses objective markers, such as: facial recognition, radar signature, biometric sensors, media address control identification (MAC ID), voice recognition, various sensors, radio-frequency identification (RFID), or other Internet-of-Things (IoT) data. Then, the OS filters out the displayed content on the shared device based on a level of trust amongst the occupants and a level of sensitivity. To …

Determining Breakpoints In Media Content, Migle Padegimaite, Sammy El Ghazzal

Defensive Publications Series

Systems and methods for determining breakpoints in media content via a machine learning architecture by accounting for the context of the media content, and for placing separate pieces of additional content into the breakpoints, are described. The machine learning architecture determines the number of pieces of additional content to insert into the media content, divides the media content into a number of scenes and tags each scene with appropriate identifiers based on comparison to reference content, identifies break points at the end of each scene, identifies which break point to insert pieces of additional content into, and then inserts the …

Method For Filtering Displayed Content Based On Occupants, Scott Randolph, Bradley Stenning, Steve Steve

Defensive Publications Series

This publication describes an operating system (OS) that re-defines a user as a sum of all occupants in a shared space, and not merely as a single individual who has signed into a shared device. To identify the occupants of a space, the OS uses objective markers, such as: facial recognition, radar signature, biometric sensors, media address control identification (MAC ID), voice recognition, various sensors, radio-frequency identification (RFID), or other Internet-of-Things (IoT) data. Then, the OS filters out the displayed content on the shared device based on a level of trust amongst the occupants and a level of sensitivity. To …

Automating Last-Mile Power Connection Discovery In Datacenters, N/A

Defensive Publications Series

Up-to-date records of last mile power connection information in a datacenter are useful, e.g., to enable data center engineers to reason about the power-network dependency of machines; to monitor aggregate power consumed by a group of machines; etc. Such records are typically obtained manually, e.g., by associating barcodes with both distribution equipment (junction boxes) and power loads (racks) and scanning the barcodes to establish distribution-to-load mapping. The present manual methods of power-connection discovery are labor-intensive.

This disclosure describes automated discovery of last-mile power connection information in a datacenter. Power cables (whips) that connect racks to junction boxes are made available …

Translation Of Electronic Dialogs

Defensive Publications Series

Techniques for translation of all or subsets of an electronic communication sent by a first user (via a corresponding client device) to one or more additional users (for consumption via corresponding client device(s)). The translation is from a source language of the electronic communication to target language(s) of the additional user(s). For example, assume a first user creates an electronic communication in Chinese and transmits the electronic communication to a second user who only understands English. Techniques described herein may translate all or portions of the communication from Chinese to English and present (e.g., visually and/or audibly) the translated English …

Automatic Dubbing Of Videos With Multiple Speakers, Sandro Feuz, Mohammadamin Barekatain

Defensive Publications Series

A machine-learning model that automatically converts audio streams from an audio-visual content from a source language to a destination language is described. In response to determining that an audio stream should be translated, a machine-learning-based dubbing model is invoked for a specific destination language. In case of multiple speakers, voice embedding techniques are used to match dubbed audio streams to the corresponding speakers. The sentiment in the original speaker’s voice is preserved by training the model with targeted data set in the destination language.

Securing Private Images On User Devices, Linden Evans, Laura Paragano

Defensive Publications Series

This disclosure describes techniques to secure and remove private images (e.g., photos) from display on user devices. With user permission, a device examines stored user images for private content and relocates detected private images to a locked album on the device. The private album can include user-adjustable settings to control the organization and distribution of private images. Such features enable the device to automatically seclude private images, thus removing such images from potential undesirable discovery by other users that may use the device or other devices that access a collection of images on the device.

Systems And Methods For Predicting Online Video Metrics, Aaron Shon, Charles Johan Larsson Tibell, Axel Erik Olov Henriksson

Defensive Publications Series

A low-latency reporting mechanism that runs as a layer on top of an analytics engine is described. The analytics engine runs a predictive model to generate one or more metrics related to online content usage in near future. The predictive model is based on artificial intelligence and machine learning. The predictive model can be trained with low-latency real-time event data as well as canonical data obtained from historical event logs.

