Low-quality Data Research Articles

Targeted Training for Multi-organization Recommendation

Making recommendations for users in diverse organizations ( orgs ) is a challenging task for workplace social platforms such as Microsoft Teams and Slack. The current industry-standard model training approaches either use data from all organizations to maximize information or train organization-specific models to minimize noise. Our real-world experiments show that both approaches are poorly suited for the multi-org recommendation setting where different organizations’ interaction patterns vary in their generalizability. We introduce targeted training , which improves on standard practices by automatically selecting a subset of orgs for model development whose data are cleanest and best represent global trends. We demonstrate how and when targeted training improves over global training through theoretical analysis and simulation. Our experiments on large-scale datasets from Microsoft Teams, SharePoint, Stack Exchange, DBLP, and Reddit show that in many cases targeted training can improve mean average precision (MAP) across orgs by 10–15% over global training, is more robust to orgs with lower data quality, and generalizes better to unseen orgs. Our training framework is applicable to a wide range of inductive recommendation models, from simple regression models to graph neural networks (GNNs).

Metabolic multireactor: Practical considerations for using simple oxygen sensing optodes for high-throughput batch reactor metabolism experiments.

We present a system for carrying out small batch reactor oxygen consumption experiments on water and sediment samples for environmental questions. In general, it provides several advantages that can help researchers achieve impactful experiments at relatively low costs and high data quality. In particular, it allows for multiple reactors to be operated and their oxygen concentrations to be measured simultaneously, providing high throughput and high time-resolution data, which can be advantageous. Most existing literature on similar small batch-reactor metabolic studies is limited to either only a few samples, or only a few time points per sample, which can restrict the ability for researchers to learn from their experiments. The oxygen sensing system is based very directly on the work of Larsen, et al. [2011], and similar oxygen sensing technology is widely used in the literature. As such we do not delve deeply into the specifics of the fluorescent dye sensing mechanism. Instead, we focus on practical considerations. We describe the construction and operation of the calibration and experimental systems, and answer many of the questions likely to come up when other researchers choose to build and operate a similar system themselves (questions we ourselves had when we first built the system). In this way, we hope to provide an approachable and easy to use research article that can help other researchers construct and operate a similar system that can be tailored to ask their own research questions, with a minimum of confusion and missteps along the way.

THE USE OF ARTIFICIAL INTELLIGENCE AND NEURAL NETWORKS IN THE EDUCATIONAL PROCESS IN PROFESSIONAL DISCIPLINES BY STUDENTS MAJORING IN PROFESSIONAL EDUCATION (DIGITAL TECHNOLOGIES)

This article examines the possibilities and prospects of using artificial intelligence and neural networks in student education. The use of these technologies can have a positive impact on the quality of education and provide more effective knowledge transfer. The authors of the article explore the possibilities of using neural networks in the study of professional disciplines by students of the specialty «Vocational Education (Digital Technologies)». Among the advantages of using artificial intelligence and neural networks in education are the optimization of the learning process, adaptive and individualized learning, increased motivation, and automatic assessment of knowledge and skills. Additionally, the possibility of using neural networks for data analysis is considered, which will allow students to improve the quality of research and increase the accuracy of results. Moreover, artificial intelligence systems can help ensure more effective remote learning. Furthermore, artificial intelligence can be applied to create virtual simulators and trainers, allowing students to acquire practical skills and experience without risking experimentation on real objects. For instance, this can be useful for students who study computer repair and modernization, where practical experience and safe and efficient equipment handling are essential.
 Despite the advantages of using these technologies, such as personalized learning and increased efficiency, potential drawbacks should also be considered. Possible drawbacks include low-quality source data, potential errors in the neural network's operation, and the possibility of ethical problems. To minimize the negative impact of these drawbacks, it is necessary to use technologies that consider their limitations and risks, as well as to constantly improve the technical base and increase the qualifications of teachers and students. Therefore, the use of artificial intelligence and neural networks in education is a promising and powerful tool that can improve the quality of education, make it more efficient and personalized, and provide students with the skills and knowledge they need to succeed in the digital age.

