Search

Since the outbreak of COVID-19, many efforts have been made to utilize the respiratory sounds and coughs collected by smartphones for training Machine Learning models to classify and distinguish COVID-19 sounds from healthy ones. Embedding those models into mobile applications or Internet of things devices can make effective COVID-19 pre-screening tools afforded by anyone anywhere. Most of the previous researchers trained their classifiers with respiratory sounds such as breathing or coughs, and they achieved promising results. We claim that using special voice patterns besides other respiratory sounds can achieve better performance. In this study, we used the Coswara dataset where each user has recorded 9 different types of sounds as cough, breathing, and speech labeled with COVID-19 status. A combination of models trained on different sounds can diagnose COVID-19 more accurately than a single model trained on cough or breathing only. Our results show that using simple binary classifiers can achieve an AUC of 96.4% and an accuracy of 96% by averaging the predictions of multiple models trained and evaluated separately on different sound types. Finally, this study aims to draw attention to the importance of the human voice alongside other respiratory sounds for the sound-based COVID-19 diagnosis.

The application of cross-dataset training in object detection tasks is complicated because the inconsistency in the category range across datasets transforms fully supervised learning into semi-supervised learning. To address this problem, recent studies focus on the generation of high-quality missing annotations. In this study, we first specify that it is not enough to generate high-quality annotations using a single model, which looks only once for annotations. Through detailed experimental analyses, we further conclude that hard-label training is conducive for generating high-recall annotations, whereas soft-label training tends to obtain high-precision annotations. Inspired by the aspects mentioned above, we propose a dynamic supervisor framework that updates the annotations multiple times through multiple-updated submodels trained using hard and soft labels. In the final generated annotations, recall and precision improve significantly through the integration of hard-label training with soft-label training. Extensive experiments conducted on various dataset combination settings support our analyses and demonstrate the superior performance of the proposed dynamic supervisor.

Drought forecasting can effectively reduce the risk of drought. We proposed a hybrid model based on deep learning methods that integrates an autoregressive integrated moving average (ARIMA) model and a long short-term memory (LSTM) model to improve the accuracy of short-term drought prediction. Taking China as an example, this paper compares and analyzes the prediction accuracy of six drought prediction models, ARIMA, support vector regression (SVR), LSTM, ARIMA-SVR, least square-SVR (LS-SVR) and ARIMA-LSTM, for SPEI. The performance of all the models was compared using measures of persistence, such as the Nash-Sutcliffe efficiency (NSE) and so on. The results show that all three hybrid models (ARIMA-SVR, LS-SVR and ARIMA-LSTM) had higher prediction accuracy than the single model. (ARIMA, SVR and LSTM), for a given lead time, at different scales. the NSEs of the hybrid ARIMA-SVR, LS-SVR and ARIMA-LSTM models for the predicted SPEI1 are 0.043,0.168 and 0.368, respectively, and the NSEs of SPEI24 is 0.781, 0.543 and 0.93, respectively. This finding indicates that when the lead time remains unchanged, the prediction accuracy of the hybrid ARIMA-SVR, LS-SVR and ARIMA-LSTM models for the SPEI at various scales is gradually improved with increasing time scale, and the prediction accuracy of the model with a one-month lead time is higher than that of the model with a two-month lead time. In addition, the ARIMA-LSTM model has the highest prediction accuracy at the 6-, 12-, and 24-month scales, indicating that the model is more suitable for the forecasting of long-term drought in China.

The aim of this paper is to develop a system for drug–target interaction prediction using artificial intelligence which involves development of both machine learning and deep learning-based systems. In this paper, we use a convolutional neural network (CNN) model, to classify drug–target interactions between drug pairs. Applied to the DDI-Corpus dataset, the single CNN model achieve performance with an F1-score of 0.82 ± 0.012 for the single model and 0.81 ± 0.015 for the ensemble model using deep learning-based CNN with an approved accuracy of 96.72% which is an extra-ordinary achievement. This work has also been performed using the machine learning-based classifiers support vector machine (SVM). For machine learning-based implementation, drug-bank dataset was used for the training and testing. The main challenge when using machine learning for this purpose is the availability of negative DTI to train on. Training machine learning model, the SVM achieved an area under the ROC curve (AUC) of 0.753 ± 0.006, which taking the difference in computational resources into consideration compares well to the AUC of 0.886 ± 0.010 network-based state-of-the-art approach. We achieved and best accuracy of 93.76% using SVM after testing several times.

Background: Evaluating the quality of life patients with subacromial impingement syndrome is important for further treatments Aim: The aim of the study was to evaluate simultaneously the quality of life with subacromial impingement syndrome using different parameters in a single model. Material and Methods: Ninety patients diagnosed subacromial impingement syndrome were included in the study. Mental component score and physical component score were measured using SF-36 in the assesment of the quality of life and different scores were evaluated. The indirect and direct effects of these tests on the quality of life were examined in a single model using the structural equation modelling. Results: The most total effect on physical component score was Shoulder Disability Questionnaire and the Quick Disabilities of the Arm, Shoulder and Hand Score, Visual Analog Score had the least total effect on physical component score. Considering the results of mental component score, the total most effect was VAS had the the minimum total effect. Conclusion: The quality of life of the patients with subacromial impingement syndrom give more detailed informations when analysing the scores dependently. Evaluating the quality of life of subacromial impingement syndrome, using different measures of patients dependently will provide a more detailed research opportunity. If the patient quality of life evaluation is done in detail, it will facilitate the selection of appropriate treatment.

