Search

Gas sensors play a crucial role in various industries and applications. In recent years, there has been an increasing demand for gas sensors in society. However, the current method for screening gas-sensitive materials is time-, energy-, and cost-consuming. Consequently, an imperative exists to enhance the screening efficiency. In this study, we proposed a collaborative screening strategy through integration of density functional theory and machine learning. Taking zinc oxide (ZnO) as an example, the responsiveness of ZnO to the target gas was determined quickly on the basis of the changes in the electronic state and structure before and after gas adsorption. In this work, the adsorption energy and electronic and structural characteristics of ZnO after adsorbing 24 kinds of gases were calculated. These computed features served as the basis for training a machine learning model. Subsequently, various machine learning and evaluation algorithms were utilized to train the fast screening model. The importance of feature values was evaluated by the AdaBoost, Random Forest, and Extra Trees models. Specifically, charge transfer was assigned importance values of 0.160, 0.127, and 0.122, respectively, ranking as the highest among the 11 features. Following closely was the d-band center, which was presumed to exert influence on electrical conductivity and, consequently, adsorption properties. With 5-fold cross-validation using the Extra Tree accuracy, the 24-sample data set achieved an accuracy of 88%. The 72-sample data set achieved an accuracy of 78% using multilayer perceptron after 5-fold cross-validation, with both data sets exhibiting low standard deviations. This verified the accuracy and reliability of the strategy, showcasing its potential for rapidly screening a material's responsiveness to the target gas.

Background: This study aims to develop a machine learning-based application in a real-world medical domain to assist anesthesiologists in assessing the risk of complications in patients after a hip surgery., Methods: Data from adult patients who underwent hip repair surgery at Chi-Mei Medical Center and its 2 branch hospitals from January 1, 2013, to March 31, 2020, were analyzed. Patients with incomplete data were excluded. A total of 22 features were included in the algorithms, including demographics, comorbidities, and major preoperative laboratory data from the database. The primary outcome was a composite of adverse events (in-hospital mortality, acute myocardial infarction, stroke, respiratory, hepatic and renal failure, and sepsis). Secondary outcomes were intensive care unit (ICU) admission and prolonged length of stay (PLOS). The data obtained were imported into 7 machine learning algorithms to predict the risk of adverse outcomes. Seventy percent of the data were randomly selected for training, leaving 30% for testing. The performances of the models were evaluated by the area under the receiver operating characteristic curve (AUROC). The optimal algorithm with the highest AUROC was used to build a web-based application, then integrated into the hospital information system (HIS) for clinical use., Results: Data from 4,448 patients were analyzed; 102 (2.3%), 160 (3.6%), and 401 (9.0%) patients had primary composite adverse outcomes, ICU admission, and PLOS, respectively. Our optimal model had a superior performance (AUROC by DeLong test) than that of ASA-PS in predicting the primary composite outcomes (0.810 vs. 0.629, p < 0.01), ICU admission (0.835 vs. 0.692, p < 0.01), and PLOS (0.832 vs. 0.618, p < 0.01)., Conclusions: The hospital-specific machine learning model outperformed the ASA-PS in risk assessment. This web-based application gained high satisfaction from anesthesiologists after online use., (© 2022. The Author(s).)

Zero-shot learning (ZSL) aims to learn a classifier for unseen classes by exploiting both training data from seen classes and external knowledge. In many visual tasks such as image classification, a set of high-level attributes that describe the semantic properties of classes are used as the external knowledge to bridge seen and unseen classes. While the attributes are usually treated equally by previous ZSL studies, we observe that the contribution of different attributes varies significantly over model training. To adaptively exploit the discriminative information embedded in different attributes, we propose a novel encoder-decoder framework with attention mechanism on the attribute level for zero-shot learning. Specifically, by mapping the visual features into a semantic space, the more discriminative attributes are emphasized with larger attention weights. Further, the attentive attributes and the class prototypes are simultaneously decoded to the visual space so that the hubness problem can be eased. Finally, the labels are predicted in the visual space. Extensive experiments on multiple benchmark datasets demonstrate that our proposed model achieves a significant boost over several state-of-the-art methods for ZSL task and comparative results for generalized ZSL task., Competing Interests: Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2021 Elsevier Ltd. All rights reserved.)

