Digital Commons Network^™

Adaptive Reasoning And Evaluation Framework For Multi-Agent Intelligent Systems In Debate-Driven Decision-Making, Subhadip Mitra

Defensive Publications Series

Collective decisions require debate to surface contrastive points of view and to enable balanced and well-informed decision-making. Large language models (LLMs) can provide an opportunity to formulate deliberative decision making through multiple artificial intelligence (AI) agents. However, existing LLMs struggle with complex group discussions. This disclosure describes an artificial intelligence framework and techniques to support complex debate scenarios and group decision-making. A tiered structure of language models is deployed to dynamically generate and evaluate arguments. The techniques can handle multi-participant debates while considering ethical implications, producing well-reasoned outcomes. The described framework addresses limitations in existing debate systems by offering improved …

Reliable Multi-Tenant Solid State Drives With Flexible Write Throughput Sharing, N/A

Defensive Publications Series

Solid state drives (SSDs) that are shared among multiple tenants struggle to balance guaranteed throughput, burst capabilities, and power loss protection (PLP). Existing solutions either compromise fairness, limit burst throughput, or sacrifice data integrity during power outages. This disclosure describes techniques to share SSD write throughput between multiple tenants in a way that does not compromise PLP support by the SSD. This is accomplished through dynamic allocation of the SSD write cache to the individual tenants upon need. Tenants can be provided with a guaranteed share of the total SSD write throughput. A subset of the tenants can be allowed …

Mixed-Precision Quantization For Machine Unlearning Through Contrastive Learning, Gaowen Liu, Yuguang Yao, Yihua Zhang, Charles Fleming, Ramana Kompella

Defensive Publications Series

Machine learning models (e.g., generative models) are useful for automating many tasks, including generating content (e.g., text, image, video, etc.). However, outputs generated by these models may include undesirable content that reflects biases and stereotypes. Thus, machine unlearning approaches have been leveraged to remove bias in models. Conventional techniques for removing bias from models typically rely on contrastive learning to learn unbiased models, but these techniques generate full-precision models that require significant resources for implementation and deployment. To address these issues, techniques proposed herein provide mixed-precision quantization for machine unlearning through contrastive learning, thus facilitating the implementation of a quantized …

Unified Robotic Benchmarking Platform For Augmented Or Virtual Reality, Yang Chen, Narges Noori, Chao Guo, Chen Lu, Luke Jia, Chris Kenyon, Sean Yuan

Defensive Publications Series

Capturing and quantifying the performance of augmented reality (AR) or virtual reality (VR) devices during test and development has traditionally been a labor-intensive and subjective task undertaken by human testers. This disclosure describes robot-based techniques of testing AR/VR devices. By integrating robotic arms, computer vision, speakers, displays, and microphones into a unified testing suite, the techniques address the challenges of labor-intensiveness, variability, and uncertainty of traditional AR/VR testing. The techniques provide objective and consistent ground-truth data, enabling precise measurement of both hardware and software performance. The techniques can enable AR/VR developers and manufacturers to make data-driven decisions to optimize their …

Automated Password Change Using A Large Language Model For Website Navigation, Ece Scheuss, Fiorella Barrientos Villalta, Vasilii Sukhanov, Viktor Semeniuk

Defensive Publications Series

Some web browsers include password-check features that indicate insecure credentials, e.g., credentials that may have been leaked to third parties. However, users may not change such credentials, potentially leaving their accounts vulnerable. This disclosure describes techniques to automate password change, such that users can stay safe online without having to undertake manual effort to change their passwords with websites or applications. Multiple entry points are offered to users to enable them to start the automated password-change flow. With user permission, the web browser navigates to a known password-change website; auto-generates a strong, new password; fills in a password-change form with …

