Search

AbstractProbability surveys are a major source of population representative data for policy research and program evaluation. However, the data come with the added complications of being observational and selected with unequal probabilities. Propensity score adjustments have become increasingly popular for inferring causal relationships in non-randomized studies, but when using survey data, estimates of the population level causal effect may be biased if the sampling design is not adequately adjusted for. The current practice of using propensity score estimators with complex surveys is somewhat ad-hoc. We propose a potential-outcome super-population framework to streamline the causal analysis. We also develop propensity-score-and-survey weighted estimators and corresponding variance estimators, as well as their asymptotic properties. Our framework clarifies the confusion regarding the use of survey weighted propensity score in practice. The choice actually depends on the available sampling weights. Various estimators are compared in a simulation study, which shows that the proposed estimators perform better than the competing methods in terms of bias and confidence interval coverage when treatment effects are heterogeneous. To address an important public health issue, we evaluate the impact of e-cigarette use on future tobacco use intention in teens, using a large nationally representative survey in the United States. [ABSTRACT FROM AUTHOR]

With the phasor measurement units (PMUs) being widely utilized in power systems, a large amount of data can be stored. If transient stability assessment (TSA) method based on the deep learning model is trained by this dataset, it requires high computation cost. Furthermore, the fact that unstable cases rarely occur would lead to an imbalanced dataset. Thus, power system transient stability status prediction has the bias problem caused by the imbalance of sample size and class importance. Faced with such a problem, a TSA model based on the sample selection method is proposed in this paper. Sample selection aims to optimize the training set to speed up the training process while improving the preference of the TSA model. The typical samples which can accurately express the spatial distribution of the raw dataset are selected by the proposed method. Primarily, based on the location of training samples in the feature space, the border samples are selected by trained support vector machine (SVM), and the edge samples are selected by the assistance of the approximated tangent hyperplane of a class surface. Then, the selected samples are input to stacked sparse autoencoder (SSAE) as the final classifier. Simulation results in the IEEE 39‐bus system and the realistic regional power system of Eastern China show the high performance of the proposed method. [ABSTRACT FROM AUTHOR]

ABSTRACT This paper presents a novel approach to pollution assessment by investigating support vector machines (SVM) with an uncertainty option to overcome the limitations of traditional kriging. While kriging is a major tool for geostatistical modelling, allowing to estimate the distribution of contaminants in a region from a small set of samples, it does not allow to extract also the uncertainty map. An uncertainty map is of great interest, as it allows to identify regions of high uncertainty where one should sample in order to reduce high level of uncertainties. In this paper, we propose two variants of the SVM with an uncertainty option, each using a different hinge loss to improve the accuracy and efficiency. These losses allow to estimate different levels of contaminations, as well as uncertainty, such as the three levels: positive, uncertain and negative, namely for pollution estimation: high‐pollution, uncertain and low‐pollution. In addition to the exploration of SVM variants, we propose an innovative active sample selection strategy based on the uncertainty criterion. This strategy is designed to systematically reduce uncertainties in pollution assessment, thus providing adaptability to dynamic environmental changes. An incremental SVM with an uncertainty option is introduced to further optimise the sample selection process. Furthermore, the decision‐making process is refined through the introduction of a novel three‐hinge loss. The corresponding optimization problem and its resolution allow for a more nuanced contamination assessment with multiple levels of estimation, providing a valuable tool for characterising contamination levels with increased granularity. Extensive experiments on synthetic and real data validate the proposed methodology. Synthetic data simulations assess the quality of the approach, while real data from a two‐dimensional porosity measurement demonstrate practical applicability. This research contributes to the advancement of pollution assessment methodologies, providing an adaptable solution for environmental monitoring. [ABSTRACT FROM AUTHOR]

Combustion optimization is an effective way to improve the efficiency of thermal power generation and reduce carbon and NOx emissions. Real-time and precise NOx emission prediction is the basis for combustion optimization control of thermal power plants. To construct an accurate NOx concentration prediction model, a novel just-in-time learning (JITL) method based on random forest (RF) is proposed in the present work. With this method, first, an improved permutation importance algorithm is proposed to extract important variables. In addition, a similarity index that incorporates temporal and spatial measures is defined to select a local training set representative of the process data. Moreover, considering the influence of model parameters on prediction performance under different working conditions, a process monitoring method based on a moving window (MW) is used to monitor the change in working conditions and guide online updating. The experimental results show that the proposed method has excellent prediction accuracy, with a coefficient of determination of 0.9319, a root-mean-square error of 3.6960 mg/m3, and an average absolute error of 2.7718 mg/m3 on the test set, making it superior to other traditional methods. [ABSTRACT FROM AUTHOR]

This paper seeks to identify the causal impact of educational human capital on social distancing behavior at workplace in Turkey using district-level data for the period of April 2020 - February 2021. We adopt a unified causal framework, predicated on domain knowledge, theory-justified constraints anda data-driven causal structure discovery using causal graphs. We answer our causal query by employing machine learning prediction algorithms; instrumental variables in the presence of latent confounding and Heckman's model in the presence of selection bias. Results show that educated regions are able to distance-work and educational human capital is a key factor in reducing workplace mobility, possibly through its impact on employment. This pattern leads to higher workplace mobility for less educated regions and translates into higher Covid-19 infection rates. The future of the pandemic lies in less educated segments of developing countries and calls for public health action to decrease its unequal and pervasive impact. [ABSTRACT FROM AUTHOR]

This paper addresses the performance degradation of deep neural networks caused by learning with noisy labels. Recent research on this topic has exploited the memorization effect: networks fit data with clean labels during the early stages of learning and eventually memorize data with noisy labels. This property allows for the separation of clean and noisy samples from a loss distribution. In recent years, semi-supervised learning, which divides training data into a set of labeled clean samples and a set of unlabeled noisy samples, has achieved impressive results. However, this strategy has two significant problems: (1) the accuracy of dividing the data into clean and noisy samples depends strongly on the network's performance, and (2) if the divided data are biased towards the unlabeled samples, there are few labeled samples, causing the network to overfit to the labels and leading to a poor generalization performance. To solve these problems, we propose the curriculum regularization and adaptive semi-supervised learning (CRAS) method. Its key ideas are (1) to train the network with robust regularization techniques as a warm-up before dividing the data, and (2) to control the strength of the regularization using loss weights that adaptively respond to data bias, which varies with each split at each training epoch. We evaluated the performance of CRAS on benchmark image classification datasets, CIFAR-10 and CIFAR-100, and real-world datasets, mini-WebVision and Clothing1M. The findings demonstrate that CRAS excels in handling noisy labels, resulting in a superior generalization and robustness to a range of noise rates, compared with the existing method. [ABSTRACT FROM AUTHOR]