Wind serves as natural resources as the solution to minimize global warming and has been commonly used to produce electricity. Because of their uncontrollable wind characteristics, wind speed forecasting is considered one of the best challenges in developing power generation. The Autoregressive Integrated Moving Average (ARIMA), Simple Exponential Smoothing (SES) and a hybrid model combination of ARIMA and SES will be used in this study to predict the wind speed. The mean absolute percentage error (MAPE) and the root mean square error (RMSE) are used as measurement of efficiency. The hybrid model provides a positive outcome for predicting wind speed compare to the single model of ARIMA and SES.

Preclinical testing platforms have been instrumental in the research and development of thrombectomy devices. However, there is no single model which fully captures the complexity of cerebrovascular anatomy, physiology, and the dynamic artery-clot-device interaction. This article provides a critical review of phantoms, in-vivo animal, and human cadaveric models used for thrombectomy testing and provides insights into the strengths and limitations of each platform. Articles published in the past 10 years that reported thrombectomy testing platforms were identified. Characteristics of each test platform, such as intracranial anatomy, artery tortuosity, vessel friction, flow conditions, device-vessel interaction, and visualization, were captured and benchmarked against human cerebral vessels involved in large-vessel occlusion stroke. Thrombectomy phantoms have been constructed from silicone, direct 3D-printed polymers, and glass. These phantoms represent oversimplified patient-specific cerebrovascular geometry but enable adequate visualization of devices and clots under appropriate flow conditions. They do not realistically mimic the artery-clot interaction. For the animal models, arteries from swine, canines, and rabbits have been reported. These models can reasonably replicate the artery-clot-device interaction and have the unique value of evaluating the safety of thrombectomy devices. However, the vasculature geometries are substantially less complex and flow conditions are different from human cerebral arteries. Cadaveric models are the most accurate vascular representations but with limited access and challenges in reproducibility of testing conditions. Multiple test platforms should be likely used for comprehensive evaluation of thrombectomy devices. Interpretation of the testing results should take into consideration platform-specific limitations.

In the article we present a single-run approach to recognizing nested named entities using neural networks with transformers. The main advantage of this approach is that a single model is trained to recognize all entity types. The model can identify all entities in a single pass. Our main contribution is the simplified representation of nested named entities on the token level, and evaluation of the presented approach on three languages — Polish (PolEval 2018 dataset), German (GermEval 2014 dataset), and Czech (CNEC 2.0 dataset). For each dataset we obtained state-of-the-art results. For Polish and German we obtained a significant improvement — 1.4 pp and 3.0 pp respectively. For Czech we obtained an improvement of 0.5 pp.

In artificial intelligence (AI), knowledge is the information required by an intelligent system to accomplish tasks. While traditional knowledge bases use discrete, symbolic representations, detecting knowledge encoded in the continuous representations learned from data has received increasing attention recently. In this work, we propose a method for building a continuous knowledge base (CKB) that can store knowledge imported from multiple, diverse neural networks. The key idea of our approach is to define an interface for each neural network and cast knowledge transferring as a function simulation problem. Experiments on text classification show promising results: the CKB imports knowledge from a single model and then exports the knowledge to a new model, achieving comparable performance with the original model. More interesting, we import the knowledge from multiple models to the knowledge base, from which the fused knowledge is exported back to a single model, achieving a higher accuracy than the original model. With the CKB, it is also easy to achieve knowledge distillation and transfer learning. Our work opens the door to building a universal continuous knowledge base to collect, store, and organize all continuous knowledge encoded in various neural networks trained for different AI tasks.

Recently, Generative Adversarial Network (GAN) based approaches are applied in facial attribute translation. However, many tasks, i.e. multi facial attributes translation and background invariance, are not well handled in the literature. In this paper, we propose a novel GAN-based method that aims to get the target image that performs better within modifying one or more facial attributes in a single model. The model generator learns multi-points by inputting a re-coded transfer vector, ensuring the single model could learn multiple attributes simultaneously. It also optimizes the cycle loss to enhance the efficiency of transferring multi attributes. Moreover, the method uses the adaptive parameter to improve the calculation method of the loss function of the residual image. The results are also compared with the StarGAN v2, which is the current state-of-the-art model to prove the effectiveness and advancedness. Experiments show that our method has a satisfactory performance in multi facial attributes translation.

In this paper we demonstrate that $\textit{context free grammar (CFG) based methods for grammar induction benefit from modeling lexical dependencies}$. This contrasts to the most popular current methods for grammar induction, which focus on discovering $\textit{either}$ constituents $\textit{or}$ dependencies. Previous approaches to marry these two disparate syntactic formalisms (e.g. lexicalized PCFGs) have been plagued by sparsity, making them unsuitable for unsupervised grammar induction. However, in this work, we present novel neural models of lexicalized PCFGs which allow us to overcome sparsity problems and effectively induce both constituents and dependencies within a single model. Experiments demonstrate that this unified framework results in stronger results on both representations than achieved when modeling either formalism alone. Code is available at https://github.com/neulab/neural-lpcfg., Accepted at TACL 2020

Accurate prediction of bus delays improves transit service delivery and can potentially increase passenger use and satisfaction. To date, models developed for predicting bus delays have been restricted to single routes because of their poor performance on a wide network, due to reliance on simplistic model architectures and limited sources of data. This paper proposes a deep learning-based framework for predicting bus delays at the network level. The framework is fueled by large, heterogenous bus transit data (GTFS) and vehicle probe data. We utilize entity embeddings to enable the framework to simultaneously fit functions and learn patterns from both categorical and continuous data streams. The framework results in a single model that is able to characterize the factors influencing delays on multiple routes, for multiple stations at a time, at different times of the day, and during different seasons. A case study was conducted in Saint Louis, Missouri, with data collected over a 1-month period. The results indicate that the developed modeling framework is high performing, predicting delays for multiple stops at a mean absolute percentage error (MAPE) of about 6%. For different routes and trips, the observed prediction errors were stable across days of week, bus stops, holidays, and time of day. Although peak hour factors and the distance to bus stop influenced the model’s prediction errors for some routes, the observed differences were not significant. Compared to previous research, the ability to simultaneously model continuous and categorical data with deep learning and the use of heterogenous data contributed to such high performance on multiple routes.