Machine-Learning Of Leading Indicators From A Plurality Of Services To Perform Memory Management On Another Plurality Of Services With Correlating Memory Usage Profile, Jayesh Wadikar, Pok Wong, Ajay Madhavan, Smruti Lele

Defensive Publications Series

Presented herein are techniques for machine-learning of leading indicators from a plurality of services and the use of these leading indicators to perform memory management operations on another plurality of services with correlating memory usage profiles. The techniques presented herein include collecting, as leading indicators, service event logs from a plurality of services. Machine-learning is then used to cross-correlate the leading indicators with memory use pattern snapshots at subsequent times for another plurality of services to predict the optimal when, how much, and how for execution of memory garbage collection on the latter set of services.

Machine Learning Based Technique To Dynamically Allocate A Network Slice Based On User Equipment Type And Network Slice Selection Function Factor For 5g Wireless Internet Of Things Networks, Akram Sheriff, Santosh Patil, M. David Hanes, Gonzalo Salgueiro

Defensive Publications Series

Techniques are described for dynamically allocating different Long Term Evolution (LTE) User Equipment (UE) variants (e.g., Cat-M1 and Narrowband Internet of Things (NB-IOT) UEs) to the correct Network Slice. This may be accomplished by using a Machine Learning (ML) based software component running on 4G LTE evolved Node B (eNB) or in 5G next generation NodeB (gNB).

Dynamic Re-Calibration Of Location Using Active Sensors And Behavioral Analytics, Robert Barton, Jerome Henry, Vishal Desai

Defensive Publications Series

Techniques are described for using Wi-Fi® active sensors where the exact location is known. A Real-Time Location System (RTLS) compares a predicted location of devices with their actual location and measures the variance. Variance data is compared with the Radio Frequency (RF) metrics reported by the sensors at the time of measurement such that a regression model is created. Thus, based on the fully trained model, a correction factor for the RTLS algorithm may be determined at any given time and place in the Wireless Local Area Network (WLAN) to improve location accuracy.

Virtual Reality In Nuisance Protection, Anonymous

Defensive Publications Series

In the current world, online presence of individuals is continuously evolving. A host of applications exist today that are used by individuals for a variety of purposes that include lifestyle, online shopping, maintaining social presence, travel and commute, healthcare, and many more. With advancements in Virtual Reality (VR), interactions have become incredibly realistic. VR finds implementation in multiple applications like social media, gaming, online shopping experience, virtual sight-seeing tours, cinema, web pages, etc. As may be true for every technology, VR also has some drawbacks. In this disclosure, we address one such drawback of VR namely trolls, which comprises of …

Hand Of The Invisible Poet, Anonymous

Defensive Publications Series

The present disclosure describes a method for automatically superimposing content of a user’s preference on targeted portions of images captured by the user in real time. The targeted portions to display the content, also known as secondary spaces, are detected via heat maps, which are based on the user’s focus on the screen. The secondary spaces are regions, where the user’s attention is least centered, that is, the regions, where the color density is least. Moreover, the method provides the user with the content depending upon the user’s location, so that the user can relate and make use of the …

Motion Sickness Mitigation Using Vr/Ar Devices, Anonymous

Defensive Publications Series

Motion sickness is a common problem experienced by individuals travelling in enclosed vehicles. Such condition occurs in a ship/boat encountering rough waves (sea-sickness), or a land vehicle going through a rugged terrain, in aircraft, etc. It happens when there is a mismatch between the motion perceived by the human vestibular system in the ear and the motion seen by the eyes. The interior of a ship/motor vehicle is visually observed by person to be stationary, while the vestibular system (which is a fluid filled structure inside the ear) still perceives the motion of the ship/vehicle. The present disclosure aims to …

Internet Protocol Version 6 Prefix Coloring In Software Defined Access Fabric For Differentiated Policy Enforcement, Sri Gundavelli, Shree Murthy, Sudhir Jain, Indermeet Gandhi

Defensive Publications Series

Techniques are described herein for Internet Protocol version 6 (IPv6) prefix coloring in Software Defined Access (SDA) fabric. These techniques may enable differentiated policy enforcement.