Crop mapping using remote sensing data is essential for predicting crop yield and future requirements. However, it can be accomplished by having full ground truth data corresponding to the samples captured by remote sensors. In a practical scenario, generally, we have limited ground-truth data where available crop samples are less than the number of spectral bands of a hyperspectral image. This paper proposes a sample selection method to select samples from a hyperspectral image based on similarity measure and fuzziness in the samples. We use a single-ground truth pixel of each crop for crop-sample selection. Firstly, we select the number of samples for each crop class using Spectral Information Diversion and tangent of Spectral Angle Mapper ( SS tan ) measure from a selected region of the hyperspectral image. Then, we compute fuzziness for each selected crop sample to generate a set of final samples. We have evaluated the proposed method on an Airborne Visible near InfraRed Imaging Spectrometer-Next Generation (AVIRIS-NG) dataset and two benchmark datasets, i.e., Indian Pines and Salinas. Finally, classification results using three supervised classifiers, namely, support vector machine, k-nearest neighbourhood and random forest, have been shown in terms of overall accuracy and kappa coefficient. It can be demonstrated that the proposed method outperforms some prevailing methods by at least 4.5 % for 5 samples and 11% for 10 samples for the Indian Pines dataset. [ABSTRACT FROM AUTHOR]

Örnekler; bir kavrama ait tanımları açıklamak, kavrama ait olan ya da olmayan durumları detaylandırmak veya somutlaştırmak amacıyla, kavramın özel durumlarını oluşturan yapılardır. Bu araştırmada ortaokul matematik öğretmenlerinin; oran-orantı konusunda örnekleri nasıl belirlediklerini, derslerin farklı aşamalarında ne tür örnekleri kullandıklarını tespit etmek ve örnekleri kullanırken yararlandıkları kaynakların neler olduğunu ortaya çıkarmak amaçlanmıştır. Araştırmada tarama yöntemi türlerinden biri olan kesitsel tarama yöntemi kullanılmıştır. Araştırmanın katılımcıları, uygun örnekleme yöntemiyle belirlenen 97 ortaokul matematik öğretmeninden oluşmaktadır. Araştırmanın verileri, öğretmenlere uygulanan açık uçlu sorulardan oluşan anket ve yarı-yapılandırılmış mülakatlar yardımıyla toplanmıştır. Elde edilen veriler betimsel olarak analiz edilmiştir. Öğretmenlerin verdikleri cevaplara göre kategoriler oluşturulmuştur. Araştırma sonucunda öğretmenlerin kullandıkları örnek türlerinin dersin aşamalarına göre farklılık gösterdiği tespit edilmiştir. Bununla birlikte öğretmenlerin günlük hayatla ilişkili örnekleri dersin her aşamasında kullandıkları belirlenmiştir. Öğretmenlerin örnekleri kullanırken yararlandıkları kaynakların; “yardımcı kaynaklar”, “z-kitap”, “ders kitabı”, “internet” ve “EBA” ile “ÖDSGM” olduğu görülmüştür. Bazı öğretmenler ise herhangi bir kaynaktan yararlanmadan kendilerinin örnekler hazırladıklarını ifade etmişlerdir. Öğretmenlerin örnek seçiminde sadece bir kaynağa bağlı kalmadıkları, farklı kaynaklardan yararlandıkları tespit edilmiştir. Bunun yanı sıra öğretmenlerin mesleki kıdemi arttıkça EBA ve internet gibi dijital kaynaklar yerine daha çok ders kitaplarındaki örnekleri tercih ettikleri belirlenmiştir.

Consuming too much fat, sugar, and salt is associated with adverse health outcomes. Food reformulation is one possible strategy to enhance the food environment by improving the nutritional quality of commercial products. However, food reformulation faces many hindrances. One way to alleviate some of these hindrances is to embrace a multicriteria approach that is based on a market inventory. In this objective, additional sensory screening and water content analyses allow going beyond nutrition and composition information on the packaging. However, due to feasibility reasons for later in-depth analyses, it is necessary to work with several reduced and manageable products. To the best of the authors' knowledge, in the literature, there is no sample selection approach taking into account multiple criteria as a base for future food reformulation. The overall aim of this paper is to propose a method to select the best representative products from the market base, for future reformulation by going beyond nutrition and composition information on the packaging. This approach considered therefore nutrition, composition, economic, water content, and sensory information with the example of the cookies market. The first step is the creation of an extensive cookie database including sensory and water content information. In total 178 cookies among the French market were identified, then focus was placed on 62 chocolate chip cookies only. Sensory screening and water content analyses of all 62 products were conducted. The second step is to make an informed subset selection, thanks to a cluster analysis based on 11 nutrition, composition, and water content variables. A representative subset of 18 cookies could be derived from the obtained clusters. The representativity was evaluated with statistical uni- and multivariate analyses. Results showed a broad variety of chocolate chips cookies with a large nutritional, compositional, water content, and sensory differences. These results highlighted the first paths for future reformulation in this product category and showed the importance to include physical product information beyond the information on the packaging. This complete database on the selected cookies constituted a solid base for identifying future reformulation levers, in order to improve the nutritional quality and health.

Estimating true treatment effects in the presence of selection bias is a common problem in program evaluation, but it is uncommon to quantify the share of a total apparent effect that is “true”. We identify multiple methods for doing so, in the context of isolating the true causal influence of the built environment on travel behavior in the presence of residential self-selection (RSS). Among the different approaches for dealing with self-selection, we use three: statistical control (SC) modeling, propensity score-based techniques (PS), and sample selection (SS) modeling. Our objective is to identify multiple ways to estimate the proportion of the total apparent effect of the built environment on travel behavior that is due to the built environment itself, which we call the “built environment proportion”, or BEP. We present and evaluate three categories of methods for estimating this measure (each native to one of the approaches): variance explained, modular effects, and treatment effects. A BEP formula associated with a given method can be applied to its native approach or cross-applied to other approaches. We identify and enumerate 47 potential BEPs. The methods presented here can be applied in other contexts involving treatment evaluation in the presence of selection bias.

Combustion optimization is an effective way to improve the efficiency of thermal power generation and reduce carbon and NOx emissions. Real-time and precise NOx emission prediction is the basis for combustion optimization control of thermal power plants. To construct an accurate NOx concentration prediction model, a novel just-in-time learning (JITL) method based on random forest (RF) is proposed in the present work. With this method, first, an improved permutation importance algorithm is proposed to extract important variables. In addition, a similarity index that incorporates temporal and spatial measures is defined to select a local training set representative of the process data. Moreover, considering the influence of model parameters on prediction performance under different working conditions, a process monitoring method based on a moving window (MW) is used to monitor the change in working conditions and guide online updating. The experimental results show that the proposed method has excellent prediction accuracy, with a coefficient of determination of 0.9319, a root-mean-square error of 3.6960 mg/m3, and an average absolute error of 2.7718 mg/m3 on the test set, making it superior to other traditional methods.

