FedISM: Enhancing Data Imbalance via Shared Model in Federated Learning

Considering the sensitivity of data in medical scenarios, federated learning (FL) is suitable for applications that require data privacy. Medical personnel can use the FL framework for machine learning to assist in analyzing large-scale data that are protected within the institution. However, not all clients have the same distribution of datasets, so data imbalance problems occur among clients. The main challenge is to overcome the performance degradation caused by low accuracy and the inability to converge the model. This paper proposes a FedISM method to enhance performance in the case of Non-Independent Identically Distribution (Non-IID). FedISM exploits a shared model trained on a candidate dataset before performing FL among clients. The Candidate Selection Mechanism (CSM) was proposed to effectively select the most suitable candidate among clients for training the shared model. Based on the proposed approaches, FedISM not only trains the shared model without sharing any raw data, but it also provides an optimal solution through the selection of the best shared model. To evaluate performance, the proposed FedISM was applied to classify coronavirus disease (COVID), pneumonia, normal, and viral pneumonia in the experiments. The Dirichlet process was also used to simulate a variety of imbalanced data distributions. Experimental results show that FedISM improves accuracy by up to 25% when privacy concerns regarding patient data are rising among medical institutions.

The Influence of Distance Education and Peer Self-Regulated Learning Mechanism on Learning Effectiveness, Motivation, Self-Efficacy, Reflective Ability, and Cognitive Load

COVID-19 has resulted in the increased use of distance learning around the world. With the advancement of information technology, traditional classroom teaching has gradually integrated the Internet and distance learning methods. Students need to be able to learn on their own in a distance learning environment, so their ability to self-regulate their learning in a distance learning environment cannot be ignored. However, in previous studies on self-regulated learning, most learners learn alone. When they have academic doubts, they cannot obtain help and support from their studies, resulting in reduced learning outcomes. This study uses the peer self-disciplined learning mechanism to establish a distance teaching system that assists students and to improve their own learning status by meeting with peers at a distance. It can also help learners orient themselves by observing their peers' learning status and goal considerations. The participants in this study were 112 college students in the department of information management. The control group used a general self-regulated teaching system for learning, and the experimental group used a distance learning system, incorporating peer self-regulated learning. The results of the study found that learners who used the distance peer learning mechanism were more effective than those who used the general distance self-regulated learning system;learners who used the distance peer-regulated learning mechanism had better motivation, self-efficacy, and reflection after the learning activity than those who used the general distance self-regulated learning system. In addition, with the aid of such mechanisms, learners' cognitive load can be reduced, and learning effectiveness can be improved. © 2023 by the authors.

Family income loss and child health inequality during the COVID-19 pandemic in Taiwan

Objectives: This study aimed to examine the relationship between family income loss and child health during the COVID-19 pandemic. Methods: Data for the analysis were obtained from the Taiwan Birth Cohort Study, a nationally representative sample of babies born in 2005, and 18, 024 caregivers participated in the survey as their children aged 15. In analysis, we first conducted descriptive analyses to test the correlation between socioeconomic variables and family income loss. We next assessed whether there was a gradient relationship between family income loss and child health using Cochran-Armitage trend test. Finally, multiple logistic regression was used to estimate the relationship between family income loss and child health. Results: Our findings indicated that (1) lower socioeconomic families were at a greater risk of suffering income loss during COVID-19;(2) children in the families experiencing a more severe loss of income had worse health, but the gradient relationship was not significant for those in higher income families;and (3) significantly higher risk of fair/poor health of children was found in the severe (OR: 1.3, 95% CI 1.2-1.5) and mild (OR: 1.2, 95% CI 1.1-1.3) income loss groups than in the no income loss group after adjustment for socioeconomic variables. Conclusions: Family income loss due to COVID-19 was significantly associated with child health inequality. To avoid widening the health gap, children in families experiencing financial impacts during the COVID-19 pandemic should be protected and supported, particularly those in lower socioeconomic groups. © 2022 Chinese Public Health Association of Taiwan. All rights reserved.

Adaptable eye tracking technology of obscured face

Humans around the world are affected by special infectious pneumonia (COVID-19). There are more and more people wearing masks that are necessary for daily or workplace use. However, the sensitivity of face detection will be affected by feature obscuration, and most of them cannot be performed. Obscured face detection and gaze tracking. This paper proposes a face detection and landmark repair, and then realizes the tracking of the eye trajectory of the obscured face. Model database with obscured face image data can also include unobscured face image data. After calibrated eye area, machine learning [1] algorithm is used for model database training to achieve eye area detection and provide real-time position coordinates. The eye information of the partial simulation model is superimposed and calculated to complete the feature point restoration, feature point detection and definition. Finally, K-means [2] is used to classify the image around the eyes to distinguish the eyeball from the white of the eye and calculate the position of the eyeball center. The face wearing a mask will affect the sensitivity of face detection, and the person wearing a mask cannot be detected. We use a two-stage method to locate the eyeball center of the face wearing a mask. We use the machine learning algorithm to detect the bounding box near the eyes, and we use the obscured image to train our model. Then attach the chin pattern to the place that is expected to be covered. Use a general cross-platform machine learning library [3] to locate area near the eyeball. Then use an unsupervised learning clustering algorithm to classify the image near the eyeball to analyze the eyeball area and find the center of the eyeball, to achieve the purpose of eye tracking. © 2020 ITE and SID.

Enhancing the fluorescence and cycle threshold of qPCR devices through excitation time point adjustment

Quantitative polymerase chain reaction (qPCR) has been widely employed for the positive or negative detection of bacteria or viruses, particularly SARS-CoV-2. Fluorescence signal and cycle threshold information is critical for the positive and negative detection of target test samples in qPCR systems. To determine viral concentration, the fluorescence intensity of each cycle must be recorded using a qPCR system. In general, the time points of fluorescence excitation and excitation light intensity affect fluorescence intensity. Thus, this study proposed an effective excitation method for enhancing fluorescence intensity. Several parameters, including excitation light intensity, the excitation time point, and the reaction time of the reagent at each temperature stage, were modified in assessing fluorescence performance and determining suitable parameters for fluorescence excitation in a qPCR system. Fluorescence intensity resulted in the most optimal fluorescence performance;specifically, excitation was triggered by using a 30 mA current, and the excitation light was activated when the temperature decreased to 60 °C. Total reaction time was 1 s, and the concentrated fluorescence value and suitable cycle threshold value were obtained. Overall, high efficiency, low fluorescence decay, and high light stability were observed. The present findings demonstrate that controlling the time point of excitation light can enhance the fluorescence efficiency and performance of qPCR systems, with relevant benefits in medical diagnostics and rapid viral detection, among other applications. © 2021 IEEE.