Air quality forecast is an important technical means to ensure timely and proper response to heavy pollution weather. In this study, a hybrid deep air quality predictor (HDAQP) model consisting of one-dimensional convolutional neural network (CNN), long short-term memory (LSTM), and deep neural network (DNN) is proposed to forecast air quality indicators (mainly PM2.5 concentrations). The proposed model can overcome the limitations of the single model and meanwhile make the best of each. CNN model is used to convolve the historical PM2.5 concentration data along with meteorological data to extract shallow features, while LSTM model is used to extract the deep temporal features. Finally, the DNN model is adopted to transfer these deep features into the final forecast results. Compared with the mainstream deep learning models (e.g., RNN, LSTM, and CNN-LSTM models), the HDAQP model exhibits a better performance in short-term PM2.5 concentration forecast. With the increase of prediction time, the long-term prediction performance of the HDAQP model will be degraded, but it is still better than the mainstream deep learning models. Moreover, considering other meteorological factors, with the multi-source data, the HDAQP model can forecast PM2.5 concentrations more accurately.

One of the hottest and most popular methods in applied Machine Learning is Ensemble methods.Ensemble combines predictions from different models to generate a final prediction with better performance than any other single model. The research focused on the implementation of Ensemble method for predicting student academic performance based on their personal characteristics, family background, infrastructural environment in the college and external environment, etc...Our study uses RandomForestClassifier, Logistic Regression, and ExtraTreesClassifier as the Base Learners and AdaBoost Classifier as the Meta Learner. This result helps in predicting the accuracy of students’ academic performance and also in identifying the poor performers, so that early measures prior to final semester examination can be deployed

Image denoising is a thoroughly studied research problem in the areas of image processing and computer vision. In this work, a deep convolution neural network with added benefits of residual learning for image denoising is proposed. The network is composed of convolution layers and ResNet blocks along with rectified linear unit activation functions. The network is capable of learning end‐to‐end mappings from noise distorted images to restored cleaner versions. The deeper networks tend to be challenging to train and often are posed with the problem of vanishing gradients. The residual learning and orthogonal kernel initialisation keep the gradients in check. The skip connections in the ResNet blocks pass on the learned abstractions further down the network in the forward pass, thus achieving better results. With a single model, one can tackle different levels of Gaussian noise efficiently. The experiments conducted on the benchmark datasets prove that the proposed model obtains a significant improvement in structural similarity index than the previously existing state‐of‐the‐art techniques.

Areas that contain ecologically distinct biological content, called bioregions, are a central component to spatial and ecosystem‐based management. We review and describe a variety of commonly used and newly developed statistical approaches for quantitatively determining bioregions. Statistical approaches to bioregionalization can broadly be classified as two‐stage approaches that either ‘Group First, then Predict’ or ‘Predict First, then Group’, or a newer class of one‐stage approaches that simultaneously analyse biological data with reference to environmental data to generate bioregions. We demonstrate these approaches using a selection of methods applied to simulated data and real data on demersal fish. The methods are assessed against their ability to answer several common scientific or management questions. The true number of simulated bioregions was only identified by both of the one‐stage methods and one two‐stage method. When the number of bioregions was known, many of the methods, but not all, could adequately infer the species, environmental and spatial characteristics of bioregions. One‐stage approaches, however, do so directly via a single model without the need for separate post‐hoc analyses and additionally provide an appropriate characterization of uncertainty. One‐stage approaches provide a comprehensive and consistent method for objectively identifying and characterizing bioregions using both biological and environmental data. Potential avenues of future development in one‐stage methods include incorporating presence‐only and multiple data types as well as considering functional aspects of bioregions.

The clothing piece passes through different manufacturing stages, the most important of which is the staging stage, which is to obtain more than one measurement by drawing a single model for the pattern, then its graduation to obtain multiple measurements and a high degree of quality, and the degree of the pattern can be obtained through manual staging and staging by computer, Despite the technological advances of the garment industry, However, he is still using the manual grading to obtain the different measurements of the model, especially in specialized educational institutions and small industrial workshops that do not have an integrated manufacturing unit, so they resort to the manual grading process to obtain the different measurements of the model and this method requires more time and effort so it does not fit With the major garment factories that need to speed up production due to the large numbers that produce them according to the requirements of the market, Hence, the research problem is clear through what is how to overcome some of the problems of manual grading, which confronts the clothes manufacturers in small industrial workshops in order to obtain a high-quality product, the research aims to reach the best methods of manual grading for women in proportion to the quality of manufacturing, as well as obtain Ali is a high-quality women's robe finished product, The research assumes that there are statistically significant differences between the opinions of the arbitrators about the samples produced by the research and the most important results are that there are statistically significant differences between the samples produced according to the results of the arbitration, and the best methods of manual grading of the female rub have been reached and obtain the highest possible degree of quality, The researcher recommended shedding light on the various difficulties in the stages of manufacturing clothes and trying to provide appropriate solutions for them that employers face in small industrial workshops.