Cropland is the foundation of agriculture and the cornerstone of ensuring national food security. Ongoing advancement of China’s urbanization and social economy results in a tendency toward ‘non-agriculturalization’ of cropland in pursuit of economic benefits, which greatly threatens food security. Therefore, long-term monitoring of cropland is urgently needed. Based on the Google Earth Engine (GEE) platform, this paper uses the Landsat series and Sentinel-2 multisource images from 1990–2020 to dynamically monitor cropland in Yunnan Province. First, a multiresolution collaborative sample selection method was proposed and sample labels with resolutions better than 30 m from 1990 to 2015 were obtained. An ‘impervious surface-spectrum-texture-geometry’ (ISTG) feature system was constructed and the Jeffries-Matusita (JM) distance method was used for feature optimization. Finally, the proposed collaborative classification system method of land classification and binary classification was used to extract cropland in Yunnan Province. The results show that: (1) The accuracy of cropland extraction was highly reliable, with an overall accuracy (OA) wall-to-wall ranging from 93.15–98.14% and a kappa coefficient (Kappa) ranging from 86.29–96.26%. (2) The change trend of the cropland area in Yunnan Province from 1990 to 2020 exhibited certain fluctuations, but overall decreased annually. (3) The province’s average annual cropland area is 567.71 × 103 hectares. The proposed method has great potential in large-scale cropland extraction.

AbstractFinancial inclusion is one of the development priorities of the Ethiopian Government. However, factors, which might affect it are not explored adequately with appropriate methods. To examine the relationship between financial inclusion and its demand-side determinants, we applied a simple probit model and two probit models with sample selection to the 2018/19 Ethiopia Socioeconomic Survey (ESS4). Our simple probit model estimation results show that male, older, non-Muslim, married, more educated, richest, mobile users or people who are aware of how to open a bank account are more likely to have a formal account than their counterparts. However, being from a rural area or from a larger household size reduces the likelihood of having an account. Results from the probit selection model show that rural residents, employed, richer people, or people who are out of labor force are more likely to use their account to save money than their counterparts. Furthermore, the third model shows that being non-Muslim, married, widowed, or poorer reduces the likelihood of account use than their counterparts. Thus, policies should target the most vulnerable groups of the society. Especially, rural households, women, the Muslim population, the poor, less educated people, and the young should be the focus of financial inclusion endeavors of the Government. Parallel to these policy measures, awareness of opening bank account and financial education programs should be promoted.

In the paper, we first explore a novel problem of training the robust Graph Neural Networks (GNNs) against noisy graphs and noisy labels. To the problem, we propose a general Self-supervised Robust Graph Neural Network framework that consists of three modules: graph structure learning, sample selection, and self-supervised learning. Specifically, we first employ a graph structure learning approach to obtain an optimal graph structure. Next, using this structure, we use a clustering algorithm to generate pseudo-labels that represent the clusters. We then design a sample selection strategy based on these pseudo-labels to select nodes with clean labels. Additionally, we introduce a self-supervised learning technique where low-level layer parameters are shared with GNNs to predict pseudo-labels. We jointly train the graph structure learning module, the GNNs model, and the self-supervised model. Finally, we conduct extensive experiments on four real-world datasets, demonstrating the superiority of our methods compared with state-of-the-art methods for semi-supervised node classification under noisy graphs and noisy labels. [ABSTRACT FROM AUTHOR]

Copyright of Anadolu University Journal of Education Faculty (AUJEF) is the property of Anadolu Universitesi Egitim Fakultesi Dergisi and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

For the past forty years, Ethiopia has been promoting sustainable land management activities to enhance agricultural productivity. This study was intended to identify the factors determining farmers’ adoption and continued use of soil bund measures in the Handosha watershed, Omo-Gibe river basin. A multistage sampling technique was used to select 340 households using the Heckman sample selection model. A total of 235 (69.12%) households adopted soil bunds, but only 89 (37.87%) of them were sustainably practicing soil bunds on their farm plots. Most adopters widely practiced soil bunds (49.42%), followed by stone bund (15.9%), and Fanyajuu (10%). The empirical results of the Heckman sample selection model showed that the farming experience, land tenure security, and perception of profitability of conservation measures were significantly positively affected the adoption of soil bund. Whereas, farm plot size and participation in off farm activities significantly negatively influenced the adoption of soil bund. Sustainable use of soil bund measures were significantly positively influenced by land tenure security, family size, and frequency of extension contact, whereas the distance between farm plots and home, and farm plot size were negatively affected. As a result, a design of agro-ecological-based soil and water conservation (SWC) measures was essential in reducing farmland vulnerability to soil erosion and food insecurity. It has been concluded that conservation practices should not only focus on the implementation and biophysical factors but also consider the socioeconomic interests of the farmers to improve the sustainable use of conservation technologies.

Easy-to-use libraries such as scikit-learn have accelerated the adoption and application of machine learning (ML) workflows and data-driven methods. While many of the algorithms implemented in these libraries originated in specific scientific fields, they have gained in popularity in part because of their generalisability across multiple domains. Over the past two decades, researchers in the chemical and materials science community have put forward general-purpose machine learning methods. The deployment of these methods into workflows of other domains, however, is often burdensome due to the entanglement with domain-specific functionalities. We present the python library scikit-matter that targets domain-agnostic implementations of methods developed in the computational chemical and materials science community, following the scikit-learn API and coding guidelines to promote usability and interoperability with existing workflows.

Cyberbullying detection has attracted many researchers to detect negative comments deployed on communication platforms as cyberbullying can take many forms: verbal, implicit, explicit, or even nonverbal. The successful growth of social media in recent years has opened new perspectives on the detection of cyberbullying, although related research still encounters several challenges, such as data imbalance and expression implicitness. In this paper, we propose an automated cyberbullying detection framework designed to produce satisfactory results, especially when imbalanced short text and different dialects exist in the Arabic text data. In the proposed framework a new method to solve the imbalance problem is suggested, where the modified simulated annealing optimization algorithm is used to find the optimal set of samples from the majority class to balance the training set. This method has been evaluated using traditional machine learning algorithms including support vector machine, and deep learning algorithms including Long Short-Term Memory (LSTM) and Bidirectional LSTM (Bi-LSTM). To generate a framework that can detect Arabic written cyberbullying on communication platforms, the accuracy, recall, specificity, sensitivity and mean squared error are used as the main performance indicators. The results indicate that the proposed framework can improve the performance of the tested algorithms, and Bi-LSTM outperforms other methods for cyberbullying classification.

