Search Results (154)

Temporal action localization aims to identify the boundaries of the action of interest in a video. Most existing methods take a two-stage approach: first, identify a set of action proposals; then, based on this set, determine the accurate temporal locations of the action of interest. However, the diversely distributed semantics of a video over time have not been well considered, which could compromise the localization performance, especially for ubiquitous short actions or events (e.g., a fall in healthcare and a traffic violation in surveillance). To address this problem, we propose a novel deep learning architecture, namely an adaptive template-guided self-attention network, to characterize the proposals adaptively with their relevant frames. An input video is segmented into temporal frames, within which the spatio-temporal patterns are formulated by a global–Local Transformer-based encoder. Each frame is associated with a number of proposals of different lengths as their starting frame. Learnable templates for proposals of different lengths are introduced, and each template guides the sampling for proposals with a specific length. It formulates the probabilities for a proposal to form the representation of certain spatio-temporal patterns from its relevant temporal frames. Therefore, the semantics of a proposal can be formulated in an adaptive manner, and a feature map of all proposals can be appropriately characterized. To estimate the IoU of these proposals with ground truth actions, a two-level scheme is introduced. A shortcut connection is also utilized to refine the predictions by using the convolutions of the feature map from coarse to fine. Comprehensive experiments on two benchmark datasets demonstrate the state-of-the-art performance of our proposed method: 32.6% mAP@IoU 0.7 on THUMOS-14 and 9.35% mAP@IoU 0.95 on ActivityNet-1.3. Full article

Background/Objectives: Early care and education programs promote children’s social–emotional development, predicting later school success. The COVID-19 pandemic worsened an existing youth mental health crisis and increased teacher stress. Therefore, we applied an infant and early childhood mental health consultation model, Jump Start Plus COVID Support (JS+CS), aiming to decrease behavioral problems in children post-pandemic. Methods: A cluster randomized controlled trial compared JS+CS to an active control, Healthy Caregivers–Healthy Children (HC2), at 30 ECE centers in low-income areas in South Florida. Participants were not blinded to group assignment. Teachers reported on children’s social–emotional development at baseline and post-intervention using the Devereux Early Childhood Assessment and Strengths and Difficulties Questionnaire. We assessed whether teacher stress, classroom practices, and self-efficacy mediated the relationship between JS+CS and child outcomes. We also explored whether baseline behavior problems moderated JS+CS effects on child protective factors, relative to HC2. Results: Direct group-by-time differences between JS+CS and HC2 were limited. However, JS+CS demonstrated significant within-group improvements in teacher-reported child protective factors, behavior support practices, and classroom safety practices. Classroom safety practices consistently mediated positive changes in child behaviors, including the DECA total protective factor score and subdomains of initiative and self-regulation. Additionally, teacher perceptions of behavior support mediated gains in child attachment. Conclusions: JS+CS shows promise in building protective systems around children through intentional support for teachers, underscoring the value of whole-child, whole-environment approaches in early intervention. Full article

Hyperparameter Optimization (HPO) is an important and challenging problem in machine learning. Traditional HPO methods require substantial evaluations to search for superior configurations. Recent Large Language Model (LLM)-based approaches leverage domain knowledge and few-shot learning proficiency to discover promising configurations with minimal human effort. However, the repetition issues causes LLMs to generate configurations similar to context examples, which may confine the optimization process to local regions. Moreover, since LLMs rely on the examples they generate for a few-shot learning, a self-reinforcing loop is formed, hindering LLMs from escaping local optima. In this work, we propose EvoContext, which aims to intentionally generate configurations that differ significantly from examples via external interventions and actively breaks the self-reinforcing effect for a more efficient approximation of the global optimum. Our EvoContext method involves two phases: (i) initial example generation through cold or warm starting and (ii) iterative optimization that integrates genetic operations for updating examples to enhance global exploration capabilities. Additionally, it employs LLMs in-context learning to generate configurations based on competitive examples for local refinement. Experiments on several real-world datasets show that EvoContext outperforms traditional and other LLM-driven approaches on HPO. Full article

Trust management provides an alternative solution for securing open, dynamic, and distributed multi-agent systems, where conventional cryptographic methods prove to be impractical. However, existing trust models face challenges such as agent mobility, which causes agents to lose accumulated trust when moving across networks; changing behaviors, where previously reliable agents may degrade over time; and the cold start problem, which hinders the evaluation of newly introduced agents due to a lack of prior data. To address these issues, we introduced a biologically inspired trust model in which trustees assess their own capabilities and store trust data locally. This design improves mobility support, reduces communication overhead, resists disinformation, and preserves privacy. Despite these advantages, prior evaluations revealed the limitations of our model in adapting to provider population changes and continuous performance fluctuations. This study proposes a novel algorithm, incorporating a self-classification mechanism for providers to detect performance drops that are potentially harmful for service consumers. The simulation results demonstrate that the new algorithm outperforms its original version and FIRE, a well-known trust and reputation model, particularly in handling dynamic trustee behavior. While FIRE remains competitive under extreme environmental changes, the proposed algorithm demonstrates greater adaptability across various conditions. In contrast to existing trust modeling research, this study conducts a comprehensive evaluation of our model using widely recognized trust model criteria, assessing its resilience against common trust-related attacks while identifying strengths, weaknesses, and potential countermeasures. Finally, several key directions for future research are proposed. Full article