Doubts on the reliability of parallel corpus filtering

Parallel corpus filtering (PCF) aims to filter out low-quality data residing in parallel corpora. Recently, deep learning-based methods have been employed to assess the quality of sentence pairs in a parallel corpus, along with rule-based filtering that filters out noisy data depending on the pre-defined error types. Despite their utilization, to the best of our knowledge, a comprehensive investigation into the practical applicability and interpretability of PCF techniques remains unexplored. In this study, we raise two doubts on deep learning-based PCF: (i) Can deep learning-based PCF extract high-quality data? and (ii) Are scoring functions of PCF reliable? To answer these questions, we conduct comparative experiments on various PCF techniques with four datasets on two language pairs, English–Korean, and English–Japanese. Through the experiments, we demonstrate that the performance of the deep learning-based PCF highly depends on the targeting parallel corpus, and shows fluctuating adaptability depending on their characteristics. In particular, we figure out that high-scored sentences derived by the PCF technique do not necessarily guarantee high-quality results, rather it shows unintended preference.

Topic-Clustering Model with Temporal Distribution for Public Opinion Topic Analysis of Geospatial Social Media Data

Analysis of the spatiotemporal distribution of online public opinion topics can help understand the hotspots of public concern. The topic model is employed widely in public opinion topic clustering for social media data. In order to handle topic-clustering of low-quality geospatial social media data, such as microblog data, with short text and timeliness characteristics, this study proposed a Dirichlet multinomial mixture over time (DMMOT) model to cluster microblog topic for public opinion analysis. The DMMOT model assumes that a single document belongs to a single topic, in line with the characteristics of a short text, and it introduces the probability distribution of “topic-time” in the process of topic generation. The model parameter inference process was presented in detail by exploring the Gibbs sampling method. Results generated using the DMMOT model in case study show that the “topic-word” distribution is semantically aggregated within various topics, and “topic-time” distribution clustered within a time window under each topic. Furthermore, the characteristics of the trend of each topic over time are basically consistent with the corresponding trend of topic in reality in terms of content. These indicate that the DMMOT model improves topic clustering for short text to some extent. Furthermore, the DMMOT model performed well in both temporal and spatial analysis of public opinion topics based on microblog data.

Calibration Assessment of Low-Cost Carbon Dioxide Sensors Using the Extremely Randomized Trees Algorithm.

As the monitoring of carbon dioxide is an important proxy to estimate the air quality of indoor and outdoor environments, it is essential to obtain trustful data from CO2 sensors. However, the use of widely available low-cost sensors may imply lower data quality, especially regarding accuracy. This paper proposes a new approach for enhancing the accuracy of low-cost CO2 sensors using an extremely randomized trees algorithm. It also reports the results obtained from experimental data collected from sensors that were exposed to both indoor and outdoor environments. The indoor experimental set was composed of two metal oxide semiconductors (MOS) and two non-dispersive infrared (NDIR) sensors next to a reference sensor for carbon dioxide and independent sensors for air temperature and relative humidity. The outdoor experimental exposure analysis was performed using a third-party dataset which fit into our goals: the work consisted of fourteen stations using low-cost NDIR sensors geographically spread around reference stations. One calibration model was trained for each sensor unit separately, and, in the indoor experiment, it managed to reduce the mean absolute error (MAE) of NDIR sensors by up to 90%, reach very good linearity with MOS sensors in the indoor experiment (r2 value of 0.994), and reduce the MAE by up to 98% in the outdoor dataset. We have found in the outdoor dataset analysis that the exposure time of the sensor itself may be considered by the algorithm to achieve better accuracy. We also observed that even a relatively small amount of data may provide enough information to perform a useful calibration if they contain enough data variety. We conclude that the proper use of machine learning algorithms on sensor readings can be very effective to obtain higher data quality from low-cost gas sensors either indoors or outdoors, regardless of the sensor technology.

