Optimized Iterative Methods For Markov Decision Processes: A Subspace Approach

Defensive Publications Series

Markov Decision Processes (MDPs) are fundamental in computational decision-making frameworks, requiring efficient computational methods for their resolution. Traditionally, methods such as matrix splitting have been used to decompose the Bellman equations for iterative solutions. However, the Krylov Subspace Method (KSM) presents a more efficient alternative, leveraging dynamic iterative techniques that significantly outperform static ones. This paper introduces the Enhanced Iterative Subspace Method (EISM), detailing its conceptual motivations, practical implementations, and superiority through various computational experiments.

The Resolution Combining Composite Projection Configuration Which Contains 4 Cyclically Arranged Projection Configurations., Punarjeewa Abeysekera

Defensive Publications Series

This paper describes a resolution combining composite projection configuration which contains 4 cyclically arranged projection configurations. The resolution combining composite projection configuration will contain 4 cyclically arranged reflecting surfaces and 4 cyclically arranged projection configurations. Each projection configuration will have the capability to project a distributed projection. The resolution combining composite projection configuration, utilizing the 4 projection configurations, will have the capability to simultaneously project 4 distributed projections which can be combined to produce a one single projection.

Advancements In Sequence Generation: A Gan-Based Reinforcement Learning Approach

Defensive Publications Series

This paper builds upon the SeqGAN framework, aiming to introduce targeted enhancements to improve its efficacy. SeqGAN, a pivotal algorithm for generating discrete token sequences, is extensively utilized across various NLP applications. By conceptualizing the data generator as a stochastic policy within the domain of reinforcement learning (RL), SeqGAN adeptly sidesteps the challenges associated with generator differentiation through direct gradient policy updates. The RL reward signal is derived from the evaluations of a GAN discriminator assessing complete sequences, subsequently informing state-action decisions via Monte Carlo search. Notable enhancements include the incorporation of WGAN, employing Earth Mover’s distance rather than KL …

Presenting Information Based On Inferred Interpersonal Relationships, N/A

Defensive Publications Series

Users often meet someone new, or someone known that the user has forgotten. In such cases, it is typically useful to learn, deduce, or recall details of the person. This disclosure describes techniques, implemented with user permission, that help users in contextual identification of unfamiliar individuals they meet in various situations. Identification can be performed with appropriate permissions by applying computer vision and facial recognition to people’s photos and other relevant contextual information. Implementations can combine standard off-the-shelf components such as face recognition, speech recognition, OCR, AR, LLM, etc. Relevant cues about the individuals can then be proactively surfaced via …

Strategic Approaches To Mitigate The Impact Of Ebola Virus Disease In West African Nations

Defensive Publications Series

Ebola Virus Disease (EVD) poses a significant threat to global health, with mortality rates ranging from 50% to 90% during outbreaks. Identified initially in 1976 near the Ebola River in what is now the Democratic Republic of Congo, EVD has periodically led to severe outbreaks. The recent epidemic in West Africa, notably affecting Guinea, Sierra Leone, and Liberia, underscores the urgent need for robust health infrastructure and international collaboration to address such pandemics. This study delves into the outbreak's dynamics and countermeasures in these countries, aiming to shed light on enhancing global health responses to EVD and similar health emergencies. …

Optimization Strategies For Epidemic Control: A Case Study On Ebola In West Africa

Defensive Publications Series

Ebola Virus Disease (EVD) remains one of the deadliest viral infections, with mortality rates ranging from 50% to 90% during outbreaks. Identified first in 1976 near the Ebola River in what is now the Democratic Republic of Congo, EVD outbreaks have sporadically emerged, causing widespread devastation. The recent epidemic in West Africa underscores the critical need for robust health systems and international collaboration in disease management. This paper examines the outbreak dynamics and response efforts in Guinea, Sierra Leone, and Liberia—countries severely impacted due to fragile healthcare infrastructures. Our analysis aims to glean insights into enhancing healthcare responses to EVD …

Strategies For Controlling Ebola Outbreaks In West Africa: An Optimization Model Approach

Defensive Publications Series

Ebola Virus Disease (EVD) is a severe, often fatal illness in humans, with an average fatality rate of about 50%, which can rise to 90% during outbreaks. Since its first identification in 1976 near the Ebola River in the Democratic Republic of Congo, EVD has led to sporadic, yet devastating outbreaks. The most significant of these occurred recently in West Africa, majorly impacting Guinea, Sierra Leone, and Liberia due to inadequate healthcare systems. This crisis underlines the urgent need for enhanced global health security and international collaboration to manage such diseases effectively. This study analyzes the outbreak and response strategies …