Background: Dengue virus causes a wide spectrum of disease, which ranges from subclinical disease to severe dengue shock syndrome. However, estimating the risk of severe outcomes using clinical presentation or laboratory test results for rapid patient triage remains a challenge. Here, we aimed to develop prognostic models for severe dengue using machine learning, according to demographic information and clinical laboratory data of patients with dengue., Methodology/principal Findings: Out of 1,581 patients in the National Cheng Kung University Hospital with suspected dengue infections and subjected to NS1 antigen, IgM and IgG, and qRT-PCR tests, 798 patients including 138 severe cases were enrolled in the study. The primary target outcome was severe dengue. Machine learning models were trained and tested using the patient dataset that included demographic information and qualitative laboratory test results collected on day 1 when they sought medical advice. To develop prognostic models, we applied various machine learning methods, including logistic regression, random forest, gradient boosting machine, support vector classifier, and artificial neural network, and compared the performance of the methods. The artificial neural network showed the highest average discrimination area under the receiver operating characteristic curve (0.8324 ± 0.0268) and balance accuracy (0.7523 ± 0.0273). According to the model explainer that analyzed the contributions/co-contributions of the different factors, patient age and dengue NS1 antigenemia were the two most important risk factors associated with severe dengue. Additionally, co-existence of anti-dengue IgM and IgG in patients with dengue increased the probability of severe dengue., Conclusions/significance: We developed prognostic models for the prediction of dengue severity in patients, using machine learning. The discriminative ability of the artificial neural network exhibited good performance for severe dengue prognosis. This model could help clinicians obtain a rapid prognosis during dengue outbreaks. However, the model requires further validation using external cohorts in future studies., Competing Interests: The authors have declared that no competing interests exist.

This paper reports on the use of machine learning to delineate data harnessed by fiber-optic distributed acoustic sensors (DAS) using fiber with enhanced Rayleigh backscattering to recognize vibration events induced by human locomotion. The DAS used in this work is based on homodyne phase-sensitive optical time-domain reflectometry (φ-OTDR). The signal-to-noise ratio (SNR) of the DAS was enhanced using femtosecond laser-induced artificial Rayleigh scattering centers in single-mode fiber cores. Both supervised and unsupervised machine-learning algorithms were explored to identify people and specific events that produce acoustic signals. Using convolutional deep neural networks, the supervised machine learning scheme achieved over 76.25% accuracy in recognizing human identities. Conversely, the unsupervised machine learning scheme achieved over 77.65% accuracy in recognizing events and human identities through acoustic signals. Through integrated efforts on both sensor device innovation and machine learning data analytics, this paper shows that the DAS technique can be an effective security technology to detect and to identify highly similar acoustic events with high spatial resolution and high accuracies.

Segmenting bioimage based filaments is a critical step in a wide range of applications, including neuron reconstruction and blood vessel tracing. To achieve an acceptable segmentation performance, most of the existing methods need to annotate amounts of filamentary images in the training stage. Hence, these methods have to face the common challenge that the annotation cost is usually high. To address this problem, we propose an interactive segmentation method to actively select a few super-pixels for annotation, which can alleviate the burden of annotators. Specifically, we first apply a Simple Linear Iterative Clustering (i.e., SLIC) algorithm to segment filamentary images into compact and consistent super-pixels, and then propose a novel batch-mode based active learning method to select the most representative and informative (i.e., BMRI) super-pixels for pixel-level annotation. We then use a bagging strategy to extract several sets of pixels from the annotated super-pixels, and further use them to build different Laplacian Regularized Gaussian Mixture Models (Lap-GMM) for pixel-level segmentation. Finally, we perform the classifier ensemble by combining multiple Lap-GMM models based on a majority voting strategy. We evaluate our method on three public available filamentary image datasets. Experimental results show that, to achieve comparable performance with the existing methods, the proposed algorithm can save 40 percent annotation efforts for experts.

Counterfeit goods are pervasive, being found in products as diverse as textiles and optical media to pharmaceuticals and sensitive electronics. Here, an anti‐counterfeit platform is reported in which plasmonic nanoparticles (NPs) are used to create unique image tags that can be authenticated quickly and reliably. Specifically, plasmonic NPs are assembled into periodic arrays of NP clusters by template‐assisted self‐assembly (TASA), where the light scattering responses from the arrays are analyzed by dark‐field optical microscopy. Tag design proved modular as plasmonic NPs with different optical responses can be selected and paired with Templates with different features (e.g., well size, well shape, and number and arrangement of wells in an array), giving access to a variety of color responses and unique images. These images can be differentiated from one another and authenticated by image analysis. Authentication methods based on shallow and deep neural networks are compared, where deep neural networks authenticated TASA tags with higher accuracy. Given the ease of tag fabrication and rapid image analysis, these platforms are ideal for on‐the‐fly tagging and supply‐chain authentication of critical goods. [ABSTRACT FROM AUTHOR]

