Digital Commons Network^™

Measuring End-To-End Latency Of Fingerprint Authenticators, Jose Rodriguez, Firas Sammoura

Defensive Publications Series

The techniques of this disclosure enable accurate measurement of end-to-end latency of a fingerprint sensor, e.g., of a mobile device. A touch sensor in proximity to the fingerprint sensor is coupled to an oscilloscope. A photodiode in proximity to the screen of the device is coupled to another channel of the same oscilloscope. An accelerometer attached to the device to measure device vibrations, and is coupled to yet another channel of the same oscilloscope. The oscilloscope logs the signals generated by the touch sensor, the photodiode, and the accelerometer. An accept decision results in the brightening of the device screen …

Handcrafting Visual Features Of Emails Or Landing Pages To Detect Phishing, Kuntal Sengupta, Vijay Eranti

Defensive Publications Series

In a phishing attack, a perpetrator attempts to obtain the online credentials of a user by impersonating a trusted entity such as a bank, email service provider, etc. Sophisticated phishers attempt to deceive spam filters by structuring the visual look-and-feel of their fake emails to be nearly but not precisely identical to emails sent by a trusted entity, such that spam filters allow the fake email to reach a user’s inbox.

This disclosure describes use of hand-crafted visual features of emails or landing pages, and classification based on earth-mover’s distance, to assess the visual similarity of genuine and phished emails. …

Mixed Reality Interaction With Physical Keyboard, Anonymous

Defensive Publications Series

Performing certain operations when viewing an application via virtual reality devices, such as VR glasses, can be cumbersome. This disclosure utilizes a mixed reality display to provide application-specific virtual overlays for a physical keyboard. User interface elements of the application currently active are identified and displayed as virtual overlays atop a physical keyboard. The overlay matches the UI elements displayed by the application icons and allows the user to activate a UI element by simply pressing a physical key. The virtual overlay provides a contextual keyboard and can be utilized for fast navigation, accessing system functions, setting up shortcuts, playing …

Automatic Removal Of Artifacts In 360° Videos, Anonymous

Defensive Publications Series

Many 360°videos captured using a camera rig and tripod setup have artifacts related to the tripod that the rig is attached to. This disclosure describes automatic removal of such artifacts by use of machine learning techniques. Portions of the video that include artifacts are detected and removed automatically using trained machine learning model(s). The ML models are trained to recognize production equipment based on videos captured in neutral environments or based on markers on the production equipment. The detected artifacts are automatically replaced based on the context and ensuring that the resultant video is stable.

Projection Agnostic Quality Metric For 360° Images, Anonymous

Defensive Publications Series

Comparison of image quality for 360° images and videos that utilize different projections are difficult. This disclosure provides a reliable quality metric for 360° images/ videos, irrespective of the projection used. The 360° image for which the quality metric is to be computed is converted into a spherical representation. The surface of the sphere is peeled to produce overlapping loops that are rectangular in structure. A quality metric is determined by utilizing the block-based technique of SSIM to compute scores for blocks of the image and choosing a percentile based on the scores of the blocks of an image.

User Identification Across Online Providers, Anonymous

Defensive Publications Series

This disclosure describes construction of a user identification model based on confirmed user data obtained from different providers. The user identification model is usable by individual providers to provide missing pieces of user-specific data. An identification server maintains user data that includes globally unique user identifiers. The user data can be utilized to match user actions across different online providers and to perform analyses based on matched actions.

Selective Content Delivery Based On Content Emotion, Anonymous

Defensive Publications Series

Content items and associated criteria for delivery are received by a server from content provider. The server provides online services such as social networking, media hosting and sharing websites, news, etc. The server extracts content features from the content items using text extraction, image processing and other techniques. The server determines a content emotion associated with the content by applying trained machine-learning models. The server selects content items based on the content emotion, detected user emotion of a requesting user, and the criteria associated with different content items. Selected content items are delivered to a requesting user in a ranked …

