3633 research outputs found
Sort by
Digital detection of attention and distraction behaviors
Get PDFPaying attention helps us learn, advance in our careers, and build successful relationships, but when it’s compromised, achievement of any kind becomes far more challenging. Causes of not paying attention can range from common factors like sleep deprivation, stress, or a mood disorder to health difficulties such as ADHD, OCD, or a thyroid problem that affects concentrating. This work extracts paying attention and not paying attention behavior patterns in the context of learning. In early work, our study identified attention and distraction behaviors using gathered video recordings of online classes. The work found ten paying attention behaviors and six distracted behavior patterns. In this paper, we use computer vision techniques to extract features related to these behaviors. These features are distance between hand and face, pitch yaw roll, eye-to-camera distance, hand-to-camera distance, iris direction, gaze tracking, mouth aspect ratio, eye aspect ratio, distance between face and frame side, and facial landmark configuration. This research also applied three types of machine learning—logistic regression, decision trees, and random forest—and the accuracy rates were 79%, 86%, and 89%, respectively. This result is better than relying only on two extracted features in our previous work
AI-based system for in-bed body posture identification using FSR sensor
Get PDFNon-invasive sleep monitoring holds significant promise for enhancing healthcare by offering insights into sleep quality and patterns. In this context, accurate detection of body position is crucial, as it provides essential information for diagnosing and understanding the causes of various sleep disorders, including sleep apnea. The aim of this work is to develop an efficient system for sleep position detection using a minimal number of FSR (Force Sensitive Resistor) sensors and advanced machine learning techniques. A hardware setup was developed incorporating 3 FSR sensors, on-board signal processing for frequency boundary filtering and gain adjustment, an ADC (Analog-to-digital converter), and a computing unit for data processing. The collected data was then cleaned and structured before applying various machine learning models, including Logistic Regression, Random Forest Classifier, Support Vector Classifier (SVC), K-Nearest Neighbors (KNN), and XGBoost. An experiment with 15 subjects in 4 different sleeping positions was conducted to evaluate the system. The SVC demonstrated notable performance with a test accuracy of 64%. Analysis of the results identified areas for future improvement, including better differentiation between similar positions. The study highlights the feasibility of using FSR sensors and machine learning for effective sleep position detection. However, further research is needed to improve accuracy and explore more advanced techniques. Future efforts will aim to integrate this approach into a comprehensive, unobtrusive sleep monitoring system, contributing to better healthcare services
Three-level flying capacitor multilevel topology with delta-sigma modulation
No full textThis paper proposes a novel modulation concept for a 3-level fying capacitor multilevel system using an asynchronous delta-sigma modulator (ADSM). The benefts of delta-sigma modulation, including noise shaping, spread-spectrum qualities, and a lower average switching frequency while maintaining the same signal quality as traditional pulse-width modulation (PWM) methods, can be exploited in multilevel power converters. A control circuit is introduced which determines the fying capacitor voltage from the output switching terminal and keeps the fying capacitor at a constant voltage. In this paper, we present the concept of the modulator and the control circuit to control the fying capacitor voltage. We confrm the modulation method by simulations and measurements
Completely defined cell culture medium for advanced alveolar models
Get PDFIn vitro alveolar models are an important tool study respiratory physiology, investigate lung diseases, develop and test new therapies, and simulate inhalation exposure to environmental pollutants or therapeutic agents. A defined cell culture medium is essential for maintaining consistent and reliable in vitro models. In this study a defined cell culture medium for advanced alveolar models based on epithelial A549 cell line and endothelial EaHy926 cell line is developed. Cellular survival was proven by quantification of LDH release. It was shown that co-culture of the cells enhances cellular survival compared to the endothelial monoculture. The formation of a homogeneous monolayer on laminin coated membranes was confirmed by actin staining. Further, the expression of cell type specific proteins (endothelial: CD31, VE-cadherin; epithelial: ZO1 and E-cadherin) was shown. Overall, we demonstrated a completely defined cell culture medium that can be used for the setup of an advanced in vitro alveolar model for biomedical application like drug development or breath gas analysis