The building damage caused by natural disasters seriously threatens human security. Applying deep learning algorithms to identify collapsed buildings from remote sensing images is crucial for rapid post-disaster emergency response. However, the diversity of buildings, limited training dataset size, and lack of ground-truth samples after sudden disasters can significantly reduce the generalization of a pre-trained model for building damage identification when applied directly to non-preset locations. To address this challenge, a self-incremental learning framework (i.e., SELF) is proposed in this paper, which can quickly improve the generalization ability of the pre-trained model in disaster areas by self-training an incremental model using automatically selected samples from post-disaster images. The effectiveness of the proposed method is verified on the 2010 Yushu earthquake, 2023 Turkey earthquake, and other disaster types. The experimental results demonstrate that our approach outperforms state-of-the-art methods in terms of collapsed building identification, with an average increase of more than 6.4% in the Kappa coefficient. Furthermore, the entire process of the self-incremental learning method, including sample selection, incremental learning, and collapsed building identification, can be completed within 6 h after obtaining the post-disaster images. Therefore, the proposed method is effective for emergency response to natural disasters, which can quickly improve the application effect of the deep learning model to provide more accurate building damage results. [ABSTRACT FROM AUTHOR]

As an effective semi-supervised learning algorithm, the co-training method trains two classifiers on two views independently. The unlabeled sample selection strategy in the self-labeled process is crucial for co-training. However, most of the existing strategies strongly depend on parameter settings and require re-calculating the confidence of unlabeled samples in each iteration. Inspired by the concept of natural neighbors introduced recently, a co-training method based on parameter-free and single-step unlabeled data selection strategy with natural neighbors (CT-NaN) is proposed in this paper. In CT-NaN, the confidence value of unlabeled samples is calculated in a parameter-free manner by analyzing the training data based on natural neighbors before the iteration of co-training, and it requires to be calculated only once in the whole process of co-training. Besides, CT-NaN is able to mitigate the negative effect of outliers because the training stops automatically when only outliers remain. Four groups of experiments with 22 data sets are conducted, and the results verify the effectiveness of CT-NaN when compared with 8 state-of-the-art co-training methods. [ABSTRACT FROM AUTHOR]

Pipelines are the most economical and sensible way to transport oil and gas. Long-distance oil and gas pipelines consist of many steel pipes or pipe fittings joined by welded girth welds, so girth welds are an essential part of pipelines. Owing to the limitations of welding conditions and the complexity of controlling weld quality in the field, some defects are inevitably present in girth welds and adjacent weld areas. These defects can lead to pipeline safety problems; therefore, it is necessary to perform failure risk assessment of pipeline girth welds. In this study, an artificial neural network model was proposed to predict the failure risk of pipeline girth welds with defects. Firstly, many pipeline girth weld failure cases, pipeline excavation, and inspection data were collected and analyzed to determine the main factors influencing girth weld failure. Secondly, a spatial orthogonal optimization method was used to select training samples for the artificial neural network model to ensure that the training sample set could cover the feature space with a minimum number of samples. Thirdly, a prediction model based on BP neural networks was established to predict the failure risk levels. The training dataset/testing dataset was 602/4215, and the prediction accuracy for all risks of girth welds achieved an acceptable level. This study can provide a valuable reference for pipeline operators to prevent pipeline accidents. [ABSTRACT FROM AUTHOR]

Problem definition: Gig economy companies benefit from labor flexibility by hiring independent workers in response to real-time demand. However, workers' flexibility in their work schedule poses a great challenge in terms of planning and committing to a service capacity. Understanding what motivates gig economy workers is thus of great importance. In collaboration with a ride-hailing platform, we study how on-demand workers make labor decisions; specifically, whether to work and work duration. Our model revisits competing theories of labor supply regarding the impact of financial incentives and behavioral motives on labor decisions. We are interested in both improving how to predict the behavior of flexible workers and understanding how to design better incentives. Methodology/results: Using a large comprehensive data set, we develop an econometric model to analyze workers' labor decisions and responses to incentives while accounting for sample selection and endogeneity. We find that financial incentives have a significant positive influence on the decision to work and on the work duration—confirming the positive income elasticity posited by the standard income effect. We also find support for a behavioral theory as workers exhibit income-targeting behavior (working less when reaching an income goal) and inertia (working more after working for a longer period). Managerial implications: We demonstrate via numerical experiments that incentive optimization based on our insights can increase service capacity by 22% without incurring additional cost, or maintain the same capacity at a 30% lower cost. Ignoring behavioral factors could lead to understaffing by 10%–17% below the optimal capacity level. Lastly, our insights inform the design of platform strategy to manage flexible workers amidst an intensified competition among gig platforms. Funding: This study was supported by The Jay H. Baker Retailing Center, The William and Phyllis Mack Institute for Innovation Management, The Wharton Risk Management and Decision Processes Center, and The Fishman-Davidson Center for Service and Operations Management. Supplemental Material: The online appendices are available at https://doi.org/10.1287/msom.2023.1191. [ABSTRACT FROM AUTHOR]

For the past forty years, Ethiopia has been promoting sustainable land management activities to enhance agricultural productivity. This study was intended to identify the factors determining farmers' adoption and continued use of soil bund measures in the Handosha watershed, Omo-Gibe river basin. A multistage sampling technique was used to select 340 households using the Heckman sample selection model. A total of 235 (69.12%) households adopted soil bunds, but only 89 (37.87%) of them were sustainably practicing soil bunds on their farm plots. Most adopters widely practiced soil bunds (49.42%), followed by stone bund (15.9%), and Fanyajuu (10%). The empirical results of the Heckman sample selection model showed that the farming experience, land tenure security, and perception of profitability of conservation measures were significantly positively affected the adoption of soil bund. Whereas, farm plot size and participation in off farm activities significantly negatively influenced the adoption of soil bund. Sustainable use of soil bund measures were significantly positively influenced by land tenure security, family size, and frequency of extension contact, whereas the distance between farm plots and home, and farm plot size were negatively affected. As a result, a design of agro-ecological-based soil and water conservation (SWC) measures was essential in reducing farmland vulnerability to soil erosion and food insecurity. It has been concluded that conservation practices should not only focus on the implementation and biophysical factors but also consider the socioeconomic interests of the farmers to improve the sustainable use of conservation technologies. [ABSTRACT FROM AUTHOR]

Various techniques have been applied to address residential self-selection (RSS) bias in estimating the impact of the built environment on travel behavior. We investigate how much the diverse results obtained are due to the method used, focusing on the most-often applied approaches of statistical control modeling, propensity score-based techniques, and sample selection modeling. First, a companion paper identifies 47 plausible ways to estimate a key quantity of interest, namely, the proportion of the total apparent effect of the built environment on travel behavior that is due to the built environment itself, which we call the “built environment proportion”, or BEP. Using a single sample, the present paper provides empirical estimates of 28 of the 47 BEP formulas, and creates and applies a framework for comparing those BEPs. We find that BEPs based on modular effects or treatment effects involving probability weights, and those applied to two-equation models, are among the ten best (based on the goodness of fit of the underlying models and on various measures of quality of the BEP). For our dataset, the ten best BEP averages fall between 0.506 and 0.772, with an overall average of 0.617. Thus, our best estimate is that for this application, about 38% of the total apparent effect of the built environment is due to RSS. However, the range of these averages indicates that even when controlling for as many confounding factors as possible, and even when using the best methods for computing the BEP, we still find substantial variation in the answer to the question: how much of the built environment's apparent total influence on travel behavior is due to self-selection?