Using data from the 2022 and 2023 National Survey on Drug Use and Health, we examined factors associated with treatment use for substance use disorder (SUD), perceived SUD treatment needs, and reasons for treatment non-use. Of U.S. adults, 18.1% had any past-year SUD (alcohol use disorder [AUD] and/or any drug use disorder [DUD]), 14.4% of those with SUD received SUD treatment in the past year, and 5.5% of those who did not receive treatment had a perceived need for treatment. Treatment use was significantly associated with AUD and DUD severities (aOR = 3.85, 95% CI = 2.82–5.26 for severe AUD; aOR = 2.82, 95% CI = 2.27–3.47 for severe DUD), problem self-perception (aOR = 2.12, 95% CI = 1.74–2.58), and mental health treatment use (aOR = 6.07, 95% CI = 4.73–7.78). Perceived treatment needs among those who did not use treatment were also significantly associated with AUD and DUD severities, problem self-perception, and any mental illness. The most frequently reported reasons for treatment non-use among those with perceived need were self-sufficiency beliefs, lack of readiness to stop using or start treatment, stigma-related concerns, and health insurance/cost problems. The findings underscore the importance of screening SUD and educating about the harms of untreated SUD in increasing motivation and readiness for treatment use among people with SUD. Full article

This study investigates the reliability of Francis turbines and highlights the critical need for an improved deformation monitoring system for bolts that fasten a hydroturbine head-cover to its casing. During different operational stages of the hydraulic unit, such as start-up, partial load, and full load, the hydroturbine head-cover and its fastening bolts are subjected to static and cyclic loads. The loads generate vibrations and different deformations that must be monitored. Although various measuring instruments, such as vibration sensors and accelerometers, have been developed to monitor hydroturbine vibrations, only two systems—KM-Delta-8-CM and PTK KM-Delta—are currently applied to measure fastening bolt deformation. Furthermore, only one system, SKDS-SISH, was found to monitor the forces inducing this deformation. After analysis, it is evident that the described systems for monitoring the deformation of the fastening bolts do not guarantee the trustworthiness of the measuring sensors and there is a need for their improvement. The implementation of a self-checking function (including metrological features), the development of a digital twin of the sensor, and the application of technologies based on artificial intelligence could solve this problem. Full article

Many females experience physical problems caused by menstruation, such as menstrual cramps and premenstrual syndrome, which disrupt their daily lives and work. Knowing when menstruation begins is essential for managing such physical conditions. However, menstrual periods are not always cyclic and can be extended by physical and mental stress. Currently used menstrual management applications rely on self-reported cycle length and basal body temperature (BBT), which makes it challenging to predict irregular periods. Advances in smart wearables have made continuous, non-invasive health monitoring accessible, such as heart rate variability (HRV). HRV characteristics reflect autonomic nervous system activity and are used as physical and mental health status indices. This study aims to explore the relationship between HRV indices and the menstrual cycle using smart wearables. A total of 13 females aged from 18 to 20 participated in this study and measured their indices using an underwear-type wearable device for six months. The device measured HRV and body acceleration. Participants recorded their BBT every morning and answered questionnaires about their physical and mental status every morning and evening. They also reported the start and end dates of menstruation. The HRV data were split into sleep and wake phases using acceleration and calculated time- and frequency-domain HRV indices. Cross-correlation and regression analysis were conducted to assess the relation between the menstrual cycle and phases, such as follicular and luteal, and the HRV indices. A significant relationship between HRV indices and the menstrual cycle length was found, particularly in the difference between the follicular and luteal phases of HRV indices. This difference showed a relatively high association with menstrual cycle length. Importantly, the regression analysis results suggested that HRV indices can be used to predict the length of the menstrual cycle and potential physical and mental disorders. These findings significantly contributed to menstrual health management and the Femtech industry. Full article

