Digital Commons Network^™

Two-Stage Machine Learning Model For Local Processing Of Frequent Queries, Vinod Das Krishnan

Defensive Publications Series

Virtual assistants implemented on a client device can recognize and fulfill spoken queries locally on device, e.g., using trained machine learning models. However, on-device query processing requires substantial processing power and memory. Heavy use of processor and/or memory resources can slow down the operation of the entire device, including queries for which processing is offloaded to an external server. This disclosure describes techniques to enhance the performance of local processing of voice based user queries via a two-stage serial pipeline of trained machine learning models. A first stage includes a limited-scope high efficiency model that can only recognize and fulfill …

Automatically Generating Variant Queries Of An Input Query Based On Input Mode, Matthew Sharifi, Victor Carbune

Defensive Publications Series

This disclosure describes techniques for translating an input query provided to a digital assistant to equivalent queries having a different input mode than the input query, such as typed or spoken input mode. A query provided by the user in one input mode is translated to variants for one or more other input modes based on grammars, user-permitted query logs, and/or machine learning models. Components of the digital assistant are updated with both the query and the equivalent query. The described techniques can be used to adapt digital assistant speech recognition and query response generation based on previous user queries, …

Adaptive Summarization Of Digital Assistant Responses To User Queries, Matthew Sharifi, Victor Carbune

Defensive Publications Series

This disclosure describes techniques for performing adaptive summarization of responses provided by a digital assistant to a user’s queries based on explicit and/or implicit feedback from the user that is indicative of their engagement with given responses. The feedback is determined based on user-permitted data from device sensors. If the user permits, past reactions of the user to output responses are used to train machine learning models or heuristics/rules that are used to determine the length of a query response appropriate for the user. Described features enable a digital assistant to automatically match the response length (duration) to the user’s …

Automatic Reporting Of Smartphone Privacy Risks, N/A

Defensive Publications Series

This publication describes techniques for the automatic reporting of privacy risks detected on a computing device to a user. Some risks may include content collected and shared by physical sensors (e.g., camera, microphone, location, activity sensors) with applications installed on a computing device. These applications may use information from the sensors to perform operations for the user (e.g., providing directions on a map application, sharing a photo on a social media application, reporting on the health of a user). A user may opt in to a Privacy Manager to help detect risks by ranking content collected by applications, for example, …

Quickly Monitor And Control All Physical Sensors, N/A

Defensive Publications Series

This publication describes techniques to turn on and off physical sensors of a computing device to protect user privacy. Users may be afforded a quick and easy way to turn on and off all physical sensors using a Privacy Manager that can be activated, for example, with a “sensor mute” icon located on the user interface or a physical switch located on the computing device. Users may monitor and/or customize the activity (e.g., on/off settings) of each sensor in a centralized control setting on the computing device. The Privacy Manager may identify each physical sensor of the computing device and …

Techniques For Passive Sharing Of Detailed User Status, N/A

Defensive Publications Series

This publication describes techniques for passively sharing a current, detailed status of a user (e.g., availability, what they are doing, how to best communicate with the user) with close contacts. A user opts in to a Status Manager and can enable customized sharing with individuals (e.g., family, friends). The Status Manager best determines the current status of the user by passively using contextual information, for example from the computing device sensors or associated applications. Utilizing this information, a notification can be displayed on the computing device (e.g., messaging applications, communication applications, on a launcher …

Metal Products Exhibiting Improved Mechanical Properties And Methods Of Making The Same, Samuel Robert Wagstaff

Defensive Publications Series

The present disclosure relates to metal stock and methods for producing metal stock, such as coils of metal strip, and more specifically to continuously cast recrystallized alloys containing dispersoids, as well as the processes for making such alloys.