Various techniques have been applied to address residential self-selection (RSS) bias in estimating the impact of the built environment on travel behavior. We investigate how much the diverse results obtained are due to the method used, focusing on the most-often applied approaches of statistical control modeling, propensity score-based techniques, and sample selection modeling. First, a companion paper identifies 47 plausible ways to estimate a key quantity of interest, namely, the proportion of the total apparent effect of the built environment on travel behavior that is due to the built environment itself, which we call the “built environment proportion”, or BEP. Using a single sample, the present paper provides empirical estimates of 28 of the 47 BEP formulas, and creates and applies a framework for comparing those BEPs. We find that BEPs based on modular effects or treatment effects involving probability weights, and those applied to two-equation models, are among the ten best (based on the goodness of fit of the underlying models and on various measures of quality of the BEP). For our dataset, the ten best BEP averages fall between 0.506 and 0.772, with an overall average of 0.617. Thus, our best estimate is that for this application, about 38% of the total apparent effect of the built environment is due to RSS. However, the range of these averages indicates that even when controlling for as many confounding factors as possible, and even when using the best methods for computing the BEP, we still find substantial variation in the answer to the question: how much of the built environment's apparent total influence on travel behavior is due to self-selection?

Abstract Infinite numbers of real-world applications use Machine Learning (ML) techniques to develop potentially the best data available for the users. Transfer learning (TL), one of the categories under ML, has received much attention from the research communities in the past few years. Traditional ML algorithms perform under the assumption that a model uses limited data distribution to train and test samples. These conventional methods predict target tasks undemanding and are applied to small data distribution. However, this issue conceivably is resolved using TL. TL is acknowledged for its connectivity among the additional testing and training samples resulting in faster output with efficient results. This paper contributes to the domain and scope of TL, citing situational use based on their periods and a few of its applications. The paper provides an in-depth focus on the techniques; Inductive TL, Transductive TL, Unsupervised TL, which consists of sample selection, and domain adaptation, followed by contributions and future directions.

Software Defect Prediction (SDP) is an important method to analyze software quality and reduce development cost. Data from software life cycle has been widely used to predict the defect prone of software modules, and although many machine learning-based SDP models have been proposed, their predictive performance is not always satisfactory. Traditional machine learning-based classifiers usually assume that all samples have the same contribution to the training of SDP, which is not true. In fact, different training samples have different effects on the performance of the SDP model, the performance of machine learning-based SDP models is heavily dependent on the quality of training samples. For the above shortcoming of traditional machine learning-based classifiers, the contributions of this paper are as follows: (1) Inspired by the clustering algorithm, a method to calculate the contribution of each training sample to the SDP model is proposed, which not only considers the relationship between the contributions of the training samples to the SDP model, and also analyzes the influence of the distance between the sample and the category boundary on the performance of the SDP model, so it is different from the existing calculation method of sample contribution. (2) A Sample Selection (SS) method is proposed to improve the performance of the SDP model. It first calculates the contribution of each training sample based on several nearest neighbors of the sample and the label information of these neighbors, and then implements SS according to Hoeffding probability inequality and the contribution of each sample. To confirm the validity of the proposed SDP model, some experimental results are given. Both direct observations and statistical tests of the experimental results show that the SS method is very effective for improving the predictive performance of the SDP model. [ABSTRACT FROM AUTHOR]

Intelligent detection of abnormal behaviors meets the need of engineering applications for identifying anomalies and alerting operators. However, most existing methods tackle the high-dimensional sequential video data with key frame extraction, which ignore the redundancy effect of inter- and intra- video frames. In this paper, a novel Abnormal Detection method based on double sparseness LSSVMoc (AD_LSSVMoc) is proposed, which combine both sample (i.e. frame) selection and feature selection simultaneously in a uniform sparse model. For the feature extraction, both handcrafted features and learned features are aggregated into effective descriptors. To achieve feature selection and sample selection, a improved LSSVMoc is proposed with sparse primal and dual optimization strategy, and alternating direction method of multipliers is used to solve the constrained linear equations problem raised in AD_LSSVMoc. Experiments show that the proposed AD_LSSVMoc method achieves a competitive detection performance and high detecting speed compared to state-of-the-art methods. [ABSTRACT FROM AUTHOR]

This paper addresses the performance degradation of deep neural networks caused by learning with noisy labels. Recent research on this topic has exploited the memorization effect: networks fit data with clean labels during the early stages of learning and eventually memorize data with noisy labels. This property allows for the separation of clean and noisy samples from a loss distribution. In recent years, semi-supervised learning, which divides training data into a set of labeled clean samples and a set of unlabeled noisy samples, has achieved impressive results. However, this strategy has two significant problems: (1) the accuracy of dividing the data into clean and noisy samples depends strongly on the network’s performance, and (2) if the divided data are biased towards the unlabeled samples, there are few labeled samples, causing the network to overfit to the labels and leading to a poor generalization performance. To solve these problems, we propose the curriculum regularization and adaptive semi-supervised learning (CRAS) method. Its key ideas are (1) to train the network with robust regularization techniques as a warm-up before dividing the data, and (2) to control the strength of the regularization using loss weights that adaptively respond to data bias, which varies with each split at each training epoch. We evaluated the performance of CRAS on benchmark image classification datasets, CIFAR-10 and CIFAR-100, and real-world datasets, mini-WebVision and Clothing1M. The findings demonstrate that CRAS excels in handling noisy labels, resulting in a superior generalization and robustness to a range of noise rates, compared with the existing method.