Reduce Delays In Ai-Powered Calls, Liy Opor

Defensive Publications Series

Traditional AI-generated responses in sales calls often face significant delays, typically 4+ seconds. This includes 700ms for Speech-to-Text, 2 seconds for AI response generation, and 400ms for Text-to-Speech. Using Retrieval Augmented Generation can extend this to 5-7 seconds, leading to customer dissatisfaction. To address this, we've introduced technical solutions to reduce delays. Using GPT-4 streaming mode and sentence-level TTS can cut response time by about 1 second. Concurrent matching with existing responses can further reduce time. If a match is found, a pre-recorded voice response is delivered immediately. If not, transitional words buy time for GPT-4 to generate a response, …

Security And Privacy Risks Of Multiple Payment Systems, Miguel Silva

Defensive Publications Series

The proliferation of multiple payment systems has introduced significant security and privacy risks, as users navigate a complex landscape of digital transactions. These risks arise from vulnerabilities in system integration, data breaches, and insufficient encryption protocols. Additionally, the lack of standardized security practices across platforms exacerbates the challenges of ensuring user privacy. Addressing these concerns requires comprehensive solutions that prioritize robust encryption, secure data sharing, and unified regulatory frameworks.

Joint Training Of Multiple Neural Networks, Urvang Joshi, Debargha Mukherjee, In Suk Chong, Akshaya Purohit, Shan Li, Innfam Yoo, Feng Yang

Defensive Publications Series

This paper describes a technique for jointly training multiple neural network models to handle complex and diverse data distributions. The technique breaks up data into separate classes, with each model focusing on a specific subset while using a classifier-based architecture with temperature-controlled Softmax. The training process gradually transitions from uniform model contribution to specialized model selection, combining predictions through weighted summation. This enables effective distribution of different data patterns across multiple smaller neural networks while maintaining inference efficiency through single model selection. The technique is particularly valuable in scenarios where data exhibits significant variations that would otherwise require extremely large …

The Composite Light Capturing And Projection System That Contains 16 Light Capturing And Projection Systems Where The Default Image Projection Of The Scene Will Be Structured Into 16 Sub Image Projections., Punarjeewa Abeysekera

Defensive Publications Series

This paper describes a composite light capturing and projection system that contains 16 light capturing and projection configurations where the default image projection of the scene will be structured into 16 sub image projections. The composite light capturing and projection system consists of 16 light capturing and projection configurations. The entire composite light capturing and projection system will consists of 5 main entity types. They are the generally utilized concave mirror section, the generally utilized beam splitter, the generally utilized primary reflector, the generally utilized short light projection sub configuration and the generally utilized long light projection sub configuration. The …

Face Id And Roi Detection Enhanced Intelligent Auto White Balancing And Dynamic Color Tuning Applications In Cameras Using Isp Edge Ai Processing, Hp Inc

Defensive Publications Series

Dynamic Color Tuning as Auto White Balance in cameras is a critical feature for ensuring accurate color reproduction across varying lighting conditions. This paper presents an innovative method for AI based enhancement of Dynamic Color Tuning in (PC) Cameras by integrating real-time face ID and ROI (Region of Interest) detection via ISP (Image Signal Processor), involving ISP AI on edge, to improve and dynamically customize existing Auto White Balancing of Camera image for each user or camera subject. Using data trained AI models, each individual user’s face skin tone/complexion, reflective properties of skin and other facial characteristics are to be …

Solid-State Forms Of Resmetirom And Processes For Preparation Thereof, Msn Laboratories Private Limited, R&D Center; Srinivasan Thirumalai Rajan; Sagyam Rajeshwar Reddy; Kammari Balraju; Garai Abhijit; Edulakanti Jhansi And Dodle Beerappa.

Defensive Publications Series

The present disclosure relates to a novel crystalline forms of 2-[3,5-dichloro-4-(5-isopropyl-6-oxo-1,6-dihydropyridazin-3-yloxy)phenyl]-3,5-dioxo-2,3,4,5-tetrahydro[1,2,4]triazine-6-carbonitrile represented by the following structural formula-1, which is referred to as Resmetirom.

Formula-1.