PurposeThis research examines the influence of the learning environment and students' attitudes towards choosing an accounting career mediated by intention to enhance the current knowledge.Design/methodology/approachThe online survey is adopted to test the research model of this research. There are 503 useable responses collected with the effective response rate of 72 per cent. Data analysis and hypothesis testing use Partial Least Square as part of the Structural Equation Modelling technique.FindingsThe results of this research indicate that accounting students possess positive attitude both towards the intention to enhance the current knowledge and choosing their accounting career. The learning environment includes educators and friends who give significant influence on students' intention. Besides, current knowledge enhancement is also discovered to be able to mediate the link between attitude, learning environment and intention to choose accounting career.Research limitations/implicationsDifferent generations may generate either different perception or different orientation in choosing accounting career. Therefore, future research can consider wider coverage and more updated object.Practical implicationsFindings of this research suggest that periodic improvement and renewal are necessary to administer to create an optimum learning environment, in term of teachers' capacity, teaching materials and supporting social environment.Originality/valueThis research contributes to any research related to attitudes towards choosing an accounting career. This study is the leading study that combines students' attitudes, learning environment, current knowledge enhancement, and career choice in one single model.

Natural killer cells preferentially target and kill malignant and virally infected cells. Both these properties present compelling clinical utility in the field of haemopoietic progenitor cell transplantation (HPCT), potentially promoting a graft vs leukaemia effect in the absence of graft vs host disease and protecting against cytomegalovirus activation. Killer Ig-like receptors (KIR) play a central role in the cytotoxic action of natural killer cells, providing opportunity for improving transplantation outcomes by prioritising potential donors with optimal characteristics. Numerous algorithms for assessing KIR gene content as part of HPCT donor selection protocols exist, but no single model has been found to be universally applicable in all transplant centres. This review summarises several of the predominant strategies in KIR assessment algorithms, discussing their basic scientific principles, clinical utility and benefits to post-transplant outcomes. Finally, the review will consider how future donor selection protocols could develop towards unifying the concepts of KIR proteomics and genetics for optimising patient care.

This research intends to explore a variety of factors that may affect the satisfaction use of ELMSs. Therefore, various factors have been tested empirically by using another model approach integrated into a single model. Other variables that are still reliable for further research without changing the construct of the indicators are subjective norms, usability, perceived innovation and satisfaction. Some relationships between variables that have significant influence include usability on perceived innovation, perceived innovation on satisfaction. One relationships between variables that have a small influence subjective norms on perceived innovation.

Statistical network models such as the Gaussian Graphical Model and the Ising model have become popular tools to analyze multivariate psychological datasets. In many applications, the goal is to compare such network models across groups. In this paper, I introduce a method to estimate group differences in network models that is based on moderation analysis. This method is attractive because it allows one to make comparisons across more than two groups for all parameters within a single model and because it is implemented for all commonly used cross-sectional network models. Next to introducing the method, I evaluate the performance of the proposed method and existing approaches in a simulation study. Finally, I provide a fully reproducible tutorial on how to use the proposed method to compare a network model across three groups using the R-package mgm.

We apply the vision transformer, a deep machine learning model build around the attention mechanism, on mel-spectrogram representations of raw audio recordings. When adding mel-based data augmentation techniques and sample-weighting, we achieve comparable performance on both (PRS and CCS challenge) tasks of ComParE21, outperforming most single model baselines. We further introduce overlapping vertical patching and evaluate the influence of parameter configurations. Index Terms: audio classification, attention, mel-spectrogram, unbalanced data-sets, computational paralinguistics

The consequences of blast traumatic brain injury (blast-TBI) in humans are largely determined by the characteristics of the trauma insult and, within certain limits, the individual responses to the lesions inflicted [1]. In blast-TBI the mechanisms of brain vulnerability to the detonation of an explosive device are not entirely understood. They most likely result from a combination of the different physical aspects of the blast phenomenon, specifically extreme pressure oscillations (blast-overpressure wave – primary blast), projectile penetrating fragments (secondary blast) and acceleration-deceleration forces (tertiary blast), creating a spectrum of brain injury that ranges from mild to severe blast-TBI [2]. The pathophysiology of penetrating and inertially-driven blast-TBI has been extensively investigated for many years. However, the brain damage caused by blast-overpressure (primary blast) is much less understood and is unique to this type of TBI [3]. Indeed, there continues to be debate about how the pressure wave is transmitted and reflected through the brain and how it causes cellular damage [4]. No single model can mimic the clinical and mechanical complexity resulting from a real life blast-TBI [3]. The different models, non-biological (in silico or surrogate physical) and biological (ex vivo, in vitro or in vivo), tend to complement each other.

Many attempts have been made towards combining RGB and 3D poses for the recognition of Activities of Daily Living (ADL). ADL may look very similar and often necessitate to model fine-grained details to distinguish them. Because the recent 3D ConvNets are too rigid to capture the subtle visual patterns across an action, this research direction is dominated by methods combining RGB and 3D Poses. But the cost of computing 3D poses from RGB stream is high in the absence of appropriate sensors. This limits the usage of aforementioned approaches in real-world applications requiring low latency. Then, how to best take advantage of 3D Poses for recognizing ADL? To this end, we propose an extension of a pose driven attention mechanism: Video-Pose Network (VPN), exploring two distinct directions. One is to transfer the Pose knowledge into RGB through a feature-level distillation and the other towards mimicking pose driven attention through an attention-level distillation. Finally, these two approaches are integrated into a single model, we call VPN++. We show that VPN++ is not only effective but also provides a high speed up and high resilience to noisy Poses. VPN++, with or without 3D Poses, outperforms the representative baselines on 4 public datasets. Code is available at https://github.com/srijandas07/vpnplusplus., Comment: submitted to a journal

