n&k Technology (United States)

Data slice finding is an emerging technique for validating machine learning (ML) models by identifying and analyzing subgroups in a dataset that exhibit poor performance, often characterized by distinct feature sets or descriptive metadata. However, in the context of validating vision models involving unstructured image data, this approach faces significant challenges, including the laborious and costly requirement for additional metadata and the complex task of interpreting the root causes of underperformance. To address these challenges, we introduce AttributionScanner, an innovative human-in-the-loop Visual Analytics (VA) system, designed for metadata-free data slice finding. Our system identifies interpretable data slices that involve common model behaviors and visualizes these patterns through an Attribution Mosaic design. Our interactive interface provides straightforward guidance for users to detect, interpret, and annotate predominant model issues, such as spurious correlations (model biases) and mislabeled data, with minimal effort. Additionally, it employs a cutting-edge model regularization technique to mitigate the detected issues and enhance the model's performance. The efficacy of AttributionScanner is demonstrated through use cases involving two benchmark datasets, with qualitative and quantitative evaluations showcasing its substantial effectiveness in vision model validation, ultimately leading to more reliable and accurate models.

The advances in multi-modal foundation models (FMs) (e.g., CLIP and LLaVA) have facilitated the auto-labeling of large-scale datasets, enhancing model performance in challenging downstream tasks such as open-vocabulary object detection and segmentation. However, the quality of FM-generated labels is less studied as existing approaches focus more on data quantity over quality. This is because validating large volumes of data without ground truth presents a considerable challenge in practice. Existing methods typically rely on limited metrics to identify problematic data, lacking a comprehensive perspective, or apply human validation to only a small data fraction, failing to address the full spectrum of potential issues. To overcome these challenges, we introduce VISTA, a visual analytics framework that improves data quality to enhance the performance of multi-modal models. Targeting the complex and demanding domain of open-vocabulary image segmentation, VISTA integrates multi-phased data validation strategies with human expertise, enabling humans to identify, understand, and correct hidden issues within FM-generated labels. Through detailed use cases on two benchmark datasets and expert reviews, we demonstrate VISTA's effectiveness from both quantitative and qualitative perspectives.

Mental health monitoring is essential for early inter-vention and effective treatment, yet traditional methods often rely on infrequent assessments and self-reported symptoms, leading to delays in care. The proposed AI-Driven Patient Monitoring and Mental Health Analytics Platform leverages artificial intelligence, machine learning, natural language processing, and deep learning to provide continuous and real-time insights into patient well-being. By integrating IoT-enabled remote monitoring, sentiment analysis, and predictive analytics, the platform enhances clinical decision-making and facilitates proactive mental health management. The system aggregates data from daily mood logs, chat interactions, and biometric sensor inputs, using a hybrid AI model that combines sentiment analysis with predictive modeling based on LSTM and Transformer architectures. A clinician dashboard provides interactive visualizations, trend analysis, and automated alerts, enabling early detection of mental health deterioration. Data security measures, including AES-256 encryption and JWT-based authentication, ensure patient confidentiality and compliance with healthcare standards. Through continuous monitoring, AI-powered insights, and secure communication features, this platform aims to bridge the gap between patients and mental health professionals, improving accessibility, early intervention, and personalized care. Future enhancements will focus on reinforcement learning for adaptive AI responses and expanded integration with additional wearable technologies.

In this research, a web application is developed that includes the identification of medicinal plants. The system uses deep learning to enable users to upload images and scans of herbal plants, pasting and analyzing them using deep learning convolutional neural networks. The system allows users to search for the medicinal purpose of plants. A powerful convolutional neural network, Inception V3, which has been trained on a variety of medicinal plant leaves is utilized in the system. The model can classify the type of plant effectively and scan the leaf for its medicinal purpose. The system can assist researchers, doctors, and traditional health practitioners by correlating leaf traits with proven medicinal properties. It performs better for plant classification and health detection task than previous methods. To enhance classification performance and minimize overfitting, we have employed data preprocessing methods such as normalization, data augmentation, and noise reduction. The Inception V3 model can find out the different features of a leaf. Like it includes venation, texture, and color gradients. It has been found that the accuracy achieved with the system is 99.8% which is better than the manual methods as well as other variants. Besides recognizing a plant species, the application also has a database that provides medicinal information about the plant.