Giving critical theoretical and evaluation practice context to ‘Testing, Testing’ in Jamaica

This article examines how evaluability assessments can be designed and conducted to generate meaningful and actionable results for reducing youth crime in Jamaica. It draws on case illustrations from the 2021 ‘Testing, Testing’ CAPRI research study report to highlight the nexus between research and evaluation and promote adopting practices that can enhance the quality and relevance of evaluations conducted within the Caribbean region and beyond. The findings suggest that data needed to inform evaluation and evidenced-based policymaking remain limited despite re-energised regional and global efforts to address this challenge. Particularly, low data quality, production, and use continue to hinder measuring progress on the Sustainable Development Goals. The article, therefore, offers an accessible decolonising framework for understanding, supporting and critically assessing how inertia within the context of evaluation practice can be challenged. Concrete suggestions for action include using local knowledge and underutilised techniques such as evaluability assessments.

Machine Learning Approach Empowers Well Placement in Tight Gas Field

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper IPTC 22188, “Machine-Learning-Empowered Well Placement in a Large Unconventional Tight Gas Field in China,” by Ting Yu, SLB; Xiangzeng Wang, Shaanxi Yanchang Petroleum; and Alexis Carrillat, SLB, et al. The paper has not been peer reviewed. Copyright 2022 International Petroleum Technology Conference. Reproduced by permission. _ A tight gas field produces gas from a large heterogeneous fluvial reservoir with limited well control in margins and appraisal areas and sparse advanced logs. The principal goal of the study described in the complete paper was to provide a novel solution using machine-learning (ML) techniques to predict sandstone distribution and, to some extent, automate the process of optimizing well placement. The presented work flow overcomes low data quality, scaling, and inconsistency and builds the bridge between geoscience and artificial intelligence (AI) software platforms. Introduction The field is an unconventional gas reservoir covering a vast area in the Ordos Basin in China. This field produces gas from a complex and heterogeneous fluvial reservoir with limited well control for prediction of sand-body distribution. Reservoir distribution and quality prognosis carry large uncertainties. The well spacing between exploration and appraisal wells is approximately 3–4 km; well control is sparser in resource-overlaying areas. As such, reservoir prediction using only well data is marked by uncertainties and can result in high geological risk in placing development wells. In addition to well data, 2D seismic lines cover the full field with a spacing of 2×4 km. The described work flow aims to use all geological and geophysical (G&G) information effectively to optimize well placement and minimize risk of dry wells. It can combine G&G software and AI platforms to deepen data mining and analyze correlations between the seismic waveform and sandstone distribution. This is achieved by using advanced ML algorithms to construct a 3D heterogeneity model with multiple variables. It enables quick organization of data for analysis and interpretation and becomes a tool for supporting real-time drilling. Objective The challenge in this fluvial reservoir is to replicate the sand-body-stacking relationships based on sparse and multiscale data integration. With that objective in mind, ML is an effective tool to automate geoscience work flows. In this study, a new work flow was created that takes advantage of ML in subsurface modeling for efficient and computationally inexpensive forecasting. This method incorporates ML in the context of decision support. The main objectives of the study include the following: - Automatically classify seismic attributes into waveforms and establish their correspondence to log motifs - Improve the effectiveness of 2D seismic data integration with well-log data - Generate a reservoir risk map based on seismic and log-data integration - Construct an advanced 3D geological model with multiple variables - Generate a more-objective interpretation by having less human inference in the data analysis - Bridge the gap between G&G and AI software platforms and lift current G&G software limitations This enabled use of advanced decision-tree-based ML algorithms to generate a geological model.

ANALYTICAL SUPPPORT FOR BANK LOAN TRANSACTIONS

The article emphasizes the importance of analytics in the effective organization of bank loan operations. The purpose of the article is to determine the range of problems currently faced by analysts of loan departments of Ukrainian banks, highlighting urgent issues related to the economic crisis caused by the war, and to outline ways to solve them using the latest and modern methods.
 The article proves that the organization of effective analytical support of loan transactions is extremely important both for the bank and its borrowers, and for the state as a whole. However, there are a number of problems that prevent its effective development. Among the main ones, there are such issues as processing a large number of arrays of data on customers and their financial transactions as well as the low quality of such data; structuring of data from separate isolated databases, which complicates the process of their analysis and processing; need for complying with regulatory requirements and standards; deterioration in the quality of analytics due to the increased speed of response to requests for loan, etc.
 The article reveals the problems of analytical support for bank loan operations, which become relevant during the law and the economic crisis. These are an increase in the risk of loan defaults, excessive regulatory intervention, changes in market conditions and technologies, a decrease in the liquidity of the banking system, insufficient information about the client, excessive reliance on “historical” data and the rise of fraud.
 The study proposes means of solving the outlined problems, the general directions of which are risk-orientation, interaction with clients, internal communications and technology development. Attention is focused on the need for banks to improve their analytical tools using the latest technologies, such as machine learning and artificial intelligence, blockchain technologies, tools for real-time analytics, Big Data technologies, analytical robots and dashboards. The use of the specified methods will allow banks to organize proper analytical support for loan operations and will enable effective management decisions to be made quickly.