Background To examine the effectiveness of the use of machine learning for adapting an intraocular lens (IOL) power calculation for a patient group. Methods In this retrospective study, the clinical records of 1,611 eyes of 1,169 Japanese patients who received a single model of monofocal IOL (SN60WF, Alcon) at Miyata Eye Hospital were reviewed and analyzed. Using biometric metrics and postoperative refractions of 1211 eyes of 769 patients, constants of the SRK/T and Haigis formulas were optimized. The SRK/T formula was adapted using a support vector regressor. Prediction errors in the use of adapted formulas as well as the SRK/T, Haigis, Hill-RBF and Barrett Universal II formulas were evaluated with data from 395 eyes of 395 distinct patients. Mean prediction errors, median absolute errors, and percentages of eyes within ± 0.25 D, ± 0.50 D, and ± 1.00 D, and over + 0.50 D of errors were compared among formulas. Results The mean prediction errors in the use of the SRT/K and adapted formulas were smaller than the use of other formulas (P Conclusions Adapting IOL power calculations using machine learning technology with data from a particular patient group was effective and promising.

Diabetic retinopathy (DR) is a diabetes complication that damages the retina. This type of medical condition affects up to 80% of patients with diabetes for 10 or more years. The expertise and equipment required are often lacking in areas where diabetic retinopathy detection is most needed. Most of the work in the field of diabetic retinopathy has been based on disease detection or manual extraction of features. Thus, this research aims at automatic diagnosis of the disease in its different stages using deep learning neural network approach. This paper presents the design and implementation of Graphic Processing Unit (hereby GPU) accelerated deep convolutional neural networks to automatically diagnose and thereby classify high-resolution retinal images into five stages of the disease based on its severity. The accuracy of the single model convolutional neural networks presented in this paper is 71.65% from VGG-16.

This paper reports on creating a neural network model for prediction of the next action in a dialogue considering conversation history, i.e. entities, context variables and emotion indicators marking emotionally loaded user utterances. Several experiments were performed to see how the information about emotions affects the accuracy of the model. For the purposes of these experiments, a dataset containing 206 dialogs in Latvian in the transport inquiry domain was created containing both neutral and emotionally loaded utterances. To see if the proposed next dialogue action prediction model architecture is suitable for other languages, the original Latvian utterances were translated into English and a separate model was trained with English data. Some experiments were performed training the model with the data in one language and testing with the data in another language as well as training a single model using data in both languages. Experiments were performed with several fastText and Transformer pre-trained embedding models. Models for both languages Latvian and English achieved 0.91 accuracy on 10-fold cross-validation.

The COVID-19 pandemic has severely crippled the healthcare industry as a whole. Efficient screening techniques are crucial to suppress the escalation of the disease. Medical image analysis of chest X-rays has recently become increasingly important in radiology examination and screening of infected patients. Studies have shown that Deep CNN models can help in the diagnosis of this infection by automatically classifying chest X-ray images as infected or not. Ensemble modelling these Deep CNN architectures can further improve the performance by reducing the generalisation error when compared to a single model. This paper presents different Ensemble Learning approaches to synergize the features extracted by Deep CNN models to improve the classification. These automatic classification approaches can be used by radiologists to help identify infected chest X-rays and support resistance.

Despite the rise of child victimization in different societies, few researchers have examined its consequences in terms of psychological morbidity (such as depression and anxiety) and problematic online behavior (such as Internet addiction and cyberbullying) in a single study. Moreover, no study has investigated the role of psychological morbidity in mediating the impact of victimization on problematic online behavior (indexed by Internet addiction and cyberbullying) in a single model. Based on a survey of 2843 Chinese secondary students (49.3% male, Mage = 13.97) from six public secondary schools in Fujian, China, we found that experience of victimization was positively associated with depression and anxiety, as well as Internet addiction and cyberbullying. Depression mediated the links between victimization and both Internet addiction and cyberbullying, with the mediating effect on Internet addiction found to be stronger for girls. While anxiety did not mediate the association between victimization and cyberbullying, it mediated the relationship between victimization and Internet addiction in boys. These findings enrich our understanding of the negative outcomes of victimization, as well as directions for intervention.

Glaucoma, an eye condition that may slowly lead into blindness, may be diagnosed in clinical conditions by an ophthalmologist using various techniques and instruments. Known to be subjective, these methods can also be lengthy and costly. Various researchers have, instead, suggested the use of computer-aided methods for detecting glaucoma on fundus images. These methods mostly involved the use of deep learning models trained and tested on the same dataset using a single convolutional neural network (CNN) model. To observe and analyze their capabilities, this study creates a multiple fusion of CNN models and compares the outcomes to that of a single architecture. Unlike the studies before, fundus images trained and tested on different datasets are used to provide patient diversity. The results show a two-way ensemble can be preferred performance wise over a single model or a three model combination. They also show the optimum performance is obtained when the models are fused using the majority voting method.

Ensemble learning is a system which used to train multiple learning models and combine their results, treating them as a “committee” of decision makers. To explore effect of ensemble learning, this paper applied two basic ensemble systems of encoder to natural language processing. To compare the individual models and ensemble systems, this paper varied the number models which used to calculate ensemble accuracies. The result is that the decision of the model, with all models combined, usually have better overall accuracy, on average, than any single model. It shown that ensemble system used all models usually have better performance. This paper given explanation in the conclusion section of this result.