Gesture Text Input, Anonymous Anonymous

Defensive Publications Series

The present publication describes a system that allows users to generate one or more emojis based on gesture inputs, while communicating using a computing device. The computing device may include a mobile phone, tablet, PC, or any other relevant computing device that can be used for communication. In one example, the communication takes place on an Instant Messaging (IM) application (app) running on a mobile device.

Access Point Transmission Optimizations Using Machine Learning Based Traffic Classification, Vincent Cuissard, Amine Choukir, Domenico Ficara, Salvatore Valenza

Defensive Publications Series

Techniques are described herein for optimizing access point transmissions. According to these techniques, configuration, state, and tuning of an access point transmission scheduler are externalized to a network core (e.g., a wireless LAN controller (WLC) or Dynamic Network Access Control (DNAC)) and a machine-learning system is used to classify wireless clients based on traffic analysis. The described techniques permit the use of fine-tuned configurations for each access point station and these configurations can be shared across all access points in the case of roaming.

Content Provenance Attribution And Verification Via Blockchain Certification, Daniel Golovin, Daniel Klein

Defensive Publications Series

Content such as images is sometimes used online without proper permission or attribution by parties other than content creators and owners. Prior to use, such parties may also alter the content as well as associated metadata. Such practices subvert rights of the content creators and owners and can mislead readers due to a lack of cues that can help determine the provenance, veracity, and credibility of the content.

This disclosure describes a mechanism to track, verify, and attribute provenance to content such as images. Per techniques of this disclosure, provenance information signed with a user’s private key is included in …

Enhanced Electronic Whiteboard, David Tigges

Defensive Publications Series

This disclosure describes an enhanced electronic whiteboard for visualization of cloud-based computing solutions. The electronic whiteboard includes technical icons that represent cloud computing solution elements and enables users to depict and manipulate various computer architectures. The electronic whiteboard is integrated with cloud migration solution software for collaborative design, sales presentations, and simulation. Depicting technical architectures is made possible by providing icons for servers, storage, networking appliances, etc. that can be dragged and dropped onto the whiteboard. With user permission, machine learning techniques are utilized to auto-populate operating data relevant to different architectures.

Enhancing User Interface Accessibility Using Computer Vision, Juhyun Lee

Defensive Publications Series

Accessibility features for vision-impaired users are typically provided by oral readouts of on-screen icons or other material. The task of transcribing in-app icons to text for text-to-speech conversion is typically left to the app developer. Icon-to-speech translation for apps from developers that are not conscious of the need to make their products accessible is often unsatisfactory. For example, a refresh button and a back button may both simply be transcribed as “button,” which is unhelpful to a vision-impaired user.

This disclosure uses computer vision techniques to automatically infer, and provide to a vision-impaired user, text corresponding to UI elements, e.g., …

Automated End-To-End Instrumentation And Testing Of Mobile Device Camera, Yu Zhong, Gregory Eitzmann, Kenneth Goldman

Defensive Publications Series

Due to reasons of security, mobile device operating systems generally allow the camera buffer to be written only by the camera - the camera buffer cannot be filled with mock frames. Due to write-restrictions on the camera buffer, e.g., technical infeasibility of feeding arbitrary, repeatable frame sequences to the camera buffer, apps that utilize device cameras such as photography apps, computer vision apps, etc. are currently tested manually.

This disclosure presents a test framework to instrument and automatically conduct end-to-end testing of apps that use a mobile camera as an input source. Under the test framework, the mobile device is …

Visualizing Classifier Performance, Lori Levy, John Burge

Defensive Publications Series

Classifiers are characterized by many measures, e.g., precision, recall, accuracy, sensitivity, specificity, receiver operating curve (ROC), fall-out, class imbalance, etc. Some measures are synonymous, e.g., sensitivity and recall. The profusion of terms causes some difficulties.

This disclosure leverages the observation that most classifier performance metrics derive from just four elemental quantities, e.g., true positives, false positives, true negatives, and false negatives. Disclosed is a simple graphic that displays the four elemental quantities and enables quick, visual, estimation of the remaining measures of classifier performance.