Spatially-explicit mapping annual oil palm heights in peninsular Malaysia combining ICESat-2 and stand age data

Oil palm is the most efficient oil-producing crop but its extension leads to increased deforestation in Southeast Asia. Oil palm height enables the quantitative estimation for carbon stock or palm oil yield. Nevertheless, there are still no accurate characterization of oil palm height providing information for the tradeoff between forest damage and carbon stock of oil palm in Southeast Asia. The new generation of spaceborne LiDAR provides large-extent canopy height samples, offering an opportunity for mapping the oil palm height at the regional scale yet with challenge to extrapolate the footprint heights to spatially coherent maps. Here, we proposed a new method by combining Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) footprint data with stand age that is closely related to tree height. We first developed a semi-automatic filtering algorithm to filter the low-quality ICESat-2 data, and used a change detection algorithm to optimize the planting year map of oil palm. Then, an empirical age-height model was derived by linking ICESat-2 footprint canopy height with oil palm age that was used to estimate spatially and temporally oil palm height for the whole Peninsular Malaysia. A validation with independent ICESat-2 footprint data suggests a high agreement for the height estimates in 2020 from the age-height model (R2 = 0.63; RMSE = 1.64 m) with a bias within ±3 m for >90% of the height estimates. Using the age-height model and planting year map of oil palm, we produced the first comprehensive wall-to-wall maps of long-term yearly oil palm height at a spatial resolution of 30 m in Peninsular Malaysia during 2001 through 2020. Our results suggest that the mean height of oil palm in all and regionally-disturbed areas have increased by 10.82 m and 9.29 m respectively during the last two decades in Peninsular Malaysia. Our results indicate that combining stand age and ICESat-2 footprint data has great potential in spatially-explicit mapping regional oil palm height that contributes to a better quantification for regional plantation carbon stock.

Dissociating distinct cortical networks associated with subregions of the human medial temporal lobe using precision neuroimaging

Tract-tracing studies in primates indicate that different subregions of the medial temporal lobe (MTL) are connected with multiple brain regions. However, no clear framework defining the distributed anatomy associated with the human MTL exists. This gap in knowledge originates in notoriously low MRI data quality in the anterior human MTL and in group-level blurring of idiosyncratic anatomy between adjacent brain regions, such as entorhinal and perirhinal cortices, and parahippocampal areas TH/TF. Using MRI, we intensively scanned four human individuals and collected whole-brain data with unprecedented MTL signal quality. Following detailed exploration of cortical networks associated with MTL subregions within each individual, we discovered three biologically meaningful networks associated with the entorhinal cortex, perirhinal cortex, and parahippocampal area TH, respectively. Our findings define the anatomical constraints within which human mnemonic functions must operate and are insightful for examining the evolutionary trajectory of the MTL connectivity across species.

Block size, parallelism and predictive performance: finding the sweet spot in distributed learning

As distributed and multi-organization Machine Learning emerges, new challenges must be solved, such as diverse and low-quality data or real-time delivery. In this paper, we use a distributed learning environment to analyze the relationship between block size, parallelism, and predictor quality. Specifically, the goal is to find the optimum block size and the best heuristic to create distributed Ensembles. We evaluated three different heuristics and five block sizes on four publicly available datasets. Results show that using fewer but better base models matches or outperforms a standard Random Forest, and that 32 MB is the best block size.