Procedural content generation via machine learning (PCGML) is the process of procedurally generating game content using models trained on existing game content. PCGML methods can struggle to capture the true variance present in underlying data with a single model. In this paper, we investigated the use of ensembles of Markov chains for procedurally generating \emph{Mega Man} levels. We conduct an initial investigation of our approach and evaluate it on measures of playability and stylistic similarity in comparison to a non-ensemble, existing Markov chain approach., 9 pages, 7 figures, Workshop on Procedural Content Generation

When training a machine learning model, there is likely to be a tradeoff between accuracy and the diversity of the dataset. Previous research has shown that if we train a model to detect one specific malware family, we generally obtain stronger results as compared to a case where we train a single model on multiple diverse families. However, during the detection phase, it would be more efficient to have a single model that can reliably detect multiple families, rather than having to score each sample against multiple models. In this research, we conduct experiments based on byte $n$-gram features to quantify the relationship between the generality of the training dataset and the accuracy of the corresponding machine learning models, all within the context of the malware detection problem. We find that neighborhood-based algorithms generalize surprisingly well, far outperforming the other machine learning techniques considered.

Common acoustic properties that different classes share degrades the performance of acoustic scene classification systems. This results in a phenomenon where a few confusing pairs of acoustic scenes dominate a significant proportion of all misclassified audio segments. In this article, we propose adopting a knowledge distillation framework that trains deep neural networks using soft labels. Soft labels, extracted from another pre-trained deep neural network, are used to reflect the similarity between different classes that share similar acoustic properties. We also propose utilizing specialist models to provide additional soft labels. Each specialist model in this study refers to a deep neural network that concentrates on discriminating a single pair of acoustic scenes that are frequently misclassified. Self multi-head attention is explored for training specialist deep neural networks to further concentrate on target pairs of classes. The goal of this article is to train a single deep neural network that demonstrates performance equivalent to, or higher than, an ensemble of multiple models, by distilling the knowledge from several models. Diverse experiments conducted using the detection and classification of acoustic scenes and events 2019 task 1-a dataset demonstrate that the knowledge distillation framework is effective in acoustic scene classification. Specialist models successfully decrease the number of misclassified audio segments in the target classes. The final single model with the proposed method that is trained by the proposed knowledge distillation from several models, including specialists trained using an attention mechanism, shows a classification accuracy of 77.63 %, higher than an ensemble of the baseline and multiple specialists.

Vb cyclones are major drivers of extreme precipitation and floods in the study area of hydrological Bavaria (Germany). When assessing climate change impacts on Vb cyclones, internal variability of the climate system is an important underlying uncertainty. Here, we employ a 50-member single-model initial-condition large ensemble of a regional climate model to study climate variability and forced change on Vb cyclones. An artificial neural network detects cutoff lows over central Europe, which are associated with extreme precipitation Vb cyclones. Thus, machine learning filters the large ensemble prior to cyclone tracking. Our results show a striking change in Vb seasonality with a strong decrease of Vb cyclones in summer (−52%) and a large increase in spring (+73%) under the Representative Concentration Pathway 8.5. This change exceeds the noise of internal variability and leads to a peak shift from summer to spring. Additionally, we show significant increases in the daily precipitation intensity during Vb cyclones in all seasons.

Ensemble learning is a prolific field in Machine Learning since it is based on the assumption that combining the output of multiple models is better than using a single model, and it usually provides good results. Normally, it has been commonly employed for classification, but it can be used to improve other disciplines such as feature selection. Feature selection consists of selecting the relevant features for a problem and discard those irrelevant or redundant, with the main goal of improving classification accuracy. In this work, we provide the reader with the basic concepts necessary to build an ensemble for feature selection, as well as reviewing the up-to-date advances and commenting on the future trends that are still to be faced.

Attribute guided face image synthesis aims to manipulate attributes on a face image. Most existing methods for image-to-image translation can either perform a fixed translation between any two image domains using a single attribute or require training data with the attributes of interest for each subject. Therefore, these methods could only train one specific model for each pair of image domains, which limits their ability in dealing with more than two domains. Another disadvantage of these methods is that they often suffer from the common problem of mode collapse that degrades the quality of the generated images. To overcome these shortcomings, we propose attribute guided face image generation method using a single model, which is capable to synthesize multiple photo-realistic face images conditioned on the attributes of interest. In addition, we adopt the proposed model to increase the realism of the simulated face images while preserving the face characteristics. Compared to existing models, synthetic face images generated by our method present a good photorealistic quality on several face datasets. Finally, we demonstrate that generated facial images can be used for synthetic data augmentation, and improve the performance of the classifier used for facial expression recognition.

Methods to reliably estimate the accuracy of 3D models of proteins are both a fundamental part of most protein folding pipelines and important for reliable identification of the best models when multiple pipelines are used. Here, we describe the progress made from CASP12 to CASP13 in the field of estimation of model accuracy (EMA) as seen from the progress of the most successful methods in CASP13. We show small but clear progress, that is, several methods perform better than the best methods from CASP12 when tested on CASP13 EMA targets. Some progress is driven by applying deep learning and residue-residue contacts to model accuracy prediction. We show that the best EMA methods select better models than the best servers in CASP13, but that there exists a great potential to improve this further. Also, according to the evaluation criteria based on local similarities, such as lDDT and CAD, it is now clear that single model accuracy methods perform relatively better than consensus-based methods.