Improved Process For Preparation Of Ponatinib Hydrochloride, Anonymous

Defensive Publications Series

Ponatinib hydrochloride, has a chemical name 3-(imidazo [1,2-b]pyridazin-3ylethynyl)-4-methyl-N-{4-[(4-methylpiperazin-1-yl)methyl]-3-(trifluoromethyl)phenyl} benzamide hydrochloride. Ponatinib hydrochloride is a kinase inhibitor which is indicated for the treatment of chronic myeloid leukaemia (CML) that is resistant or intolerant to prior tyrosine kinase inhibitor therapy or Philadelphia chromosome positive acute lymphoblastic leukaemia (Ph+ALL) that is resistant or intolerant to prior tyrosine kinase inhibitor therapy.

Affordable Long-Distance Taxi Rides, Mordechai Teicher

Defensive Publications Series

A user initiates a taxi ride to the destination. A manually driven robotaxi transports the user to a taxi station adjacent to the highway. The taxi driver disembarks from the robotaxi, and the robotaxi transitions to self-driving mode and commences a highway journey to another taxi station near the destination. Another taxi driver joins and drives the robotaxi to the final destination. In summary, the short manually driven first and last miles are charged at regular rates, while the long-distance highway trip is self-driven and bears a discounted rate, resulting in an affordable door-to-door long-distance taxi journey.

An Improved Process For Isolation Of Dual Agonist Polypeptide From Resin, Anonymous

Defensive Publications Series

An improved process for the isolation of dual agonist polypeptide compound comprising novel scavenging agent.

Optimizing Post-Transcription Language Analysis Using Audio Signal-To-Noise Ratio Estimations, Dongeek Shin

Defensive Publications Series

This publication describes using signal-to-noise ratio (SNR) estimators for audio recordings to identify an optimal large language model (LLM) for analyzing audio transcriptions generated from the audio recordings. This publication may enable users of microphone-enabled devices (e.g., smartwatches, tablets, wearables, cellular devices, mobile phones, etc.) to obtain more accurate responses to audio queries by using SNR estimators to infer the accuracy or quality of the transcription. The device may use the SNR estimates to select an appropriate LLM to generate responses to the audio input (e.g., queries, prompts, commands, etc.). It may be more difficult for speech-to-text transcription engines to …

Intelligent Topic Merging For Highly Correlated Data In Kafka, Alok Roy

Defensive Publications Series

The present disclosure provides systems and techniques to optimize resource usage in Apache Kafka by detecting and merging topics carrying highly similar or correlated data streams. The system consists of a Topic Similarity Detection Engine that analyzes data across Kafka topics using schema comparison, content similarity, and workload characteristics. Once topics with highly correlated data are identified, the system suggests or automatically merges these topics through a Merging Decision and Topic Merge Executor Module. The merging process is executed by dynamically adjusting partitions, consumer, and producer mappings, and ensuring consistency across merged topics. By reducing the number of partitions and …

Staggered Canary Rotation For Enhanced Forensic Analysis, Ray Van Hoose

Defensive Publications Series

Abstract:

This paper introduces a novel enhancement to traditional security measures, staggered canary rotation, designed to improve forensic analysis across a range of security domains. By rotating only a subset of canaries, watermarks, or similar markers at staggered intervals, this approach offers improved temporal resolution, enabling more precise detection and investigation of breaches. The method enhances understanding of the timing and scope of security incidents and provides a valuable tool for the broader threat detection community. It is open, easily implementable, and facilitates the sharing of detection insights, contributing to collective defense.