System And Method For Predicting Hardware Failures Of Electronic Devices Using Onboard Sensors And Device Lifecycle Data, Dmitry Goloubew, Shi-Jie Wen, James Turman, Sumit Parimoo

Defensive Publications Series

Techniques herein provide a capability to predict failures of hardware by using onboard sensors and provide for the ability to move from detection to prediction for hardware failures. In turn, such techniques can help to reduce downtime due to marginal hardware and improves network availability. The techniques can also help to reduce unnecessary maintenance and changes related to replacing hardware that has not failed, which can lead to business efficiencies for both customers and vendors and may become even more important in a Network-As-A-Service (NAAS) context in which both sides are paid by the vendor.

Improving Malware Detection By Parsing Broken Code, Abhay Garg, Alexandre Ginet, Arijit De, Paneendra Ba, Yu Zhou

Defensive Publications Series

Spreadsheets, word processors, and other document editing applications enable users to write scripts or macros that automate a sequence of actions, e.g., keystrokes, mouse-clicks, etc. through code. Although macros can improve user efficiency by automating repetitive actions, executable code within a document can also potentially include malware. Macro-based malware is known to intentionally use broken syntax to bypass detection. This disclosure describes a parser that is resilient to syntax errors in code, and which can, by applying local corrections, continue to parse the rest of the code after encountering a parse error. Once corrected, the code can be subject to …

Media Presence Detection Using Light Emitter And Scanner, Hp Inc

Defensive Publications Series

This disclosure relates to the field of media presence detection in a sheet feed scanner.

A design is disclosed that a scan media presence detection is achieved by using a LED light, a light pipe

which conveys light to scanner, and scanner which scans with scanner light off.

Edlab: A Benchmark Tool For Edge Deep Learning Accelerators, Hp Inc

Defensive Publications Series

Plenty of edge deep learning accelerators are proposed to speed up the inference of deep learning algorithms on edge devices. However, Various edge deep learning accelerators feature different characteristics in terms of power and performance, which makes it a very challenging task to compare different accelerators according to their specifications and in turn prohibits a new DL model from being effectively and efficiently deployed on a suitable edge device. We introduce EDLAB, an edge deep learning accelerator benchmark tool, to evaluate the overall performance of edge deep learning accelerators. EDLAB is an end-to-end benchmark tool that provides unified workloads, deployment …

Emc Component Optimization, Hp Inc

Defensive Publications Series

This disclosure is to design a decoupling capacitor topology of IC’s in PCB layout to

effectively resolve EMI problem with minimum number of capacitors. The proposed

decoupling capacitor topology is to simulate the twisted wire principle which the

common mode ground noise is cancelled at ground when 2 current loops are in opposite

direction.

Determine the location of decoupling cap, we can control the current direction of the

IC’s since the decoupling cap would be the power source to the IC’s. By controlling the

current direction of 2 IC’s with opposite direction the common mode ground noise

would be cancelled …

Virtual Printing Expert Bot And Avatar System, Hp Inc

Defensive Publications Series

Digital printing presses are very sophisticated and consist of many features. As such, they are not easy to

operate. A virtual printing expert system was designed in order to assists the press operators personally,

and make their printing experience easy, smooth and fun, while ensuring high quality prints with high

utilization.

The system consists of a virtual figure, with: “eyes” (video camera), “voice” (speaker) “ears”

(microphone), “brain” (analytics engine, Deep Learning models, cloud and on-premise AI services) and

personality.

Like a human, the virtual expert can talk & listen, write & read, watch, recognize, ask & answer

questions, tip, direct …

Wall Ironing During Fused Filament Fabrication, Jiri Konvicny, Jaime Van Kessel

Defensive Publications Series

This idea aims to increase the 3D printed objects strength in Z-axis by ironing each layer of the printed walls of the objects.

A Dynamic Role Switching Approach To Enforce Principle Of Least Privilege, Hp Inc

Defensive Publications Series

This document issues a method to consider adding a proxy module that can be dynamically switched to

different granted roles in order to decouple different business functional requests and to be executed under the

least permissions. This document designs this module as an independent component in the example system

architecture, but this disclosure will not be limited by this architecture design.