Traditional aging education at our institution in Taiwan has primarily relied on lecture-based instruction, emphasizing teacher-centered knowledge transmission. Although effective in delivering foundational theories, this approach often overlooks active student engagement which is crucial for developing critical thinking, self-confidence, and problem-solving skills. These methods focus on transmitting theories and skills while often neglecting the cultivation of a willingness to serve older adults. This study investigates the impact of integrating diverse teaching strategies into aging education courses to enhance caregiving competence, defined as a multidimensional construct comprising critical thinking, self-confidence, problem-solving ability, and willingness to serve older adults. A quasi-experimental design was used for first-year students from the Department of Healthcare Administration at a university in Taiwan. Participants were divided into traditional (111th academic year) and diverse (112th academic year) teaching groups during gerontology courses. The traditional group employed lecture-based instruction focusing on knowledge transmission, whereas the diverse group utilized flipped teaching, case or story discussions, and expert lectures, emphasizing active learning, situated learning, and reflective practices based on constructivist learning theory. Data were collected via questionnaires at the semester’s start and end. The results showed significant improvements in critical thinking, self-confidence, problem-solving skills, and willingness to serve older adults within the diverse group (p < 0.05). However, no significant differences were found between the two groups in these measures. These findings indicate that while diverse teaching strategies effectively enhance caregiving competence, their outcomes are not significantly different from those of traditional methods. Full article

The change-point model is an established methodology for the construction of self-starting control charts. Change-point charts are often nonparametric in order to be independent from any specific assumptions about the process distribution. Nonetheless, this methodology is usually implemented by considering all possible splits of a given stream of observations into two adjacent sub-samples. This can make the recent observations too influential and the chart’s signals too dependent on limited evidence. This paper proposes to correct such a distortion by using a window approach, which forces the use of only comparisons based on sub-samples of the same size. The resulting charts are “omnibus”, with respect to their having any kind of shift and also any direction of such shifts. To prove this, this paper focuses on a chart based on the Cramér–von Mises test. We report a simulation study evaluating the average number of readings to obtain a signal after a known shift has occurred. We conclude that, beyond being stable with respect to the direction of the shift, the new chart overcomes its competitors when the distribution heads toward regularity. Finally, the new approach is shown to have successful application to a real problem about air quality. Full article

In this work, we present the semi-analytical solution for the early-time recession phase of the horizontal unconfined aquifer of finite length for steady-state initial conditions. This is a case where self-similarity arguments are not applicable. The solution is built as linear perturbations from the initial steady state. The solution is determined via a Sturm–Liouville eigenvalue problem, which should be solved numerically. On the other hand, the immediate response of the aquifer to the sudden switching off the recharge, i.e., in the earliest times, is obtained by deducing analytically the large eigenvalue asymptotic solutions of the problem. We find analytically that in this time regime, the outflow Q is given by Q = Q₀ − 1.4Q₀^5/3L^−4/3n^−2/3t^2/3, where Q₀ is the initial outflow rate, L is the length of the aquifer, n is the porosity of the formation and t is the time from the start of the recession. The stated result is very accurate for times t up to ~0.01 nL^3/2k^−1/2Q₀^−1/2, where k is the hydraulic conductivity of the formation. The analytical and quantitative relation of the presented solution with the classical recession phase asymptotic solutions derived in the past by Polubarinova (early-time solution) and Boussinesq (separable, late-time solution) is discussed in detail. The presented results can be used as a benchmark solution for modeling or numerical validation purposes. Full article

As an approach, coaching-based models have been demonstrated to enhance student self-efficacy, improve grades, and increase retention and graduation rates. Coaching-based training models are also key in mentor development, focusing on open-ended questions and active listening to create supportive environments where mentees can independently find solutions. This approach not only builds mentors’ communication and leadership skills but also enhances their adaptability and problem-solving abilities. For underrepresented groups in STEM, such training positions mentors as knowledge facilitators, helping bridge gaps in mentorship experiences and bolstering confidence in their roles, thereby contributing to a more inclusive and effective learning ecosystem. This study investigates the impact of a coaching-based approach to near-peer mentor training within the UK START program, focusing on high school student participants. Interviews revealed significant benefits, including enhanced communication skills, particularly in asking open-ended questions and avoiding judgmental language. Mentors also reported improved composure in stressful situations, often utilizing techniques such as deep breathing to manage emotions during interactions with young campers. Additionally, participants experienced personal growth, seeing themselves as leaders and role models, which they attributed to the mentorship training. The role affirmed their confidence in their STEM knowledge and sparked interest in future mentorship roles. These findings suggest that structured coaching-based training can build a supportive environment, benefiting both mentors and mentees. Full article