Three-way decision based participants selection optimization model in sparse mobile crowdsensing

Aiming at the problem of low data quality caused by the low quality of participants in sparse mobile crowdsensing, a participant's selection optimization algorithm based on the three-way decision is proposed in this paper. Firstly, a similarity matrix is predicted according to the correlation between Spatio-temporal cells, constructing the classification model. Secondly, the historical behavior of participants is analyzed to extract the characteristic information of the target participants for the task. Finally, with the help of the idea of the three-way decision, the high-quality and reliable candidates are selected according to the participant's characteristic information. Then the participants who perform the task are screened under the guidance of the classification model. The experimental results based on real datasets show that the algorithm proposed in this paper can effectively improve the quality of sensing data and reduce the sensing cost.

Context-aggregator: An approach of loss- and class imbalance-aware aggregation in federated learning

Federated Learning (FL) is an emerging distributed learning paradigm which offers data privacy to contributing nodes in the collaborating environment. By exploiting the Individual datasets of different hospitals in FL setting could be used to develop reliable screening, diagnosis, and treatment predictive models to tackle major challenges such as pandemics. FL can enable the development of very diverse medical imaging datasets and thus provide more reliable models for all participating nodes, including those with low quality data. However, the issue with the traditional Federated Learning paradigm is the degradation of generalization power due to poorly trained local models at the client nodes. The generalization power of the FL paradigm can be improved by considering the relative learning contribution of client nodes. Simple aggregation of learning parameters in the standard FL model faces a diversity issue and results in more validation loss during the learning process. This issue can be resolved by considering the relative contribution of each client node participating in the learning process. The class imbalance at each site is another significant challenge that greatly impacts the performance of the aggregated learning model. This work considers Context Aggregator FL based on the context of loss-factor and class-imbalance issues by incorporating the relative contribution of the collaborating nodes in FL by proposing Validation-Loss based Context Aggregator (CAVL) and Class Imbalance based Context Aggregator (CACI). The proposed Context Aggregator is evaluated on several different Covid-19 imaging classification datasets present on participating nodes. The evaluation results show that Context Aggregator performs better than standard Federating average Learning algorithms and FedProx Algorithm for Covid-19 image classification problems.

Maximizing the potential of high-throughput next-generation sequencing through precise normalization based on read count distribution.

Next-generation sequencing technologies have enabled many advances across diverse areas of biology, with many benefiting from increased sample size. Although the cost of running next-generation sequencing instruments has dropped substantially over time, the cost of sample preparation methods has lagged behind. To counter this, researchers have adapted library miniaturization protocols and large sample pools to maximize the number of samples that can be prepared by a certain amount of reagents and sequenced in a single run. However, due to high variability of sample quality, over and underrepresentation of samples in a sequencing run has become a major issue in high-throughput sequencing. This leads to misinterpretation of results due to increased noise, and additional time and cost rerunning underrepresented samples. To overcome this problem, we present a normalization method that uses shallow iSeq sequencing to accurately inform pooling volumes based on read distribution. This method is superior to the widely used fluorometry methods, which cannot specifically target adapter-ligated molecules that contribute to sequencing output. Our normalization method not only quantifies adapter-ligated molecules but also allows normalization of feature space; for example, we can normalize to reads of interest such as non-ribosomal reads. As a result, this normalization method improves the efficiency of high-throughput next-generation sequencing by reducing noise and producing higher average reads per sample with more even sequencing depth. IMPORTANCE High-throughput next generation sequencing (NGS) has significantly contributed to the field of genomics; however, further improvements can maximize the potential of this important tool. Uneven sequencing of samples in a multiplexed run is a common issue that leads to unexpected extra costs or low-quality data. To mitigate this problem, we introduce a normalization method based on read counts rather than library concentration. This method allows for an even distribution of features of interest across samples, improving the statistical power of data sets and preventing the financial loss associated with resequencing libraries. This method optimizes NGS, which already has huge importance across many areas of biology.