Native Americans (Indigenous Americans) have high rates of serious illness in the United States. Informal caregivers are heavily relied on in caring for patients in low-resource settings. The needs of caregivers residing on reservations are sorely underreported. Therefore, our objective was to examine relationships between facilitators (communal mastery, cultural identity, and spirituality) and barriers (depression, anxiety, stress, and burden) with supportive care needs among adult informal caregivers in 1 reservation-based community. A cross-sectional survey was distributed in July and August 2022 as part of a larger multimethod, community-based participatory research study. We used descriptive statistics and linear regression models to examine relationships against the primary outcome, the Supportive Care Needs Assessment Tool for Indigenous People. Overall, 127 participants were included; most were female (n = 92, 72.4%), were between 30 and 49 years (n = 57, 44.9%), and had 6 months or less of caregiving experience (n = 41, 32.5%). Higher depression, anxiety, stress, and burden were significantly associated with higher Supportive Care Needs Assessment Tool for Indigenous People scores. Overall, mental health is a significant barrier that may indicate greater supportive care needs among informal caregivers, although further work is needed to differentiate symptoms and their impact on caregiving from a cultural perspective.

Cloud computing (CC) delivers on-demand and powerful business environments. It can be constructed on top of virtualized data hubs. Virtualization renders a flexible infrastructure for cloud. However, scheduling the tasks and managing the resources in virtualized data hubs will be difficult. Maximizing the virtual machine utilization and minimizing the total makespan is imperative in the cloud environment. The task scheduling issue can be indicated as multi-objective optimization problem, which is solved by utilizing optimization methods. This study develops a Mantaray Foraging Optimization based Makespan Enhancement in Cloud based Scheduling Environment (MROME-CSE) technique. The presented MROME-CSE technique aims to allocate the resources proficiently in the cloud environment. In the presented MROME-CSE technique, the behavior of manta rays is mainly involved. Furthermore, the MROME-CSE method derived an objective function to allocate n individual tasks to m heterogeneous accessible resources, thereby task conclusion period was minimalized and resource was fully exploited. The experimental evaluation of the MROME-CSE technique is examined briefly. The experimental outcomes stated the superior performance of the MROME-CSE technique over other recent models.

Abstract Extended-reach drilling (ERD) from a fixed offshore platform poses unique challenges for anticollision directional drilling and wellbore construction. Wells completed in multilayered reservoirs, multilaterals with compartments of varying pressure, and ERD wells are becoming common in the Russian sector of the Caspian Sea. An integrative approach and advancements in drilling and completion technologies in the Korchagina field have enabled enhanced hydrocarbon recovery through extreme reservoir contact from ERD wells. Various issues, such as wellbore stability, borehole geometry, torque and drag, and borehole condition can affect the deployment of completion hardware to the desired depth. This paper presents the installation of ground-breaking technology in an ERD well in which the longest fully electrical intelligent completion was landed successfully through proper design and careful planning. The paper will also present the planning and execution of three ERD wells completed by use of a fully electrical intelligent completion consisting of multiple downhole active flow-control devices controlled by a single downhole electrical line. The installation of these completion systems allowed the operator to control the breakthrough of water or gas or both, to selectively produce additional hydrocarbons from multiple zones and compartments in real time, and to minimize the number of required production logging tool runs. In 3D complex well profiles, while running the long completion string, axial drag is accumulated along the string, which exposes it to sinusoidal and helical buckling. Excessive compressional loads in the string beyond the buckling limit of the pipe can lead to lockup and ultimately prevent the string from reaching the setting depth. This paper presents a structural integrated approach for successfully deploying a fully electrical multipoint pressure and temperature measurement completion systems on 3 ½-in. and tapered 3 ½ × 4 ½-in. tubing by using a wireless electric connect-disconnect system.