The precarious work construct combines employment instability and employment-contingent outcomes. Yet, I argue that confining the scope of the investigation to employed individuals creates a sample selection that disguises the heterogeneous nature of employment instability. The COVID-19 skyrocketing unemployment rate provides both a compelling motivation and a unique opportunity to revisit the construct of precarious work. Using pre-COVID and COVID-19 era data of the working-age population in Israel, the results demonstrate that by pushing less stable individuals out of employment, the COVID-19 recession strengthened the negative relationship between volatility and employment opportunities and accentuated sample selection. Because the selection into employment was not random, this introduces a bias into the measurement of precarious work, one that is more severe during a recession than in a full-employment market. The discussion highlights the broader significance of this lacuna and suggests a way to hone the conceptualization and operationalization of the precarious work construct. [ABSTRACT FROM AUTHOR]

Financial inclusion is one of the development priorities of the Ethiopian Government. However, factors, which might affect it are not explored adequately with appropriate methods. To examine the relationship between financial inclusion and its demand-side determinants, we applied a simple probit model and two probit models with sample selection to the 2018/19 Ethiopia Socioeconomic Survey (ESS4). Our simple probit model estimation results show that male, older, non-Muslim, married, more educated, richest, mobile users or people who are aware of how to open a bank account are more likely to have a formal account than their counterparts. However, being from a rural area or from a larger household size reduces the likelihood of having an account. Results from the probit selection model show that rural residents, employed, richer people, or people who are out of labor force are more likely to use their account to save money than their counterparts. Furthermore, the third model shows that being non-Muslim, married, widowed, or poorer reduces the likelihood of account use than their counterparts. Thus, policies should target the most vulnerable groups of the society. Especially, rural households, women, the Muslim population, the poor, less educated people, and the young should be the focus of financial inclusion endeavors of the Government. Parallel to these policy measures, awareness of opening bank account and financial education programs should be promoted. [ABSTRACT FROM AUTHOR]

• A pioneering framework called Cross-to-Merge Training (C2MT) is proposed for learning with noisy labels (LNL). • We are the first to perform parameter aggregation on LNL after cross-training. • We introduce a novel class-balancing strategy, the Median Balance Strategy (MBS), which proves useful for sample selection. • C2MT demonstrates robustness with respect to hyper-parameters and network architectures. • Comprehensive experimental results substantiate the advantages of C2MT. The collection of large-scale datasets inevitably introduces noisy labels, leading to a substantial degradation in the performance of deep neural networks (DNNs). Although sample selection is a mainstream method in the field of learning with noisy labels, which aims to mitigate the impact of noisy labels during model training, the testing performance of these methods exhibits significant fluctuations across different noise rates and types. In this paper, we propose Cross-to-Merge Training (C2MT), a novel framework that is insensitive to the prior information in sample selection progress, enhancing model robustness. In practical implementation, using cross-divided training data, two different networks are cross-trained with the co-teaching strategy for several local rounds, subsequently merged into a unified model by performing federated averages on the parameters of two models periodically. Additionally, we introduce a new class balance strategy, named Median Balance Strategy (MBS), during the cross-dividing process, which evenly divides the training data into a labeled subset and an unlabeled subset based on the estimated loss distribution characteristics. Extensive experimental results on both synthetic and real-world datasets demonstrate the effectiveness of C2MT. The Code will be available at: https://github.com/LanXiaoPang613/C2MT. [ABSTRACT FROM AUTHOR]

Infinite numbers of real-world applications use Machine Learning (ML) techniques to develop potentially the best data available for the users. Transfer learning (TL), one of the categories under ML, has received much attention from the research communities in the past few years. Traditional ML algorithms perform under the assumption that a model uses limited data distribution to train and test samples. These conventional methods predict target tasks undemanding and are applied to small data distribution. However, this issue conceivably is resolved using TL. TL is acknowledged for its connectivity among the additional testing and training samples resulting in faster output with efficient results. This paper contributes to the domain and scope of TL, citing situational use based on their periods and a few of its applications. The paper provides an in-depth focus on the techniques; Inductive TL, Transductive TL, Unsupervised TL, which consists of sample selection, and domain adaptation, followed by contributions and future directions. [ABSTRACT FROM AUTHOR]

The demand and supply gap in local fish production provides opportunities for investment in fish production and processing that are non-artisanal. Accordingly, this paper uses the Qatar Semi-Annual Survey, a nationally representative repeated cross-sectional dataset, to analyze consumers' willingness to consume processed fish and the consumption intensity. Given the ordinal nature of the main outcome variable, namely processed fish consumption intensity, we use the ordered probit model with sample selection. The first part of our analytical strategy helps us to examine determinants of processed fish consumption, while the second part establishes factors associated with processed fish consumption intensity. The findings suggest that the majority of the respondents have a low interest in consuming processed fish. Nonetheless, we find that demographic and socioeconomic factors influence the intensity of consuming processed fish in Qatar. Specifically, consuming processed fish sometimes or very often is positively associated with age, living in a household with no children, and labor force participation. In comparison, Qatari households, tertiary education, and living in a middle-income household are factors that are likely to reduce consuming processed fish sometimes or very often. Our findings have important implications for investment in processed fish in Qatar. Specifically, the results help in targeting different consumer segments given a better understanding of factors associated with processed fish and fishery product consumption and the intensity of consumption among consumers in Qatar. The insights drawn from this study are a fundamental precondition for advancing the processed fish and fishery product industry in Qatar, an industry with the potential to generate by-products with benefits, such as energy and animal feed, prolonging fish and fishery product shelf life and convenience for consumers. Moreover, other countries in the region are likely to benefit from the insights drawn from this study. [ABSTRACT FROM AUTHOR]

The building damage caused by natural disasters seriously threatens human security. Applying deep learning algorithms to identify collapsed buildings from remote sensing images is crucial for rapid post-disaster emergency response. However, the diversity of buildings, limited training dataset size, and lack of ground-truth samples after sudden disasters can significantly reduce the generalization of a pre-trained model for building damage identification when applied directly to non-preset locations. To address this challenge, a self-incremental learning framework (i.e., SELF) is proposed in this paper, which can quickly improve the generalization ability of the pre-trained model in disaster areas by self-training an incremental model using automatically selected samples from post-disaster images. The effectiveness of the proposed method is verified on the 2010 Yushu earthquake, 2023 Turkey earthquake, and other disaster types. The experimental results demonstrate that our approach outperforms state-of-the-art methods in terms of collapsed building identification, with an average increase of more than 6.4% in the Kappa coefficient. Furthermore, the entire process of the self-incremental learning method, including sample selection, incremental learning, and collapsed building identification, can be completed within 6 h after obtaining the post-disaster images. Therefore, the proposed method is effective for emergency response to natural disasters, which can quickly improve the application effect of the deep learning model to provide more accurate building damage results.