This paper addresses the time-dependent vehicle routing problem with drones in vehicle-restricted zones and no-fly zones (TDVRPD-VRZ-NFZ). The optimization model considers the impacts of vehicle-restricted zones, no-fly zones, and time-dependent road networks on delivery paths. The objective is to minimize the total cost, including vehicle dispatch costs, energy consumption costs for vehicles and drones, and time-window penalty costs. The model is verified for correctness using Gurobi. In response to the problem’s characteristics, a hybrid genetic algorithm and variable neighborhood search with a learning mechanism (HGAVNS-LM) is proposed to solve the problem. The algorithm starts by generating the initial population using a combination of logistic mapping and reverse learning. It then improves the genetic operators and variable neighborhood search operators to optimize the initial population. To improve the algorithm’s performance, an individual elite archive is used for knowledge learning, and a self-learning mechanism is established to dynamically adjust the algorithm’s key parameters. The solution obtained by HGAVNS-LM shows a deviation of −0.2% to −0.3% compared to Gurobi, but it saves 99.68% in solving time. Compared to the genetic neighborhood search algorithm and the hybrid genetic algorithm, the improvement rates are 5.1% and 13.0%, respectively. Through the analysis of multiple sets of test cases, it is concluded that time-varying road networks, vehicle-restricted zones and no-fly zones, and different detour rules all affect delivery costs and delivery plans. The research results provide a more scientific theoretical basis for logistics companies to customize delivery solutions. Full article

Introduction: Antibiotic resistance is a major public health problem in Europe. Most antibiotics are sold only by prescription in Poland, and it is mainly up to physicians to decide whether to start antibiotic treatment. Therefore, we analyzed the factors influencing the prescribing of antibiotics for upper respiratory tract infections by primary care physicians in Poland, attitudes toward antibiotic resistance, and knowledge of the principles of antibiotic use. Methods: We conducted a CAWI (Computer-Assisted Web Interview) survey, carried out using a proprietary survey distributed online. Results: A total of 528 doctors participated in the study. The result of the physical examination and additional tests, as well as the recommendations of scientific societies are the most important in deciding whether to start antibiotic therapy. Patient pressure (p < 0.011) and workload (p = 0.021) significantly influenced the decision to prescribe an antibiotic among primary care physicians and physicians in the course of specialization, who fear of legal consequences (p < 0.001). The habits of other physicians (p < 0.001) working at the same facility appeared to be additionally important. Conclusions: The decision to implement antibiotic therapy in upper respiratory tract infections is influenced by several factors that depend on the doctor (including place of work and seniority) and the patient (clinical symptoms, expectation of antibiotic prescription). The physician’s level of knowledge contributes to reducing antibiotic prescribing. Considering the factors associated with the level of knowledge and awareness, together with a high prevalence of self-medication with antibiotics in Polish population, there is a strong need to design educational interventions aimed at reducing inappropriate antibiotic prescribing and preventing antibiotic resistance in Poland. Full article

Background: The Mediterranean Diet is recognized as one of the healthiest dietary patterns; however, in recent years, a decline in adherence has been observed in Mediterranean countries. University students represent a particularly vulnerable population, as starting university introduces new influences and responsibilities that directly impact their lifestyle and health. Objective: Analyze adherence to the Mediterranean Diet among university students and its association with other lifestyle habits and mental and physical health indicators. Methods: A cross-sectional study was conducted with a sample of 1268 students (23.65 ± 7.84 years) from a university in northern Spain between November 2020 and March 2021. An online questionnaire was administered to assess Mediterranean Diet adherence along with variables such as perceived stress, self-esteem, life satisfaction, suicidal behavior, emotional and behavioral problems, emotional intelligence, physical activity, sedentary behavior, alcohol consumption, and compulsive internet use. Results: 29.26% of students had high adherence to the Mediterranean Diet. Regression analysis indicated that high adherence was associated with higher levels of emotional intelligence, as well as lower levels of suicidal ideation, emotional problems, and compulsive internet use. Conclusions: The associations found between Mediterranean Diet and other indicators and lifestyle habits highlight the need for interdisciplinary promotion strategies within the university ecosystem. Full article

The accurate identification of channel-coding types plays a crucial role in wireless communication systems. The recognition of convolutional codes presents challenges, primarily due to their strong temporal dependencies, varying constraint lengths, and additional contamination from noise. However, existing algorithms often rely on manual feature extraction or are limited to a restricted number of coding types, rendering them inadequate for practical applications. To tackle this problem, we propose ConvLSTM-TFN (temporal feature network), an innovative blind-recognition network that integrates convolutional layers, long short-term memory (LSTM) networks, and a self-attention mechanism. The proposed approach enhances the acquisition of features from soft-decision sequence information, leading to improved recognition performance without necessitating prior knowledge of coding parameters, sequence starting positions, or other metadata. The experimental results demonstrate that our method is effective within a signal-to-noise ratio (SNR) range of 0 to 20 dB, achieving more than 90% recognition accuracy across 17 convolutional code types, with an average accuracy of 98.7%. Our method effectively distinguishes diverse coding features, surpassing existing models and establishing a new benchmark for channel-coding recognition. Full article