Compilation Of Media Consumption For User Enjoyment, N/A

Defensive Publications Series

This publication describes techniques for compiling a history of media (e.g., movies, songs, shows) consumed by a user and providing the user with access to their media history. In aspects, a Media Manager, which identifies media as it is consumed by the user, is provided on a computing device of the user. The Media Manager may use information collected by physical sensors (e.g., microphone, camera) on the computing device to identify the media and may store information about the media (e.g., title, date consumed) on the computing device. The Media Manager may provide the user with information relating to media …

Concurrency-Aware Peer-To-Peer Negotiation On Wireless Devices, Omar El Ayach, Nihar Jindal, Kumar Anand

Defensive Publications Series

This publication describes techniques of optimally establishing peer-to-peer connections between a group of wireless devices. In aspects, the techniques are performed by a Connection Manager implemented on one or more of the wireless devices. The Connection Manager utilizes a selection process to select a wireless device from the group to act as the group owner of the peer-to-peer network. Through the selection process, the Connection Manager selects the wireless device that is most suited to become the group owner. For example, the Connection Manager may select as group owner the wireless device with the highest probability of providing optimal data-exchange …

Temporary Screen Timeout Extension Based On User Intentions, Alex Salo, Tomer Shekel

Defensive Publications Series

This publication describes techniques to provide a temporary screen timeout extension for displays on electronic devices. The timeout extension is prompted based on a Screen Manager detecting activity that indicates a user may desire to keep the screen active. The Screen Manager may give the user an option to extend the screen timeout past the default timeout setting, or it may extend the timeout automatically. The default timeout setting does not change when the temporary screen timeout extension is chosen. The extension allows the user to continue to use the device periodically without needing regular inputs to the device in …

Root Cause Identification, John Garrett, Fang Chen, Adrian Bjune

Defensive Publications Series

Presented herein is a novel machine learning approach that learns the failure patterns of a device and uses this information to drive automatic root case identification. Anomalous events can be associated with potential root causes in order to build intellectual capital for training future predictive automated remediation systems.

Fasttrack Recommender System, Adam Jannetta, John Garrett, Ammar Rayes, Altaf Karim

Defensive Publications Series

Network operators are overloaded with numerous recommendations coming from vendors, some of which come from automated recommender systems. Such automated recommendations may or may not apply to a customer’s specific environment, often lack an assessment of priority within the context of the other recommendations, and may or may not apply to an individual customer’s scenario. To address these challenges, techniques are presented herein that provide a novel approach to generating and ranking recommendations coming from a dynamic recommender system where rankings are based on enriched context from, for example, live data on real networks, activities performed by real customers, etc. …

Route Selection Based On Sunlight Exposure, Joseph Turner

Defensive Publications Series

Digital maps provided via websites, mobile apps, in-car navigation systems, etc. include features to provide route guidance. Currently, such maps compute routes based on parameters such as travel time, cost, transport mode, type of roadway, etc. This disclosure describes techniques that enable digital map applications to provide routing information based on expected exposure to sunlight along the route. The sunlight exposure information can be displayed to users and utilized in route selection to minimize or maximize sunlight exposure depending on travel parameters. For example, higher sunlight exposure may be preferred by users who drive vehicles with solar panels while lower …

Multi-Rail Integration In Vertical Power Delivery, N/A

Defensive Publications Series

Vertical power delivery (VPWR) is a layout used in high-performance semiconductor electronics to reduce power losses along the power delivery network. In vertical power delivery, the voltage regulator module is mounted directly below the processing unit to which it supplies power. Although vertical power delivery reduces power losses and enables flexible PCB floorplans, it can potentially trade off efficiency with increasing power density, and due to keep-out zones for decoupling capacitors (decaps), it can degrade transient power-rail performance of nearby power rails.