Improved Video Search By Encoding Text And Visual Query Input Using A Dual Encoder Multimodal Model, D Shin

Defensive Publications Series

This disclosure describes techniques that receive text queries and additional visual input from a user to improve video search. The text input and visual input are separately encoded using a dual encoder model and combined into a query fingerprint vector. The dual encoder model is trained using contrastive loss such that when performing feature comparison, the feature comparison space enables the use of simple L2 distances to identify videos that match the input query. The fingerprint vector is in the same feature space as pre-computed vectors for available videos, enabling fast comparison and ranking. The identified matching videos are ranked, …

Lazy Loading Of Certificates During Client Validation, Zhaowei Xu

Defensive Publications Series

This disclosure describes techniques to speed up video runtimes (e.g., implemented in a video streaming app) that cater to a limited number of clients (e.g., video streaming apps across different operating systems or platforms supported by a video streaming provider) by keeping only a small number of certificates in memory. Per the techniques, the issuer name of the trusted root certificate is maintained in memory. When a client certificate chain is presented, the issuer name is tested for a match with the certificate chain. If a match is found, the corresponding root certificate is loaded and verified on demand. The …

Positional Embedding Meets Structural Embedding Of Graph Representation Learning, Yuzhong Chen, Menghai Pan, Huiyuan Chen, Mahashweta Das, Hao Yang

Defensive Publications Series

Present disclosure discloses a computer-implemented method of positional embedding for encoding positional information. The computer-implemented method includes generating a positional embedding for a node utilizing a random walk with restart (RWR) proximity distribution to generate a pairwise proximity distribution matrix comprising a plurality of rows. The computer-implemented method includes sorting the pairwise proximity distribution matrix for the node. The computer-implemented method includes generating a structural embedding based on the sorting of the pairwise proximity distribution matrix for the node.

Consistent Partitioning Of A Dataset Undergoing Streaming Updates, Jonathan Halcrow, Siddhartha Jayanti

Defensive Publications Series

This disclosure describes techniques to partition a stream of N items into groups approximately of size S under the constraint that a slight alteration in the stream changes the partitions only slightly. Per the techniques, the items are sorted by a hash of their identifier. The length N of the stream is estimated using, e.g., the Flajolet-Martin algorithm. The number of bits B of the hash to be considered is given by B = floor(log₂(N/S)), where S is the target partition size. A point in the stream is assigned to a bucket identified by …

Deepfake-Resistant Video Using Blockchain Ledgers And Machine Learning, Stephane Belmon, Roshan Rammohan

Defensive Publications Series

With the advent of deepfakes videos, it is important for the purposes of promoting societal trust to be able to verify the authenticity of a person who appears in a social media audiovisual stream or a broadcast/cable television video. This disclosure describes techniques that enable the viewer of a particular piece of video purporting to show a person to verify that the video indeed represents the person authentically in all respects, e.g., that the person truly made the statements heard in the video and did so with their authentic tone, cadence, delivery, body language, etc. A video camera certified as …

Authentication Based On Wifi Auto-Connect And Device Location, Hui Wang, Xiaoying Liu

Defensive Publications Series

This disclosure describes techniques that leverage WiFi authentication to enable certain tasks or run certain applications on a mobile device when the user is in a specific location and their device is connected to a specific WiFi network. When the device automatically connects to a WiFi access point, the WiFi network is tested for membership in a predefined list. If the WiFi network is found in the predefined list, it is determined whether the device is within a predefined geofence. If the device is in the geofence, one or more predefined actions are automatically performed. In secure and familiar environments …

Generative Video Ads Tailored To Market Segments, Manoj Muthamizh Selvan

Defensive Publications Series

With user permission, advertisements that are delivered to a user can be selected based on user data in a manner that the advertisements are likely to be useful to the user. However, the content of the advertisement itself is not personalized to the viewer or viewer group/ market segment. This disclosure describes techniques that utilize generative artificial intelligence (gen AI) to automatically create video ad content customized to particular market segments. Inputs from the advertiser such as brand images, logos, props, etc., together with market segment descriptions are provided to a large language model (LLM) that generates text prompts. The …

User Friendly Playback For Advertising-Based Video On Demand Videos, Veenu Mishra

Defensive Publications Series

Systems and techniques are proposed that provide an improved experience for a user when viewing advertising-based video on demand (AVOD) media content. A television (TV) application may interface with an over-the-top (OTT) media service to deliver the AVOD media content to a media player by way of the internet. The media player may be a computing device (e.g., a Smart TV) connected to a network and interfaced to the internet. The television application can automatically mute the sound during the playing of the advertisements (ads) (during the ad time) in the media content improving the experience of the user when …