Interpreting XGB using LIME and SHAP for otosclerosis and disarticulation diagnosis
Get PDFMachine learning algorithms and neural networks have recently been used for the classification of middle ear disorders using wideband acoustic immittance and wideband tympanometry data. This study applies the extreme gradient boosting (XGB) classifier, trained on simulated WAI data, to classify real measured data for normal, otosclerotic, and disarticulated ears. The achieved macro recall of 82 % is comparable to other approaches trained with real measurement data. The interpretability methods LIME and SHAP are used to quantify each feature’s contribution, both revealing energy reflectance between 600-800 Hz as a key feature for all classes. The key feature identified matches the differences that can be visually observed in the training and test data. However, the obtained feature contributions don’t provide enough distinguishable information to recognise incorrect or uncertain classifications
Der Einsatz künstlicher neuronaler Netze in der generativen KI aus der Perspektive des Marketing
No full textAufgrund des transformatorischen Potenzials von generativer künstlicher Intelligenz (GKI) ist ein grundlegendes Verständnis der zugrunde liegenden Modelle erforderlich, um sie effektiv in der Marketingpraxis einzusetzen. Der Beitrag beleuchtet daher die zentrale Rolle neuronaler Netze innerhalb der GKI und erläutert ihre Funktionsmechanismen, wesentlichen Entwicklungen und Einsatzmöglichkeiten. Neben dem Potenzial generativer Netze, synthetische Daten mit hoher Ähnlichkeit zu realen Daten zu erzeugen, werden auch relevante Anwendungsherausforderungen aufgezeigt. Dazu gehören technische Implikationen wie die Herstellung eines Nash-Gleichgewichts zwischen konkurrierenden Netzkomponenten sowie erweiterte praktische Implikationen wie zuverlässige Metriken zur Leistungsbewertung. Daraus ergibt sich ein praxisorientierter Einblick in den aktuellen Diskurs der GKI im Marketing aus der algorithmischen Perspektive neuronaler Netze
Wirtschaftlichkeitsbewertung von KI-Projekten : ein Beispiel für erklärbare KI zur Rüstzeitoptimierung in der CNC-Werkzeugherstellung
Get PDFMithilfe von Modellen der erklärbaren Künstlichen Intelligenz (XAI) können die Rüstzeiten in der CNC-Werkzeugherstellung reduziert werden, wodurch nicht nur die betriebliche Effizienz und Produktionskapazität gesteigert, sondern auch die wirtschaftliche Rentabilität von KI-Projekten nachgewiesen werden kann
The role of family dynamics on transition processes in family firms
No full textIn a qualitative study of 16 family firms in Germany, we examine four family dynamics, namely cohesion of extended families, multiple involved family members, predecessor-successor-relationship, and parents and partner influence. 31 semi-structured interviews with predecessor and successor CEOs identified that the family dynamics have different influences on four major transition mechanisms, namely control, knowledge transfer, harboring, and emotional alignment that lead to two outcomes, heritage maintenance and transformation. Our findings are encapsulated in a family dynamic-transition-outcome model
Nachhaltige Entwicklung: Ursprung und aktuelle Herausforderungen verstehen und Zukunftsperspektiven entwickeln, damit alle mitreden und handeln können
No full textIn diesem Kapitel wird ein kurzer Überblick über die Geschichte der Bewegung für nachhaltige Entwicklung, über Modelle und zeitgenössische Perspektiven zu diesem Thema sowie über zukunftsweisende Mechanismen, die eine stärkere Beteiligung an der Diskussion über Nachhaltigkeit ermöglichen, vorgestellt. Die Informationen bieten ein umfassendes und zukunftsorientiertes Verständnis davon, was nachhaltige Entwicklung ist und was sie sein könnte. Angesichts der planetarischen Grenzen scheint die Dringlichkeit zum Handeln offensichtlich zu sein. Dennoch gibt es zahlreiche Diskurse zum Thema Klimaverzug. Die Möglichkeiten, kritische, aber oft vernachlässigte Interessengruppen einzubinden, zeigen, wie durch die Verbindung von indigener Weisheit mit neuen Technologien von einer kurzfristigen und rein menschenzentrierten Weltsicht zu einer integrativeren und gerechteren gelangt werden könnte
Architectures and systems for identifying assets in circular supply chains using digital product passports and the asset administration shell
Get PDFThe identification and traceability of assets throughout their lifecycle is an important prerequisite for the efficient management of circular value creation and the reuse of components and products. This is a particular problem in complex and dynamic value chains. This paper examines the challenges of identifiers in circular supply chains and presents current solution approaches for identification systems by the means of the Digital Product Pass and the Asset Administration Shell. Architectures are presented for ensuring consistent identification in a circular economy and effective traceability systems in global supply chains. The paper highlights the need for both the physical identification features and the associated digital identification information to make an object consistently and uniquely identifiable