Techniques For Deriving An Llm Agent Trust Score For Dynamically Triggering Human-In-The-Loop (Hil) Feedback In Realtime For An Llm Agentic Workflow, Akram Sheriff

Defensive Publications Series

Techniques are proposed herein for deriving an adaptive and coherent agentic trust score in a multi-agentic system, which can be used to trigger a human-in-the-loop (HIL) workflow. Since not all workflows may require HIL feedback, the agent trust score can be used for decision making, in real-time, to determine whether or not HIL feedback is required for a given workflow. The techniques prioritize trust by establishing metrics across data boundaries, Application Programming Interface (API) reliability, and metadata compliance, each contributing to the agent trust score derivation. Thus, the real-time agentic scoring system proposed herein may foster confidence in Large Language …

Reactive Workflow Optimization Methods Through Emotion Detection And Machine Learning, Hp Inc

Defensive Publications Series

This idea relates to a machine learning algorithm that optimizes workflow through user interface modifications based on the user’s detected emotion. Emotion detection is done through the notebook computer’s sensor hardware and user inputs to the system. The algorithm analyzes the user’s facial expressions, voice, eye movements, keyboard strokes, mouse clicks, and other indicators of emotion and adjusts the user interface accordingly. The algorithm aims to enhance the user’s productivity, satisfaction, and well-being by providing personalized and adaptive user interface elements, such as colors, fonts, layouts, menus, notifications, app management, and feedback mechanisms. The algorithm also learns the user’s responses …

Enhancing Graph Learning Via Adaptive Neighborhood Feature Mixing

Defensive Publications Series

Graph Neural Networks (GNNs) have demonstrated remarkable success across various graph-related tasks; however, their performance often suffers when dealing with heterophilic graphs, where connected nodes tend to have dissimilar characteristics. This paper introduces a novel approach, Adaptive Neighborhood Feature Mixing (ANFM), that addresses the limitations of traditional GNNs when applied to heterophilic networks. ANFM dynamically learns how to mix feature information from a node's neighborhood based on node and edge attributes. We evaluate the effectiveness of ANFM on several benchmark heterophilic datasets and demonstrate that it outperforms existing state-of-the-art models. Our results highlight the importance of adaptive feature mixing for …

Addressing Challenges In Graph Neural Networks Through Adaptive Feature Transformation

Defensive Publications Series

Graph Neural Networks (GNNs) have demonstrated remarkable performance on various graph-related tasks. However, their effectiveness often diminishes when applied to heterophilic graphs, where interconnected nodes exhibit dissimilar attributes. This paper presents a novel approach, Adaptive Feature Transformation (AFT), designed to mitigate the challenges posed by heterophily. AFT incorporates a dynamic feature transformation mechanism, allowing nodes to adaptively adjust their representations based on the properties of their neighbors. We evaluate AFT on several benchmark heterophilic datasets and demonstrate that it achieves substantial performance gains over existing state-of-the-art GNN models. Our results underscore the importance of adaptive feature learning for processing complex, …

Adaptive Graph Learning With Node-Specific Aggregation

Defensive Publications Series

Graph Neural Networks (GNNs) have shown remarkable performance in various graph-based tasks, but their effectiveness often diminishes when applied to heterophilic graphs, where connected nodes tend to have dissimilar features. This paper addresses the challenges of learning on such graphs by proposing a novel approach, Adaptive Node-Specific Aggregation (ANSA), which dynamically adjusts the aggregation of neighbor information based on node-specific characteristics. ANSA employs learnable node embeddings and edge attributes to generate node-specific aggregation weights. We evaluate ANSA on several benchmark heterophilic datasets, demonstrating that it outperforms state-of-the-art GNN models designed for heterophilic graphs. Our results highlight the importance of adaptive …

Inserting Threaded Screws With Furrow Grooves, Marcel Stranz

Defensive Publications Series

A screw with longitudinally aligned furrow grooves can be used and lead to increased/ improved process reliability (screw seizure, increased torque).

Bumper Attachment To The Side Wall Frame (Zero Joint), Marcel Stranz

Defensive Publications Series

The principle of a carcase connector is combined in a plastic holder for a stable, positive and non-positive, detachable connection.