A new formulation of optical dispersion in terms of complex index of refraction, N(E) as a function of photon energy, E, applicable to insulators and semiconductors, is introduced. Previous formulations, developed over the past 200 years, are either valid only over a limited spectral range, or do not accurately fit experimental data, or do not conform to the Principle of Causality. The new formulation overcomes these shortcomings. Its validity is established based on theoretical and experimental considerations. The theoretical basis stems from consistency with Titchmarsh’s Theorem, a comprehensive mathematical expression of the Principle of Causality. In accordance with Titchmarsh’s Theorem, the expressions for the real and imaginary parts of N(E), the refractive index, n(E), and extinction coefficient, k(E), are shown to be skew reciprocal Hilbert Transforms. Also, the expression for the inverse Fourier Transform of N(E) is shown to be a complex-valued causal function of time. The causal function is identified as the polarization of the medium, P(t), if photons reach the medium at an initial time equal to zero. P(t) is associated with photon excitations of electrons and other quantum constituents of the media. The experimental basis of the formulation stems from demonstrating exceptionally close agreement with published experimental n(E) and k(E) data of different forms of insulators and semiconductors. Specifically, the formulation is applied to data of Water, Ice, SiO2-Glass, LiF, Polyethylene, a-Si, InSb, GaP, and As2Se3. Data for Water span thirteen orders of magnitude, and fourteen for Ice, over the spectral range of Radio-Waves to EUV. Data of the other materials involved smaller spectral ranges. Results indicate that fewer fitting parameters are needed over any segment of the electromagnetic spectrum, compared to other formulations. These theoretical and experimental findings suggest the formulation represents a universal description of optical dispersion of insulators and semiconductors.

Optically activated, high gain GaAs switches are being tested for many different pulsed power applications that require long lifetime (longevity). The switches have p and n contact metallization (with intentional or unintentional dopants) configured in such a way as to produce p-i-n or n-i-n switches. The longevity of the switches is determined by circuit parameters and by the ability of the contacts to resist erosion. This paper will describe how the switches performed in test-beds designed to measure switch longevity. The best longevity was achieved with switches made with diffused contacts, achieving over 50 million pulses at 10 A and over 2 million pulses at 80 A.

Abstract A 52 well heavy oil field development targeting shallow (3,400-4,000 ft TVD) sands on the north slope of Alaska was initiated in 2008. Horizontal wells of 11,000 to 13,000 ft MD were drilled early in the program. These initial wells served as "data-gathering and technology-proving" opportunities leading up to the eventual 23,000 ft MD wells. Key technical challenges include equivalent circulating density (ECD) and drag management. ECD management became essential in the 8½" production hole section of the longer wells. A relatively narrow (<600 psi) mud weight window necessitated changes to casing, drill-string, drilling fluids, and operational parameters. Lighter weight production casing allowed drilling a larger production hole (8¾" vs. 8½"). A tapered drill-string, reduced mud rheology, and reduced flow rate all became a necessary part of the ECD management solution. Advanced drag management techniques are required to install the 9⅝" production casing, 5½" production liner, and 4½"x3½" intelligent inner-string completion. The 9⅝" casing is installed using the "buoyancy assist" method (ie, "floatation") so it may be "pushed" and reamed in the hole beyond the point of negative weight. The lower completion consists of a 5½" slotted liner with swell packers. Centralizers on the liner were changed from non-rotating to fixed, which allowed breaking axial drag while reaming the liners to depth. Extensive torque and drag modeling was used to plan intelligent inner-string completions on the injector wells, which included injection control devices, swell packers, and a fiberoptic distributed temperature sensor (DTS) to monitor injectivity. The full length paper will discuss the technical challenges, well design solutions, and operational practices that were trialed and implemented to enable extended reach wells to be successfully drilled on the edge of the industry experience envelope with all wells meeting targeted objectives.