Enhancing earthquake detection from distributed acoustic sensing data by coherency measure and moving-rank-reduction filtering

Distributed acoustic sensing (DAS) enables the recording of earthquake signals at an unprecedented low cost with high durability using fiber optic cables. It is often compromised by relatively lower data quality in single-channel measurements compared with traditional seismic receivers but compensated by high-density recordings from hundreds of channels per earthquake event. The multichannel nature of DAS data sets facilitates the applications of well-developed coherency-based denoising methods arising from reflection seismology for detecting more earthquakes. We first introduce a coherency measure for detecting earthquake signals from DAS data sets. Then, we apply a moving-rank-reduction (MRR) filter to enhance the DAS data quality so as to improve the earthquake detectability. The MRR filter is tailored from a rank-reduction filter that is widely used in processing multichannel reflection seismic data. We find that a simple band-pass or median filter is incapable of revealing weak signals generated from small-magnitude or far-away earthquake events, whereas the MRR filter significantly improves the signal-to-noise ratio that enables the detection of those weak signals. We apply the MRR method and the coherency measure on the San Andreas Fault Observatory at Depth DAS data sets for denoising and earthquake detection. As a result, our framework detects all 31 cataloged events, outperforming the previous detection of 25 events using the same data set.

A general data quality evaluation framework for dynamic response monitoring of long-span bridges

The structural health monitoring system (SHM) of long-span bridges inevitably produces low-quality data. It is important to evaluate the data quality and screen out normal data. Most existing studies on data anomaly detection have focused on abnormal data with abnormal waveforms in time domain. Usually there are pseudo-normal data in the SHM system, which seems normal in time domain. However, the pseudo-normal data of bridge dynamic responses may cause wrong dynamic characteristics identification. In this study, the existing data anomaly detection was further expanded to data quality evaluation, and both obvious abnormal and pseudo-normal data are evaluated. A general data quality evaluation framework for dynamic response monitoring of long-span bridge structures was proposed. The time–frequency characteristics of monitoring data were obtained by continuous wavelet transform, and the data quality was classified and evaluated by constructing and training a convolutional neural network (CNN). The proposed framework was verified by using acceleration monitoring data of a cable-stayed bridge. The results reveal that this framework solves the problem that CNN cannot detect pseudo-normal data through time-domain characteristics, and improves the accuracy of data quality evaluation by using time–frequency information of the monitoring data. The acceleration monitoring data of a suspension bridge and a single pylon cable-stayed bridge were used to demonstrate the feasibility of the cross-object application of the proposed framework. The evaluation accuracy for the acceleration data quality of the suspension bridge exceeded 96%. After the reinforced training of a small number of pseudo-normal data samples, the evaluation accuracy for the acceleration data quality of the cables of the cable-stayed bridge reached 96.8%. The framework shows a good generalization capacity and robustness in cross-object applications.

Impact of household income and Special Supplemental Nutritional Program for Women, Infants, and Children on feeding decisions for infants in the United States

Despite the known benefits of breastfeeding to infants and mothers, previous studies have demonstrated that underserved women are less likely to exclusively breastfeed. Existing studies on the impact of Special Supplemental Nutritional Program for Women, Infants, and Children enrollment on feeding decisions for infants have conflicting results with low-quality data and metrics. This study aimed to examine infant feeding trends nationally in the first week postpartum over a 10-year period, comparing breastfeeding rates for primiparous women with low income who used Special Supplemental Nutritional Program for Women, Infants, and Children resources with those women who did not enroll. We hypothesized that although the Special Supplemental Nutritional Program for Women, Infants, and Children is an important resource for new mothers, free formula associated with enrollment in the Special Supplemental Nutritional Program for Women, Infants, and Children may disincentivize women to exclusively breastfeed. This was a retrospective cohort study of primiparous women with singleton gestations who gave birth at term and who responded to the Centers for Disease Control and Prevention Pregnancy Risk Assessment Monitoring System between 2009 and 2018. Data were extracted from phases 6, 7, and 8 of the survey. Women with low income were defined as those with a reported annual household income of $35,000 or less. The primary outcome was exclusive breastfeeding after 1 week postpartum. Secondary outcomes included ever breastfeeding, any breastfeeding after 1 week postpartum, and introduction of other liquids within 1 week postpartum. Multivariable logistic regression was used to refine risk estimates with adjustment for mode of delivery, household size, education level, insurance status, diabetes, hypertension, race, age, and BMI. Among the 42,778 women with low income who were identified, 29,289 (68%) of these women reported receiving Special Supplemental Nutritional Program for Women, Infants, and Children resources. There was no significant difference in the rates of exclusive breastfeeding after 1 week postpartum between those enrolled in the Special Supplemental Nutritional Program for Women, Infants, and Children and those not enrolled (adjusted risk ratio, 1.04; 95% confidence interval, 1.00-1.07; P=.10). However, those enrolled were less likely to ever breastfeed (adjusted risk ratio, 0.95; 95% confidence interval, 0.94-0.95; P<.01) and were more likely to introduce other liquids within 1 week postpartum (adjusted risk ratio, 1.16; 95% confidence interval, 1.11-1.21; P<.01). Although exclusive breastfeeding rates after 1 week postpartum were similar, women enrolled in the Special Supplemental Nutritional Program for Women, Infants, and Children were significantly less likely to ever breastfeed and more likely to introduce formula within the first week postpartum. This suggests that Special Supplemental Nutritional Program for Women, Infants, and Children enrollment may impact the decision to initiate breastfeeding and may represent an important window to test future interventions.