Attribute Prediction Based On Online Connections, Anonymous

Defensive Publications Series

Techniques to predict attributes for a user or user group are described. User profiles and connections to other users are analyzed to determine user groups such as a household. Clusters are formed based on matching attributes of similar user groups. Available databases are queried using the user group identifiers to obtain a group-wide attribute value, e.g., household income, for a subset of user groups of a cluster. The obtained values are used to predict missing attribute values that were not found in the database, after adjusting for variations between user groups, such as a number of users in the group. …

User Emotion Based Selective Delivery Of Content, Anonymous

Defensive Publications Series

User interaction with a server that provides online services such as social networking, media hosting and sharing websites, news, etc. is detected. The user interaction and other data such as user profile information is analyzed using machine-learning models trained to detect emotion. Available content items, such as user-provided content items and sponsored content items, are matched to the detected emotion based on various criteria to select and deliver particular content items to deliver to the requesting user. The described techniques for selection and delivery of content items based on detected emotion enable delivery of content items that have a high …

Smart Text Tool For Capturing And Sharing Visual Content

Defensive Publications Series

A system, method and computer readable media to facilitate capturing and sharing images through a device application is disclosed. The application includes smart and AR text tools for automatically placing the text in the image and choosing the color or the background of the text based on one or more rules. The rules may be based on analysis of the image such as facial recognition, contrast detection, color detection and content detection. The application additionally includes an augmented reality text filter for text placement on a capture screen. The AR text tool allows user to add expressive motion elements to …

Automated Augmented Reality Context Stickers

Defensive Publications Series

The disclosure relates to a method, system and computer readable media for automatically placing relevant augmented reality (AR) stickers in user’s field of view on opening a device application. The method places dynamic location, weather or time stickers on the ground plane or above the field of view, once the user opens the camera and also during the capture flow of moments. The user may either flick or dismiss them, or press-and-hold the stickers to move them to a desired position and then capture. The method brings the most used 2D post-capture sticker learnings to life automatically. The method gives …

Capture Modes, Filters And Frames For Visual Content

Defensive Publications Series

A system, method, and computer readable media are disclosed for performing swipe operations to edit visual content. The device application may include a smart feature to provide a plurality of subtle filters and/or frames for easy post-capture use with a swipe. The swipe filters and/or frames may be customized based on geographic location or nationality. Subtle filters with skin smoothing or improved lighting effects and boost filters may be used. Frames may be based on a location or city, or time - a day of week or time of day or relevant holiday frame may be used. The filters enhance …

Tool For Measuring End To End Latency, Michal Kielan, Jose M. Gonzalez, Michal Godek

Defensive Publications Series

A system, method, and computer readable media for measuring end to end (e2e) latency in a video calling system is disclosed. The latency measuring system includes a video calling system having at least two user devices that may be connected over a network, one or more external cameras, a device producing luminescence and a computing device. The camera is placed in front of each of the user devices and is configured to record the video call in both the devices. The luminescence device may be turned on for a short moment during the recording of the video call. The system …

Streaming Rate Based Playback Speed Control For Live Video

Defensive Publications Series

A system, method, and computer readable media for automatically varying the playback speed of live media content to match the measured streaming rate is disclosed. The system includes one or more live streaming devices, host devices or server, and client playback devices connected through a network. The method includes running an application configured to display an interface for accessing the built-in camera functionalities and for posting live feeds to an online system at the broadcaster side and viewing the live stream on the client side. The system includes a smart speed control module configured to automatically limit the rate of …

Simulating Web Page Loading Performance, Patrick Hulce

Defensive Publications Series

Determining the load time for a web page in any specific environment can be challenging and slow owing to the diversity of devices and network types. This disclosure provides techniques to estimate the loading time of a web page in a given environment via simulation of a dependency graph constructed by tracking the page loading process in any available environment. The simulated traversal activates a node only when all of its dependencies in the graph have completed. The time taken for node completion in the simulation is varied depending on the corresponding estimate for the target environment in which the …

On-Device Analytics: Data Collection, Representation, And Feedback Loop, Vishal Murgai, Nikul Bhatt, Srinivas Addanki

Defensive Publications Series

Described herein are techniques for a Machine Learning (ML) model to learn from a training set and predict the cause of a failure accurately. Although it is challenging for a human to identify the root cause of an issue when configuration and show command outputs are very large (e.g., for thousands of Internet Protocol (IP) routes), troubleshooting on a network device typically has to rely on a manual process with the help of some show commands.