The face identification problem is the task of identifying incoming users from their face images. The applications of this task include automating user identification at the building entrance. Consequently, many proposed algorithms are proposed for the given task ranging from handcraft computational models to deep learning models. This paper utilizes the existing effective algorithms and proposes the template selection strategy to enhance its recognition performance when the enrollment of multiple samples for each user is proposed. Besides, the paper investigates the effect of image quality on the recognition performance of the system and the efficacy of the proposed template selection strategy when applied in such a situation. Experiments are performed on the in-house dataset collected from the Thai population -- to evaluate the empirical system performance when the system is deployed in Thailand to automatically identify the user at the building gate -- as well as on the LFW public dataset consisting of 13,000 face images of 5,749 individuals with multiple ethnicities. The results show that, with the proposed sample selection, the identification error on the in-house dataset for the close-set identification decreased from 3.11 to 2.46% at 0.11% and 0.06% FMR. For the open-set identification, the FNMR decreased from 9.36% to 5.40% at 6.20% and 3.23% FAR, respectively. In addition, the experiments on the LFW dataset have also demonstrated the efficacy of the proposed system consisting of the proposed sample selection method and the selection of face recognition modules. That is, the performance improvement is noticeable as compared to the baseline where the sample selection method is not deployed and to the previous work when multiple samples are used for enrollment. [ABSTRACT FROM AUTHOR]

Misclassification of a binary response variable and nonrandom sample selection are data issues frequently encountered by empirical researchers. For cases in which both issues feature simultaneously in a data set, we formulate a sample selection model for a misclassified binary outcome in which the conditional probabilities of misclassification are allowed to depend on covariates. Assuming the availability of validation data, the pseudo-maximum likelihood technique can be used to estimate the model. The performance of the estimator accounting for misclassification and sample selection is compared to that of estimators offering partial corrections. An empirical example illustrates the proposed framework. [ABSTRACT FROM AUTHOR]

The rhenium-osmium (Re-Os) geochronometer is newly developed technology and has been regarded as a promising radiometric tool for the direct dating of the deposition of source rock, hydrocarbon generation, migration and consequent alteration (e.g., thermal cracking, thermochemical sulphate reduction), and oil-source correlation. However, the complex nature of Re-Os systems in organi-rich samples, including their relatively low elemental contents, multiple origins and occurrences, ambiguous homogenization mechanisms of 187Os/188Os and our limited understanding of the closed system of Re-Os systematics etc., often result in the relatively large uncertainties of obtained isochrones. In this paper, to improve their data quality and yield ideal Re-Os isochrons, the existing Re-Os isotopic analysis method was improved by the terms of sample selection and screening, chemical pretreatment, total analytical blank control and international reference material monitoring. Results showed that a comprehensive guide to researching working in Re-Os geochronology for petroleum system analysis can be available.

Existing literature has applied the sample selection modeling approach to disentangle the influence of the built environment (BE) and residential self-selection (RSS) on travel behavior. However, a limitation of the existing sample selection models is that they can handle only continuous or ordinal outcomes. The contribution of this study is twofold. First, we develop a sample selection model that can handle binary travel behavior outcomes in the RSS context. When the travel behavior outcome is binary, this approach's potential parameter identification issue can become serious. We employ a non-flat prior and Watanabe-Akaike information criterion in the Bayesian framework to address this issue. Second, we apply this proposed model to travel survey data in Kumamoto City, Japan, to disentangle the BE influence of a neighborhood type and RSS on car ownership. The neighborhood type is defined as the neighborhood being either less than 1,500 m (A) or greater than 1,500 m (B) from a station. We reveal that the true influence of the neighborhood type results in a mere 2.1 percentage point decrease in the car ownership probability. Additionally, we find that the share of the total BE influence (including the RSS influence) owing to RSS on the householder's car ownership is 45.7%. The proposed model is a new and useful tool for quantifying the influence of BE and the relative influence of RSS on binary travel behavior. • Sample selection model handling a binary travel behavior outcome. • It quantifies residential self-selection (RSS) and built environment (BE) effects. • Parameter identification issue is addressed using informative priors. • We check the influence of informative priors on predictive accuracy. • An empirical example that RSS offsets the true influence of BE. [ABSTRACT FROM AUTHOR]

Visual tracking is typically solved as a discriminative learning problem that usually requires high-quality samples for online model adaptation. It is a critical and challenging problem to evaluate the training samples collected from previous predictions and employ sample selection by their quality to train the model. To tackle the above problem, we propose a joint discriminative learning scheme with the progressive multistage optimization policy of sample selection for robust visual tracking. The proposed scheme presents a novel time-weighted and detection-guided self-paced learning strategy for easy-to-hard sample selection, which is capable of tolerating relatively large intraclass variations while maintaining interclass separability. Such a self-paced learning strategy is jointly optimized in conjunction with the discriminative tracking process, resulting in robust tracking results. Experiments on the benchmark datasets demonstrate the effectiveness of the proposed learning framework. [ABSTRACT FROM AUTHOR]

Exploiting informative endogenously selected samples while minimizing sample selection bias with minimum modeling assumptions and analytical burden. Unlike random sampling, selective sampling draws units based on the outcome values, such as oversampling rare events in choice outcomes and extreme activities on continuous and count outcomes. Despite high cost-effectiveness for marketing research, such endogenously selected samples must be carefully analyzed to avoid selection bias. We introduce a unified and efficient approach based on semiparametric odds ratio (SOR) models applicable for categorical, continuous and count response data collected using selective sampling. Unlike extant sampling-adjusting methods and Heckman-type selection models, the proposed approach requires neither modeling selection mechanisms nor imposing parametric distributional assumptions on the response variables, eliminating both sources of mis-specification bias. Using this approach, one can quantify and test for the relationships among variables as if samples had been collected via random sampling, simplifying bias correction of endogenously selected samples. We evaluate and illustrate the method using extensive simulation studies and two real data examples: endogenously stratified sampling for linear/nonlinear regressions to identify drivers of the share-of-wallet outcome for cigarettes smokers and using truncated and on-site samples for count data models of store shopping demand. The evaluation shows that selective sampling followed by applying the SOR approach reduces required sample size by more than 70% compared with random sampling and that in a wide range of selective sampling scenarios SOR offers novel solutions outperforming extant methods for selective samples with opportunities to make better managerial decisions. [ABSTRACT FROM AUTHOR]

Abstract Background Population-representative household survey methods require up-to-date sampling frames and sample designs that minimize time and cost of fieldwork especially in low- and middle-income countries. Traditional methods such as multi-stage cluster sampling, random-walk, or spatial sampling can be cumbersome, costly or inaccurate, leading to well-known biases. However, a new tool, Epicentre’s Geo-Sampler program, allows simple random sampling of structures, which can eliminate some of these biases. We describe the study design process, experiences and lessons learned using Geo-Sampler for selection of a population representative sample for a kidney disease survey in two sites in Guatemala. Results We successfully used Epicentre’s Geo-sampler tool to sample 650 structures in two semi-urban Guatemalan communities. Overall, 82% of sampled structures were residential and could be approached for recruitment. Sample selection could be conducted by one person after 30 min of training. The process from sample selection to creating field maps took approximately 40 h. Conclusion In combination with our design protocols, the Epicentre Geo-Sampler tool provided a feasible, rapid and lower-cost alternative to select a representative population sample for a prevalence survey in our semi-urban Guatemalan setting. The tool may work less well in settings with heavy arboreal cover or densely populated urban settings with multiple living units per structure. Similarly, while the method is an efficient step forward for including non-traditional living arrangements (people residing permanently or temporarily in businesses, religious institutions or other structures), it does not account for some of the most marginalized and vulnerable people in a population–the unhoused, street dwellers or people living in vehicles.