Dynamic Credit Limit Adjustemnt System, Palakh Shangle Visa

Defensive Publications Series

The present disclosure relates to a Dynamic Credit Limit Adjustment System which is an innovative financial technology solution designed to transform the way credit limits are managed by financial institutions. Utilizing advanced artificial intelligence (AI) and machine learning algorithms, this system dynamically adjusts an individual's credit limit in real-time based on a comprehensive analysis of their financial data. This system integrates various financial indicators such as income changes, spending patterns, payment history, and current economic trends to provide a more accurate and responsive credit limit. By doing so, it addresses the limitations of traditional credit limit assignment methods, which are …

The Concave Mirror Based Light Capturing Configuration., Punarjeewa Abeysekera

Defensive Publications Series

This paper describes a concave mirror based light capturing configuration. The concave mirror based light capturing configuration consists of a concave mirror which have a diameter equivalent to its focal length, 4 cyclically arranged beam splitters and 4 cyclically arranged light collecting sensors. The configuration, by combining the image projection sections captured utilizing the 4 cyclically arranged light collecting sensors, will have the capability to generate the complete image projection of the scene in front of it.

Method And System To Classify Regional Vs Non-Regional Profile, Jyothi Jayakumar, Velkumar Thangavel, Soman Sounderrajan

Defensive Publications Series

The present disclosure includes methods, systems and computer program products for implementing technical design details for an application change. Particularly, the present disclosure provides a method comprises enabling 302 the prompt page for the given dossier. Further, the method 300 comprises accessing 304 the user profiles based on the given profile filter. Furthermore, the method 300 comprises determining 306 to classify the user profiles based on regional and non-regional profiles, for implementing the changes in the user application. Particularly, the present disclosure provides a system that may include the application change for modifying file system, user interface, and business …

The Projection Array Which Have The Capability To Simultaneously Project 9 Projections Associated With The Same Scene Where Each Projection Will Have A Unique Brightness Level., Punarjeewa Abeysekera

Defensive Publications Series

This paper describes a projection array which have the capability to simultaneously project 9 projections associated with the same scene where each projection will have a unique brightness level. The projection array consists of 9 identical projection configurations arranged in a 3*3, 2 dimensional array system. Each identical projection configuration will contain a configuration that is capable of displaying pixel patterns that is associated with a high brightness light source, a convex lens stack and an enclosure. The projection array will have the capability to collectively project 9 overlapping projections which are associated with displaying content of the same scene. …

A Workflow To Automatically Generate The Stabilization Gaps Between Levels In A Multi-Level Build, Hp Inc

Defensive Publications Series

3D Printing technologies frequently allow to apply different parameters to specific regions of the build (levels). The builds with these characteristics are called multi-level builds. This possibility is very relevant, since it allows to print in the same build 3d models with different parameters, meeting the expected printed characteristics. But this functionality has a limitation: for some parameters (i.e. temperature), it is not possible to transition directly from one value to the other, but it is necessary to print some layers (stabilization gap) to transition from one value to the other. This disclosure provides a workflow to be able to …

Monochrome Temporal Dithering Using A Color Dither Mask, Edward Buckley, Pin-Chieh Huang

Defensive Publications Series

No abstract provided.

Reinforcement Learning Based Mechanism For Automatic Exploitability Testing, Chris Schneider, Nic Watson, Martin Sablotny, Ross Anderson

Defensive Publications Series

This disclosure describes the use of reinforcement learning techniques to generate code exploits that can validate findings of vulnerability in an automated fashion. Using static analysis, a machine learning model is built with a large corpus of code with known exploits. A cost function and associated criteria are updated to determine legitimate vulnerabilities. The model output is a number of proof-of-concept exploits that can be reused and run across the output of the static analysis tool. Executions resulting in an exploit provide proof of validity and can be flagged for remediation. Validation criteria for software mitigation and procedural controls are …

Self-Driven Issue Tackling Aid, Rutuja Pravin Kharche, Shweta Humnabadkar, Homarghya Homroy

Defensive Publications Series

Current network monitoring and health visualization tools and systems provide the results of their analysis only in the form of a report. To address that limitation, techniques are presented herein that support a range of enhancements to such tools.

• One such enhancement allows the tools to automatically diagnose an issue, and provide a solution to the same, using predictive analysis.
• Another such enhancement encompasses training the tools to provide detailed and forthright solutions to an end user.