Advertising Software/Security Usage Descriptions With Policy Response, Eliot Lear, Brian Weis, Chris Steck, Nancy Cam-Winget

Defensive Publications Series

Techniques are provided for an organization-maintained server which takes three inputs: (1) a set of Uniform Resource Identifiers (URIs) from Internet of Things (IoT) devices, each of which point to a manifest; (2) a set of manifests resolved from the URIs; and (3) a set of threat feeds. The server periodically compares the vulnerabilities in the threat feeds to the manifests. When a vulnerability is found, steps are taken to protect the rest of the network from the vulnerable devices until they can be remediated.

Obfuscation And Anonymization Techniques For Network Data Sets For Machine Learning, Ralf Rantzau, Sebastian Jeuk, Gonzalo Salgueiro

Defensive Publications Series

Techniques are described herein for securing data used for a machine learning algorithm. The frequency or top-k values calculated over time of the respective network traffic feature data sets are used instead of the actual data or a set thereof (this can also be extended to any other data sets). Here, the frequency represents the actual data and thereby obfuscates potential sensitive information that should not be used within an oftentimes shared cloud machine learning application.

Provisioning Day-Zero Configurations Through Passive Radio-Frequency Identification, Santosh Patil, Shyam Vaidyanathan, Prashant Kumar, Manoj Gupta

Defensive Publications Series

Embodiments presented herein provide a mechanism for setting the day-zero configuration of a network device without having to power on or unbox the device. Using an embedded passive radio-frequency identification (RFID) tag situated in a device, the device can be programmed at a distance using a mobile device.

Visualization For Identifying Service Responses, Wagesh Kulkarni, Rakesh Sharma, Hitesh Manwar

Defensive Publications Series

A secure web gateway is a type of security solution that prevents unsecured traffic from entering an internal network of an organization. By translating static log data from a secure web gateway into a meaningful and sensible format, an end user may identify issues that may cause delayed responses from services. Incorporating a visualization of a health view of a system into web gateway software may provide clarity to end users. By binding logged data into five-minute intervals for a selected daily or weekly duration and displaying the data on a single screen, an end user may easily view the …

Identifying Ai Participants In Calls And Escalating Ai-Only Calls Into An Optimized Machine Communication Mode, Denis Savostiyanov, Rob Hanton

Defensive Publications Series

The embodiments presented herein enable two human-emulating artificial intelligence (AI) speech agents to identify each other in situations where the two are engaged in a conversation. Once two speech agents recognize each other as artificially intelligent, they may transition to a more efficient method of communication in order to complete tasks more quickly, accurately, and inexpensively.

Machine Learning Framework For Prioritizing Location Measurements Of Multiple Devices, Xu Zhang, Huy Tran, Matt Silverman, Abhishek Mukherji, Vinay Raghuram, Abhishek Bhattacharyya, Santosh Pandey

Defensive Publications Series

Presented herein is a framework for prioritizing location measurements of multiple client devices. In particular, rather than using a round robin scheduling approach, the techniques presented herein utilize a machine learning block (e.g., random forests) to predict a score for each client device, along with a score-based scheduler.

Method Of Passive Participant Tracking In Video Conference, Morgan Fang, Yasi Xi, Zhaocai Wang, Janus Wang

Defensive Publications Series

An innovative technique is proposed to highlight video streams of participants mentioned by the current active speaker in a video conference. By immediately highlighting mentioned participants, valuable visual information can be delivered and perceived by meeting attendees so that meeting productivity is improved.