This disclosure describes techniques that integrate multiple power rails into a single VPWR module. To accommodate non-core power …

Sensemaking For Broad Topics Via Automated Extraction And Recursive Search, Alankar Jain, William M Leszczuk, Sitaram Iyer, Mehrbod Sharifi, Seher Aylin Altiok

Defensive Publications Series

The availability of vast amounts of diverse information related to a broad topic makes it difficult and time-consuming for users to find and digest the right information regarding various low-level topics within the broader space. Current approaches to addressing these challenges include providing curated topical pages, relevant query refinement suggestions, list of subtopics, etc. However, these approaches do not scale and offer inadequate support for sensemaking. This disclosure describes automated techniques that extract information from online information sources by using a query related to a high-level topic to recursively formulate additional queries for subtopics to construct a hierarchical set of …

Method To Secure Telemetry Logs In A Multi-Tenant Environment, Robert Barton, Vinay Saini, Cesar Obediente, Jerome Henry, Amit Singh

Defensive Publications Series

Modern service/hosting models utilize the concept of a multi-tenant infrastructure to share common network services while providing secure logical network segmentation between tenant customers. An on-going problem with network telemetry tools in multi-tenant and shared environments is that often they do not provide data confidentiality between tenants. Presented herein are techniques in which an access policy to telemetry logs can be generated in a multi-tenant environment through the use of role-based policy tokens that are attached to log entries as they are received at a collector.

Use Of Temperature In Dynamic Knowledge Distillation For Joint Optimization Model, Anonymous

Defensive Publications Series

In a joint optimization model, information from a large, complex teacher model is transported to small, light student models using knowledge distillation. Dynamic knowledge distillation allows the student models to learn from the teacher model on the fly. However, the performance of a joint optimization model that uses dynamic knowledge distillation suffers if the teacher model contains too much noise from the negative labels, or does not have enough information from the negative labels. This disclosure describes techniques to implement dynamic knowledge distillation by using temperature to control the amount of information transmission about negative labels from a teacher model …

Pics Tread Depth Measurement (Ptdm) Next Smartdiag Features, Philippe Lefaure

Defensive Publications Series

No abstract provided.

On-Device Query Cache For Digital Assistant, Xin Li, Chen Zhang

Defensive Publications Series

There may be failures or delays in query fulfillment when a digital assistant utilizes server-side query processing to interpret a query and provide an appropriate response. The delay can be worse in certain situations, e.g., a poor network connection when a user places queries while in a moving vehicle. This disclosure leverages the observation that many queries are repeated verbatim by users. The described techniques, with user permission, utilize an on-device query cache to reduce latency in query fulfillment. With user permission, spoken commands are matched with prior queries stored in the on-device cache and if a match is found, …

Speech Recognition Correction Based On Detected Topic In Speech, Victor Carbune, Matthew Sharifi

Defensive Publications Series

This disclosure describes techniques for adjusting speech recognition of a voice dictation based on topics that are detected from the speech during and after the dictation. A topic of the speech content is classified based on recognized speech. Speech recognition of further speech is adjusted to the detected topic, and text that was previously-recognized during the dictation is re-recognized and corrected using the adjustment. Described features can improve accuracy of text transcripts obtained automatically from dictation, reducing the amount of manual corrections required for recognized text.

Small Form Factor Wide Field Of View Augmented Reality Head Mounted Display, Punarjeewa Abeysekera

Defensive Publications Series

In this disclosure, a method to create a wide field of view augmented reality head-mounted display which has a small form factor is disclosed. Two transparent video displays are used as a front-facing main display and two other transparent video displays are used from the left and right sides of the front-facing main display slightly tilted. A beam splitter is used to combine the real-world view and the computer-generated view. Two overhead displays are used with two convex lenses to project two images into the beam splitter. The result is a small form factor augmented reality head-mounted display with a …

Http Restful Api Design Using Freeform Url To Interact With Hierarchical Json Data Stores, Santosh Nair, Sachin Yerrawar

Defensive Publications Series

Techniques are presented herein that support a Hypertext Transfer Protocol (HTTP) Representational State Transfer (RESTful) Application Programming Interface (API) employing a freeform Uniform Resource Locator (URL) to interact with a hierarchical JavaScript Object Notation (JSON) data store. Additionally, aspects of the techniques presented herein incorporate a smart caching strategy to, for example, optimize cache size, support reads and writes to an underlying database, and reduce cache misses.