This study presents a model to analyse the effect of an increasing level of residential and commercial photovoltaic and storage devices, intended for self-consumption, on a power system. A whole year is focused and modelled through a set of selected representative days. A 24-h two-stage stochastic unit commitment is solved for each representative day and expected results are weighted to achieve annual expected results. This optimisation problem includes appropriate representation of the day-ahead and on-line operations and an adequate representation of the uncertainty associated with renewable sources and demand. The main contributions are: (i) the definition of a single model in which customers are price-makers and large-scale and customers’ installations are appropriately modelled, therefore bridging the current macroscopic and microscopic modelling approaches, and (ii) the division of customers into four non-overlapping groups, which allows a sensitivity analysis of the effect of each group. Five main results are focused: net demand, CO2 emissions, load factors, system operation cost, and customer savings. The model is applied to a case study based on the Spanish power system to show the model capability to quantify the effect of an increasing level of these promising technologies on a power system.

The purpose of this study is to explore and explain the relationship between networking, external financing by banks and equity investors, and export effort in medium sized enterprises. Our research model was empirically tested on a sample of 143 MSEs in the manufacturing sector and our PLS results do not confirm the existence of a direct association between networking and export effort by MSEs. They show, however, that networking indirectly enhances MSEs’ export effort through facilitating their access to equity financing and reducing managerial risks. Whilst networking are not found to be associated with a higher access to bank financing by MSEs, this type of external financing seems to have a greater impact on export effort than equity financing. This research it is one of the first studies to explore the role of networking with stakeholders in facilitating MSEs’ access to external financing and to international markets and to merge these variables in a single model which consider simultaneously the MSE access to bank financing and to equity financing.

This study aims to present an integrated Islamic financial accountability model, which is different from previous accountability models and frameworks. In this case, all three accountability aspects which are: (i) accountability to whom (accountability groups); (ii) accountability for what (accountability dimensions); and (iii) accountability on how (accountability tools and processes) were integrated into a single model. By using these three accountability aspects, this model is specifically focused on financial accountability area with additional Islamic elements that suits Islamic Social Enterprise (ISE). In order to develop this model, this study applies the comparison method on previous accountability and Islamic accountability models with additional literature inputs from financial accountability aspects and Islamic accountability perspectives. Based on the findings, this study introduces an integrated Islamic financial accountability model specifically for ISE, one of the emerging institution types under the non-profit sector. It is hoped that the integrated Islamic financial accountability model developed in this study will be implemented by the management of ISE in order to discharge their accountability especially to donors, beneficiaries and other stakeholders.

BACKGROUND AND PURPOSE: Accurate automated infarct segmentation is needed for acute ischemic stroke studies relying on infarct volumes as an imaging phenotype or biomarker that require large numbers of subjects. This study investigates whether an ensemble of convolutional neural networks (CNN) trained on multiparametric DWI maps outperforms single networks trained on solo DWI parametric maps. MATERIALS AND METHODS: CNNs were trained on combinations of DWI, ADC, and low b-value-weighted images from 116 subjects. The performances of the networks (measured by Dice score, sensitivity and precision) were compared to one another and to ensembles of 5 networks. To assess the generalizability of the approach, the best performing model was applied to an independent evaluation cohort of 151 subjects. Agreement between manual and automated segmentations for identifying patients with large lesions volumes was calculated across multiple thresholds (21 cm(3), 31 cm(3), 51 cm(3), and 70 cm(3)). RESULTS: An ensemble of CNNs trained on DWI, ADC and low b-value-weighted images produced the most accurate acute infarct segmentation over individual networks (p

Owing to its recent advance, machine learning has spawned a large collection of solar forecasting works. In particular, machine learning is currently one of the most popular approaches for hourly solar forecasting. Nevertheless, there is evidently a myth on forecast accuracy—virtually all research papers claim superiority over others. Apparently, the “best” model can only be selected with hindsight, i.e., after empirical evaluation. For any new forecasting project, it is irrational for solar forecasters to bet on a single model from the start. In this article, the hourly forecasting performance of 68 machine learning algorithms is evaluated for 3 sky conditions, 7 locations, and 5 climate zones in the continental United States. To ensure a fair comparison, no hybrid model is considered, and only off-the-shelf implementations of these algorithms are used. Moreover, all models are trained using the automatic tuning algorithm available in the R caret package. It is found that tree-based methods consistently perform well in terms of two-year overall results, however, they rarely stand out during daily evaluation. Although no universal model can be found, some preferred ones for each sky and climate condition are advised.

Introduction Expectations for assessment in higher education have increased in recent decades, prompting institutions to invest additional resources in this area. This study aimed to determine the resources, structure, and perception of assessment resources in United States schools and colleges of pharmacy (S/COPs). Methods Assessment personnel in S/COPs were surveyed electronically. Information collected included S/COP demographics, composition of assessment positions, experience and training of assessment personnel, and structure and responsibilities of committees engaged in assessment. Respondents’ perception of their S/COPs having sufficient assessment personnel, recent changes in assessment, and the factors that prompted assessment changes were also surveyed. Results Respondents included individuals from 113 S/COPs (84% response rate). Most S/COPs had 1–2 assessment positions and 1–2 assessment-related committees. The most common assessment position titles were assistant/associate dean, director, coordinator/specialist, and administrative assistant. Dean-level administrators typically had worked in assessment the longest, whereas directors were more likely to have formal assessment training. Most respondents (75%) agreed they had sufficient assessment personnel to meet the 2007 Accreditation Council for Pharmacy Education Standards. Nearly two-thirds of respondents agreed they had sufficient personnel to meet the 2016 Standards and support their current assessment plan/process. Most S/COPs had a formal assessment committee (93%) and an average of two committees overseeing assessment. Conclusion Most S/COPs reported having sufficient resources to support assessment activities. Although there were some consistent themes, there does not appear to be a single model for structuring assessment resources or committees. Effectiveness of various assessment structures represents an area for future research.