Device History Recall Optimization With Virtual, Locally-Stored Acquisition Databases Using Previously Encountered Communication Access Points And Services, Jibing Wang, Qin Zhang, Srinivasa Prasad Vangaru

Defensive Publications Series

A mobile communication device of an end user (user equipment or “UE”) can use more than one radio access technology (“RAT”) such as GSM, WCDMA, TDS-CDMA, LTE, and NR. Searching for information needed by the UE to use a specific RAT can consume large amounts of power and time. Locally stored acquisition databases (“PLMN Info DBs”) of previously encountered communication access points and services can speed up searches, but can be limited by the storage capacity of the UE. Also a locally stored PLMN Info DB may be invalided when a UE moves to a location that is distant from …

Crowdsourced Market Research And Survey Completion Using Intelligent Conversational Assistants, Reena Singhal Lee

Defensive Publications Series

Insights into customer behavior are important for making business decisions. Yet current market research methods, e.g., surveys, multiple-choice questionnaires, etc., are long or complex, and have little or no incentive for customer participation. The quality and coverage of data that results from traditional surveys is sometimes found to be insufficient.

This disclosure leverages conversational assistants to conduct market research. At appropriate times, and with user permission, a conversational assistant asks a user questions relating to their everyday purchases, behaviors, or preferences. The questions may be longitudinal, audio-based, location-based, etc. For users that provide permission, user responses can be studied, without …

Lightfield Compression Using Commodity Hardware Video Codecs, Sai Deng, Pavel Krajcevski, Matthew Pharr, Stuart Abercrombie

Defensive Publications Series

A lightfield video is a series of lightfield stills, which include a number of reference views and residual data describing differences between the reference views and various viewpoints. The set of reference views or images, though different enough to provide the required basis for predicting intermediate views in the dataset, are correlated enough to approximate a traditional video stream. Additionally, reference images from a single time point may be compressed as a collage of smaller images that make up a single frame of a traditional video. Temporal correlation between reference images at successive time points is exploited, similar to standard …

Micro Touchpad For Wearables, Chun Yat Li, Guangzhe Cui, Debanjan Mukherjee

Defensive Publications Series

Users interact with electronic devices via a variety of input devices, such as buttons, touchpads, touchscreens, keyboards, styluses, and mice. Touchpads allow users to provide the electronic devices with a variety of commands and instructions because of their flexibility in receiving many types of input, including touches, swipes, gestures, taps, and holds from one or more fingers or other input devices. Touchpads also allow for robust construction and ergonomic variability in environments in which other input devices may not function as effectively. In particular, a micro touchpad sized for wearable devices, such as watches, activity trackers, or headphones, that retains …

Path Information-Driven Quic Path_Challenge To Optimize Path Selection In 5g Networks, Sebastian Jeuk, Gonzalo Salgueiro, David Hanes

Defensive Publications Series

Proposed herein are techniques that make use of an extended Quick UDP Internet Connection (QUIC) PATH_CHALLENGE frame to perform dynamic path selection operations within a 5G Wi-Fi/cellular network to optimize end-to-end performance. The techniques presented herein expands the capabilities of the current PATH_CHALLENGE function, as defined in the Internet Engineering Task Force (IETF) QUIC drafts. The additional information gathered as part of the PATH_CHALLENGE and the PATH_RESPONSE frame are used to understand path characteristics and dynamically optimize traffic flows accordingly.

Techniques For Checking As-Worn Alignment Of Eyewear With Display Optics Or Progressive Prescription Lenses, Ozan Cakmakci, John Stuart Fitch, Vladimir Krneta, David Hoffman

Defensive Publications Series

Fitting eyewear with display optics to a user’s head can be problematic because lenses of the eyewear are typically not on hand during product development. This can cause problems with alignment of eyeboxes of the lenses when the user wears the eyewear with finished display optics. Generally, the eyebox of a lens or display optic is a volume within the lens such that parameters of image quality criteria is met. For example, if the eyeboxes of the lenses do not align correctly with the user’s eyes, then the eyewear does not fit or align with the user’s head. Incorrect alignment …

Wifi Network Selection Based On Rssi Velocity, Rebecca Silberstein

Defensive Publications Series

A mobile device typically connects to a WiFi network based on received signal strength indicator (RSSI). In some situations, e.g., when passing an access point at a relatively high speed, the RSSI may be high enough to connect, but only for a short period of time.

This disclosure describes techniques to select a WiFi network based on RSSI and the rate of change of RSSI (RSSI velocity). If the RSSI velocity for a WiFi network is beyond a certain threshold, the WiFi connection is determined to be temporary, and the device does not connect to the WiFi network. WiFi connections …