Event Scheduling And Time Zone Management For Online Calendars, Steffen Meschkat

Defensive Publications Series

The present disclosure relates to a calendar system and related method for determining the time zone associated with an event and generating a calendar that shows multiple time zones corresponding to events in the calendar. The calendar system can include a computing system that receives event data including information associated with an event and an event location. The event data can be used to determine a time zone associated with the event, which can in turn be used to generate an output including a calendar showing, at the same time, different time zones and the events that occur in those …

A Reinforcement Learning (Rl) Model For Dynamic Optimization Of Switch Port/Queue Buffer Allocation, Imran Pasha

Defensive Publications Series

Techniques are described herein for a reinforcement learning (RL) model for dynamic optimization of switch port/queue buffer allocation. According to the described techniques, a trained neural network model can be installed on a switch or on a network management server to dynamically adjust the shared/dedicated buffer allocation for the low/high priority queues in case of a frame loss. The dynamic buffer adjustment continues till the port/queue no longer experience a frame loss.

System For Supplying High Shear Starch To A Gypsum Board Machine And Methods Of Manufacturing Gypsum Board Using High Shear Starch, National Gypsum Company

Defensive Publications Series

The present disclosure provides systems for supplying high shear starch to a gypsum board machine and methods of manufacturing gypsum board using high shear starch. These systems and methods may improve the effectiveness and efficiency of starch utilized as a gypsum core binder and may further improve the core qualities of wallboard.

All or a portion of the starch used in the gypsum board manufacturing process may be high shear starch. The starch may be sheared continuously and fed to the slurry in the pin mixer. The high shear starch may be produced by feeding desired amounts of dry starch, …

Rationalized Clustering In Machine Learning, Diveplane Corporation, Christopher Hazard Phd

Defensive Publications Series

[0001] In machine learning, there are currently debates about what an explanation or explainable model is and what is necessary for a given purpose. This post details the concepts of explanation and interpretation to help clarify the difference between the two; discusses how, although interpretation is preferable, explanation is the only option for many machine learning techniques; and then details a clustering technique that aids explanation for unsupervised machine learning.

Rationalized Clustering In Machine Learning, Diveplane Corporation

Defensive Publications Series

In machine learning, there are currently debates about what an explanation or explainable model is and what is necessary for a given purpose. This post details the concepts of explanation and interpretation to help clarify the difference between the two; discusses how, although interpretation is preferable, explanation is the only option for many machine learning techniques; and then details a clustering technique that aids explanation for unsupervised machine learning.

Wireless Network Load Corroboration Using Machine Learning (Ml) Based Video Analytics, Smruti Lele, Samer Salam, Ajay Madhavan, Jayesh Wadikar

Defensive Publications Series

Presented herein are techniques for correlating the output of a crowd counting machine learning (ML) algorithm, which operates on surveillance video, with observed network load to determine if a load spike is due to a valid network usage or an attacker trying to sabotage the network. The techniques presented herein include vision field classification based on access point (AP) coverage, linking of vision fields to AP coverage in DNAC UI, and consensus-based threat assessment and alerts.

Automated Client Classification And Policy Enforcement Through Cloud Application Programming Interface Gateway, Xueqiang Ma, Kumar Sankaralingam, Shubhojit Chattopadhyay, Kalyan Ghosh

Defensive Publications Series

Techniques are described for identifying key parameters for classifying interactive clients and daemon clients. Machine learning algorithms based on Principal Component Analysis (PCA) and enhanced Decision Tree (DT) are utilized together with HyperText Transfer Protocol (HTTP) header analysis to accurately classify clients. Rate-limiting effects are monitored as a feedback to perform re-enforcement learning to improve accuracy of the algorithm. The Application Programming Interface (API) gateway is one of the most important components to any cloud service and it needs to apply different rate-limiting policies on API requests based on different requirements of different types of clients. Traditional client identification with …

Retrospective Ads To Obtain User Feedback, Tuna Toksoz, John Dukellis

Defensive Publications Series

Software application developers seek user feedback about their apps. In particular, app developers are interested in learning why users stop using their apps. However, if a user leaves, e.g., uninstalls an app, it is difficult to bring such users back to the app, even for the purposes of feedback.

With user permission, the techniques of this disclosure use an identity service provider to associate a user with an app. After a user leaves an app, an ad network is utilized to reach the user via other apps to obtain user feedback regarding the uninstalled app via a feedback form. User …