To achieve the goal of preventing serious injuries and fatalities, it is important for a mine site to analyze site specific mine safety data. The advances in natural language processing (NLP) create an opportunity to develop machine learning (ML) tools to automate analysis of mine health and safety management systems (HSMS) data without requiring experts at every mine site. As a demonstration, nine random forest (RF) models were developed to classify narratives from the Mine Safety and Health Administration (MSHA) database into nine accident types. MSHA accident categories are quite descriptive and are, thus, a proxy for high level understanding of the incidents. A single model developed to classify narratives into a single category was more effective than a single model that classified narratives into different categories. The developed models were then applied to narratives taken from a mine HSMS (non-MSHA), to classify them into MSHA accident categories. About two thirds of the non-MSHA narratives were automatically classified by the RF models. The automatically classified narratives were then evaluated manually. The evaluation showed an accuracy of 96% for automated classifications. The near perfect classification of non-MSHA narratives by MSHA based machine learning models demonstrates that NLP can be a powerful tool to analyze HSMS data.

Multilingual neural machine translation with a single model has drawn much attention due to its capability to deal with multiple languages. However, the current multilingual translation paradigm often makes the model tend to preserve the general knowledge, but ignore the language-specific knowledge. Some previous works try to solve this problem by adding various kinds of language-specific modules to the model, but they suffer from the parameter explosion problem and require specialized manual design. To solve these problems, we propose to divide the model neurons into general and language-specific parts based on their importance across languages. The general part is responsible for preserving the general knowledge and participating in the translation of all the languages, while the language-specific part is responsible for preserving the language-specific knowledge and participating in the translation of some specific languages. Experimental results on several language pairs, covering IWSLT and Europarl corpus datasets, demonstrate the effectiveness and universality of the proposed method., ACL 2021

For many supervised learning tasks, ensemble classifiers - which make predictions by combining multiple simple models - outperform single model classifiers. While genetic programming can be used to evolve populations of simple classifiers, it tends to produce populations of highly similar models. In this work we propose Neuro MAP-Elites (NME) as a method for evolving populations of high performing models which produce diverse predictions, making them suitable for constructing ensembles.

Sports injuries will have an impact on the consistency and systemicity of the training process, as well as athlete training and performance improvement. Many talented athletes have had their careers cut short due to sports injuries. Preventing sports injuries is the best way for basketball players to reduce sports injuries. Many coaches and athletes on sports teams, on the other hand, are unaware of the importance of sports injury prevention. They only realize that the body’s sports functions are abnormal when it suffers from sports injuries. As a result, this paper proposes a gray theory neural network-based athlete injury prediction model. First, from the standpoint of a single model, the improved unequal interval model is used to predict sports injury by optimizing the unequal interval model in gray theory. The findings show that it is a good predictor of sports injuries, but it is a poor predictor of the average number of injuries. Following that, in order to overcome the shortcomings of a single model, a gray neural network combination model was used. A combination model of the unequal time interval model and BP neural network was determined and established. The prediction effect is significantly improved by combining the gray neural network mapping model and the coupling model to predict the two characteristics of sports injuries. Finally, simulation experiments show that the proposed method is effective.

In drug discovery, classification is a well established in silico method based on machine learning algorithms. However, since the activity value on a target protein is described as a continuous value, a regressional approach is worth being considered as well. The results of the regression can then be turned into classification results with a given threshold value. To further improve the results, a new method called optimally weighted ensembles that uses a combination of more than one model to build a better performing ensemble of models, is applied here. This method is used in the context of drug discovery for the first time. Naturally, it is crucial to choose suitable models for the specific problem, if any prior knowledge is available. Statistical significance of the approach is verified using a second dataset with a different target. In this work, we show to what extent the obtained classification results compare to previous, highly optimized, single model results as presented in previous work. Furthermore, the results are compared to ensembles where none of the contributing models were optimized beforehand. All case studies are performed using the publicly available database ChEMBL.

Background: In the 2018 AHA/ACC Cholesterol guideline, risk stratification is an essential element. The use of a Pooled Cohort Equation (PCE) is recommended for individuals without atherosclerotic cardiovascular disease (ASCVD), and the new dichotomous classification of very high-risk vs. high-risk has been introduced for patients with ASCVD. These distinct risk stratification systems mainly rely on traditional risk factors, raising the possibility that a single model can predict major adverse cardiovascular events (MACEs) in persons with and without ASCVD. Methods: We studied 11,335 ARIC participants with (n=885) and without (n=10,450) a history of ASCVD (myocardial infarction, ischemic stroke, and symptomatic peripheral artery disease) at baseline (1996-98). We modeled factors in the PCE and the new classification for ASCVD patients (Figure legend) in a single CVD prediction model. We examined their associations with MACEs (myocardial infarction, stroke, and heart failure) using Cox models and evaluated the discrimination and calibration for a single model including those factors. Results: During a median follow-up of 18.4 years, there were 3,658 MACEs (3,105 in participants without ASCVD). In general, the factors in the PCE and the risk classification system for ASCVD patients were associated similarly with MACEs regardless of baseline ASCVD status, although age and systolic blood pressure showed significant interactions. A single model with these predictors and the relevant interaction terms showed good calibration and discrimination for those with and without ASCVD (c-statistic=0.729 and 0.704, respectively) (Figure). Conclusion: A single CVD prediction model performed well in persons with and without ASCVD. This approach will provide a specific predicted risk to ASCVD patients (instead of dichotomy of very high vs. high risk) and eliminate a practice gap between primary vs. secondary prevention due to different risk prediction tools.