Creating Personalized Audio Playlist From Browser Tabs, Bookmarks, And Reading Lists, Na

Defensive Publications Series

There is no easy way for users to consume their browser content such as tabs, reading lists, and bookmarks in audio form. In many situations, users may benefit from read-aloud features in the web browser, allowing them to listen to articles. However, currently, users need to manually access a web page and initiate the browser read aloud feature. This disclosure describes techniques, implemented with user permission, to automatically create a personalized audio playlist for a user based on the user's web browser content such as tabs, bookmarks, and/or reading lists. Per the techniques, browser content is analyzed to identify web …

Secure Attestation During Device Reset, Subrata Banik, Andrey Pronin

Defensive Publications Series

Resetting a computing device can transition it to an insecure boot mode, thereby exposing the device to potential security breaches. This disclosure describes techniques that leverage a secure device ecosystem to achieve secure device reset or boot. Per the techniques, a device reset or boot can proceed only with an attestation by the true device owner, made secure by a cross-verification from another device within the device ecosystem. Specifically, when one device from the ecosystem receives a reset request, the device owner is required to approve the reset via another device from the ecosystem. Shared capabilities and login credentials across …

Payment Method Recommendations For In-Store Transactions, Ajay Prasad

Defensive Publications Series

Digital payment applications enable users to pay merchants via their smartphone or another device, e.g., by scanning a merchant quick response (QR) code displayed at the merchant location. However, such payment apps do not personalize the user experience or help users save money by availing relevant offers. This disclosure describes techniques that, with user permission, enhance the peer-to-merchant digital payment experience by utilizing contextual information including merchant identification and the user’s available payment methods, vouchers, coupons, etc. to automatically suggest a suitable payment method or coupon/voucher and/or to identify available offers as applicable to a transaction. The described techniques provide …

Power-Aware Detection Of User Gestures And Determination Of User Intent, Sabarish Sridhar, Gordon Wan

Defensive Publications Series

This disclosure describes techniques for accurate detection of user gestures and determination of user intent with low power consumption. Per techniques of this disclosure, sensors that are utilized to perform user intent determination are operated in a low power consumption mode prior to detection of user activity. The low power consumption mode includes a low resolution and/or low frame rate mode for operating the sensors. In the lower resolution mode, only a subset of the image sensors is active, and signals from the smaller set of image sensors are processed. Based on the preliminary detection of user activity, the computing …

Anomaly Detection On Multidimensional Time Series, Michael Yeh Visa, Xin Dai Visa, Yan Zheng Visa, Zhongfang Zhuang Visa, Laing Wang Visa, Yujie Fan Visa, Junpeng Wang Visa, Prince Osei Aboagye Visa, Uday Singh Saini Visa, Wei Zhang Visa

Defensive Publications Series

The present disclosure describes a method and system for detecting anomalies in multidimensional time series data using multidimensional matrix profiles are disclosed. A computer system can generate a multidimensional matrix profile corresponding to a multidimensional time series using a pre-sorting or post-sorting technique. The computer system can detect anomalies in the multidimensional time series using the multidimensional matrix profile, e.g., using thresholding or machine learning. The computer system can issue an alert to a requestor if an anomaly has been detected. Additionally disclosed is a k-nearest-neighbour extension to the multidimensional matrix profile anomaly detection methods.

Method For Minimizing Connector Stress When Installing Field-Replaceable Units Across Multiple Boards Simultaneously, Anonymous

Defensive Publications Series

This disclosure addresses the challenge of aligning multiple PCBs—such as a backplane and multiple switch cards—with a single Field Replaceable Unit (FRU) during installation. Traditionally, fixed mounting of all PCBs can cause misalignment and introduce harmful shear forces on connector pins, leading to deformation, signal degradation, or permanent damage. By fixing the backplane and allowing the FRU to self-align within its rail-mounted tolerances, the system ensures proper initial positioning. Additionally, each switch card is mounted on elastic neoprene grommets, enabling the card to “float” and adjust its position in any direction. This innovative approach accommodates manufacturing tolerances, reduces stress on …