Accordingly, aspects of the techniques encompass a solution-oriented approach and support a monitoring and solutioning tool MAST that is able to interact with …

Natural Conversational Interaction With Ai Agent Using Spoken Interrupt Context, Will Walker

Defensive Publications Series

Virtual assistant applications and other artificial intelligence (AI) agents can use large language models (LLMs) to generate responses to spoken user queries. However, responses from LLMs can be long and include more information than the user’s specific need. As a result, it can be difficult for users to understand and remember the answers easily simply by listening. This disclosure describes techniques to enhance the conversational capabilities of voice based interactions with virtual assistant applications and other AI agents by enabling users to interrupt spoken responses with a new query. The context of the interruption, including contents of the spoken response …

Additional Control Data Path Into Legacy Pluggable Form Factors, Brandon Dusek, Aj Velarde, Latha Sathappan, Hailing Wang

Defensive Publications Series

Techniques are presented herein that support an additional control data path for a pluggable module that employs the module’s Vcc power pins. Adding a signal modulation path to a pluggable module’s Vcc power delivery and then using the same as a management communication interface provides a form-factor agnostic approach to expand a module’s management channel. Such a data path can operate in addition to the Inter-Integrated Circuit (I2C) interface communication lines to support the sending of larger amounts of data at cumulatively faster speeds. Additionally, the presented techniques provide an approach to backward compatibility that will be tolerable by older …

A Process For The Preparation Of (6s)-N-[4-[[(2s,5r)-5-[(R)-Hydroxy(Phenyl) Methyl]Pyrrolidin-2-Yl]Methyl]Phenyl]-4-Oxo-7,8-Dihydro-6h-Pyrrolo[1,2-A]Pyrimidine-6-Carboxamide And Its Intermediates Thereof, Msn Laboratories Private Limited; R&D Center; Srinivasan Thirumalai Rajan; Sajja Eswaraiah; Gade Srinivas Reddy; Revu Saryanarayana; Challa Suresh

Defensive Publications Series

The present disclosure relates to process for the preparation of (6S)-N-[4-[[(2S,5R)-5-[(R)-hydroxy(phenyl) methyl]pyrrolidin-2-yl]methyl]phenyl]-4-oxo-7,8-dihydro-6H-pyrrolo[1,2-a]pyrimidine-6-carboxamide of formula-1 and its intermediates.

A Process For The Preparation Of (6s)-N-[4-[[(2s,5r)-5-[(R)-Hydroxy(Phenyl) Methyl]Pyrrolidin-2-Yl]Methyl]Phenyl]-4-Oxo-7,8-Dihydro-6h-Pyrrolo[1,2-A]Pyrimidine-6-Carboxamide And Its Intermediates Thereof.

Defensive Publications Series

The present disclosure relates to process for the preparation of (6S)-N-[4-[[(2S,5R)-5-[(R)-hydroxy(phenyl) methyl]pyrrolidin-2-yl]methyl]phenyl]-4-oxo-7,8-dihydro-6H-pyrrolo[1,2-a]pyrimidine-6-carboxamide of formula-1 and its intermediates.

Efficient Augmentative Communication Using Large Language Models, Adriana Guevara-Rukoz, Atsuko Yamagami, Atsushi Yamashita, Satoru Arao, Yasuaki Takebe, Shuhei Iitsuka

Defensive Publications Series

For augmentative and alternative communications (AAC), there exists an inverse relationship between input expressivity and input efficiency: A technique that selects characters one-by-one provides substantial freedom of expression but requires many actions to compose a sentence, while techniques that select from a preset vocabulary can enable quicker input that is limited to the vocabulary. This disclosure describes techniques of augmentative and alternative communication that leverage a large language model (LLMs) to simultaneously optimize both input expressivity and input efficiency. Upon the entry of a few characters of a sentence or command (e.g., using eye trackers, finger tapping, etc.), word and …

On-The-Fly Measurement Unit Conversion, Daniel V. Klein, Jonathan C. Chu

Defensive Publications Series

Mobile devices include features to provide translation of text, audio, and video between languages. Multimodal translation enables users to interpret information in languages that they do not understand by having it automatically translated to their preferred language. However, current translation techniques do not convert measurement units. This disclosure describes techniques that provide on-the-fly conversion of measurement units and currency based on user preference. The user can specify their preferred unit for temperature, weight, distance, etc. as well as their preferred currency. When providing a translation or otherwise displaying text, the user device can automatically detect information that is in a …

Selective Enhancement Of Timeboxed Video Segments For Security Camera Recordings, Alexander Haubold

Defensive Publications Series

Video captured by security cameras is of low resolution and quality, especially in low-light environments. Currently, such cameras do not have the capability to enhance the recorded videos to fix such issues. While cloud-based video enhancement is feasible it is computationally resource intensive and slow, making it unsuitable for use in enhancing security camera footage. This disclosure describes techniques to selectively apply video enhancement to timeboxed segments within the video content recorded by a security camera. A human operator or a computer-vision based automated technique identifies a segment or a set of segments related to an event of interest within …