As the early stage of Alzheimer's disease (AD), mild cognitive impairment (MCI) has high chance to convert to AD. Effective prediction of such conversion from MCI to AD is of great importance for early diagnosis of AD and also for evaluating AD risk pre-symptomatically. Unlike most previous methods that used only the samples from a target domain to train a classifier, in this paper, we propose a novel multimodal manifold-regularized transfer learning (M2TL) method that jointly utilizes samples from another domain (e.g., AD vs. normal controls (NC)) as well as unlabeled samples to boost the performance of the MCI conversion prediction. Specifically, the proposed M2TL method includes two key components. The first one is a kernel-based maximum mean discrepancy criterion, which helps eliminate the potential negative effect induced by the distributional difference between the auxiliary domain (i.e., AD and NC) and the target domain (i.e., MCI converters (MCI-C) and MCI non-converters (MCI-NC)). The second one is a semi-supervised multimodal manifold-regularized least squares classification method, where the target-domain samples, the auxiliary-domain samples, and the unlabeled samples can be jointly used for training our classifier. Furthermore, with the integration of a group sparsity constraint into our objective function, the proposed M2TL has a capability of selecting the informative samples to build a robust classifier. Experimental results on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database validate the effectiveness of the proposed method by significantly improving the classification accuracy of 80.1 % for MCI conversion prediction, and also outperforming the state-of-the-art methods.

This paper employs a dataset of 7,500 Northern Irish farms over the period 2015-2019 to investigate the factors that affect the number of workdays lost in agriculture, one of the most hazardous sectors in terms of occupational injuries. Results indicate that public policies aimed at improving farm safety should focus on dairy farms, young workers, family members other than the main farmer, and dangerous working practices related to machineries and vehicles. Additionally, results indicate that more than 18,000 workdays are lost every year on Northern Irish farms. [ABSTRACT FROM AUTHOR]

Special core analysis (SCAL) measurements play a noteworthy role in reservoir engineering. Due to the time-consuming and costly character of these measurements, routine core analysis (RCAL) data should be inspected thoroughly to select a representative subset of samples for SCAL. There are no comprehensive guidelines on how representative samples should be selected. In this study, a new framework is presented for selection of representative samples for SCAL. The foundation of this framework is using methods of PSRTI, FZI∗ (FZI-star) and TEM-function for the early estimation of petrophysical static, dynamic, and pseudo-static rock types at RCAL stage. The global hydraulic element (GHE) approach is benefitted and a FZI∗-based GHE method (i.e., GHE∗) is presented for partitioning data. The framework takes into consideration different laboratory, reservoir engineering, geological, petrophysical and statistical factors. A carbonate reservoir case is presented to support our methodology. We also show that the current forms of Lorenz and Stratigraphic Modified Lorenz Plots in reservoir engineering are not appropriate, and present new forms of them.

In order to solve the problem that low classification accuracy caused by the different distribution of training set and test set, an unsupervised domain adaptation method based on discriminant sample selection (DSS) is proposed. DSS projects the samples of different domains onto a same subspace to reduce the distribution discrepancy between the source domain and the target domain, and weights the source domain instances to make the samples more discriminant. Different from the previous method based on the probability density estimation of samples, DSS tries to obtain the sample weights by solving a quadratic programming problem, which avoids the distribution estimation of samples and can be applied to any fields without suffering from the dimensional trouble caused by high-dimensional density estimation. Finally, DSS congregates the same classes by minimizing the intra-class distance. Experimental results show that the proposed method improves the classification accuracy and robustness.

Due to the wide applications of deep learning in the field of urban rail transit passenger flow forecasting, the selection problem of training samples has become increasingly more worthy of researchers' attention, as it is closely related to urban rail transit passenger flow time series. Therefore, it is necessary to study the distribution characteristics of the contribution degree of the training sample to guide sample selection in the deep learning training process. In this study, based on the prediction accuracy and the sample contribution degree, the optimal sample contribution combination algorithm (GWO-SCBP) was ultimately generated by the grey wolf optimizer (GWO) and error back propagation (EBP) algorithms. The contribution of training samples for each station of the Xi'an metro network was calculated and analyzed. The results show that the sample contribution is not only related to the distance between the sample and predicted value, but is also closely related to the station flow characteristics. By classifying the network stations and fitting the contribution degree of the central station of each type of station, linear equations of sample contribution degree were obtained, and the R2 values attained at least 0.65, indicating a good fitting effect.

Vision-based human detection is a fundamental task in visual content analysis. It has a wide range of applications, especially for person search and retrieval. To reduce the reliance of detection models on large amount of labeled data, we modify Faster R-CNN to facilitate semi-supervised human detection. Specifically, a Reliability Analysis (RA) module is included as an add-on into our Self-Enhanced R-CNN (SE-RCNN) model. The unlabeled images can be pseudo-annotated reliably under the help of this module. As a result, both labeled and unlabeled data are fed simultaneously for model optimization. The additional supervision, in turn, guides the training of a detection module in our model. The two aspects, extracting precise proposals and generating reliable pseudo annotations, can be mutually reinforced. Unlike previous related works, it is the first attempt to build a single-stage semi-supervised human detection model. In our experiment, we observe that the RA module plays an important role in exploiting unlabeled data and leads to state-of-the-art results of SE-RCNN on multiple benchmarks.

Consuming too much fat, sugar, and salt is associated with adverse health outcomes. Food reformulation is one possible strategy to enhance the food environment by improving the nutritional quality of commercial products. However, food reformulation faces many hindrances. One way to alleviate some of these hindrances is to embrace a multicriteria approach that is based on a market inventory. In this objective, additional sensory screening and water content analyses allow going beyond nutrition and composition information on the packaging. However, due to feasibility reasons for later in-depth analyses, it is necessary to work with several reduced and manageable products. To the best of the authors' knowledge, in the literature, there is no sample selection approach taking into account multiple criteria as a base for future food reformulation. The overall aim of this paper is to propose a method to select the best representative products from the market base, for future reformulation by going beyond nutrition and composition information on the packaging. This approach considered therefore nutrition, composition, economic, water content, and sensory information with the example of the cookies market. The first step is the creation of an extensive cookie database including sensory and water content information. In total 178 cookies among the French market were identified, then focus was placed on 62 chocolate chip cookies only. Sensory screening and water content analyses of all 62 products were conducted. The second step is to make an informed subset selection, thanks to a cluster analysis based on 11 nutrition, composition, and water content variables. A representative subset of 18 cookies could be derived from the obtained clusters. The representativity was evaluated with statistical uni- and multivariate analyses. Results showed a broad variety of chocolate chips cookies with a large nutritional, compositional, water content, and sensory differences. These results highlighted the first paths for future reformulation in this product category and showed the importance to include physical product information beyond the information on the packaging. This complete database on the selected cookies constituted a solid base for identifying future reformulation levers, in order to improve the nutritional quality and health.