The role of entrepreneurship, cooperative innovation, environmental investment in relationship between the Belt and Road Initiative and green innovation upgrading

PurposeThis paper analyzes the effect of the Belt and Road Initiative (BRI) on firm green innovation upgrading using data on Chinese firms between 2009 and 2021.Design/methodology/approachThe author adopts the staggered difference-in-difference (DID) method to estimate regressions, treating the proposal of the BRI in 2013 as a policy shock. Our analysis yields few findings.FindingsThe author yields few findings. First, the BRI can significantly promote Chinese firms green innovation upgrading. Specifically, the BRI can promote firm green innovation upgrading by 0.9%. Second, the BRI mainly promotes firms green innovation upgrading by promoting firms to increase green entrepreneurship, cooperative innovation and environmental investment. Finally, the BRI has a greater impact on the green innovation upgrading of firms in the digital industrialization industry rather than digital industry and firms with low pollution emissions rather than firms with high-pollution emissions. This research indicates that the BRI is not only an important platform for sustainable development and also an important opportunity for green entrepreneurship.Research limitations/implicationsFirst, due to the low quality of data and the lack of detailed information on some firms' patents owned after 2018, fully applying data of all years for regression was not possible. Second, the author did not construct a theoretical model to explore the impact of the BRI on green innovation upgrading of firms from the perspective of outward foreign direct investment (OFDI), which is also the direction of future research. Finally, there are still some unexplored mechanisms of the BRI on firms green innovation upgrading, which should be further explored in the future.Originality/valueFirst, from the micro perspective, the author measures the quality of firms' green patents, further measuring the firms' green innovation upgrading. Second, the author discusses the impact of the BRI on firm green innovation upgrading with the method of staggered DID, so that the policy effect of the BRI can be more accurately evaluated. Third, the author comprehensively analyzes the mechanism of cooperative innovation and green infrastructure investment, as well as analyzing the heterogeneity from the perspective of industry digital transformation and firm pollution emissions. Lastly, the author provides specific paths for firms to make high-quality investment from the green BRI construction.

Improve the performance of CT-based pneumonia classification via source data reweighting

Pneumonia is a life-threatening disease. Computer tomography (CT) imaging is broadly used for diagnosing pneumonia. To assist radiologists in accurately and efficiently detecting pneumonia from CT scans, many deep learning methods have been developed. These methods require large amounts of annotated CT scans, which are difficult to obtain due to privacy concerns and high annotation costs. To address this problem, we develop a three-level optimization based method which leverages CT data from a source domain to mitigate the lack of labeled CT scans in a target domain. Our method automatically identifies and downweights low-quality source CT data examples which are noisy or have large domain discrepancy with target data, by minimizing the validation loss of a target model trained on reweighted source data. On a target dataset with 2218 CT scans and a source dataset with 349 CT images, our method achieves an F1 score of 91.8% in detecting pneumonia and an F1 score of 92.4% in detecting other types of pneumonia, which are significantly better than those achieved by state-of-the-art baseline methods.