River algal blooms have become a global environmental problem due to their large impact range and environmental hazards. However, the complex mechanisms underlying these blooms make prediction and prevention challenging. Here, we employed empirical dynamic modeling (EDM) and machine learning to reveal the causes and predict diatom blooms from 2003 to 2017 in the Han River of China. The diatom cell density ranged from 0.1 to 5.1 × 107 cells L−1, whereas algal blooms often lasted for 10 days with density exceeding 107 cells L−1. The EDM results elucidated that, under consistent high nutrient concentrations, algal blooms were primarily regulated by eight environmental factors: water temperature in the Han River; water levels, flow velocities, and streamflow discharges in the Han River and the Yangtze River; and water level variation in the Han River. The poor performance (coefficient of determination R2 < 0) of the multiple linear regression, EDM, and random forest models indicated the challenge of predicting algae density. Therefore, we used machine learning classification models to predict algal blooms occurrences. With the resampling techniques to account for imbalanced data, machine learning models achieved perfect classification prediction (Kappa value = 1) of the 13 algae‐bloom events with a 10‐day lead time during a 15‐year period, providing an important reference for preemptive warning of riverine algal blooms. Plain Language Summary: River algal blooms have become a global environmental problem due to their large impact range and environmental hazards. The Han River is one of the most frequent areas of river algal blooms in China, but there is no clear conclusion on the causes of algal blooms formation. In this study, we combined two methods to reveal the causes and predict the occurrence of algal blooms in the Han River. Our results show that the prediction of algal blooms in the Han River is feasible. These blooms are primarily affected by the water temperature and seven other hydrological variables in the Han and Yangtze River. This study provides a starting point for the preemptive warning of riverine algal blooms. Key Points: Empirical dynamic modeling detects causality for river algal bloomsMachine learning accurately predicts the occurrence of algal blooms in the Han RiverResampling techniques improve robustness and accuracy of machine learning [ABSTRACT FROM AUTHOR]

Recently, denoising diffusion probabilistic models (DDPMs) have been effective in speech enhancement. However, existing models largely follow the original diffusion training method, ignoring the trade-off between optimization goals in different training phases. This is not only responsible for the slow training of diffusion models, but also limits their performance. And most current research based on diffusion models predicts Gaussian noise added during the training process or clean data, without considering their relationship to the model. Moreover, the reverse process still does not fully exploit the predictive power of the model. In this work, we integrate the Min-SNR weighting strategy into the training process of the existing probabilistic conditional diffusion model to efficiently exploit different time steps. Based on this strategy, we propose a joint training algorithm and objective to use intermediate data generated during the training phase to assist model learning. Furthermore, we adopt a deterministic reverse process, which is quite different from the existing one, to speed up the inference and improve the quality of the enhanced speech. Experiments show that our proposed method has faster convergence and better performance than related diffusion models and other generative models. [ABSTRACT FROM AUTHOR]

The electromagnetic (EM) detection and sensing research of human signs are becoming much more attractive, especially with the rise of intelligent metasurface which can strongly enhance the control abilities and functionalities of transmitting and receiving EM fields. It proposes the design of integrating‐type intelligent metasurface to realize the indoor positioning and posture recognition of the human body using the support vector machine algorithm and a new deep learning algorithm model it proposed called MetaNet. The integrating‐type metasurface with the in‐plane feed rather than horn excitation is real low profile, and the varactor based programmable phase shifters achieving 360‐degree cover and continuous control of phase are independently embedded in each unit cell for further synthesized manipulation of apertures. One pair of integrating‐type intelligent metasurfaces are fabricated for transmitting and receiving EM fields, respectively. Also, the corresponding prototype of radar transceiver system is developed and manufactured to rapidly acquire massive data. The total experimental flow can include two main steps that first, the transmitting and receiving beams of metasurfaces should scan dynamically and synchronously under the method of far‐field synthesis for detecting the direction of the human. Second, the transmitting beam should keep the detected direction and the receiving apertures of metasurface should be varied by randomly control the phase shift of each unit cell for multiple sensing. Based on the technology of deep learning to process the massive receiving data, the profile images and key points of various human postures can be reconstructed very well in the measurement that can verify the validity of the proposed sensing system. [ABSTRACT FROM AUTHOR]

Machine learning (ML) has been applied in civil engineering to predict the compressive strength of concrete with high accuracy. In this paper, five boosting ensemble algorithms, i.e., XGBoost, AdaBoost, GBDT, LightGBM, and CatBoost, were used to predict the compressive strength of high-performance concrete (HPC). The models were evaluated using performance indicators such as R2, root mean square error (RMSE), and mean absolute error (MAE). The results showed that the CatBoost model had the highest accuracy with a R2 (0.970) and a RMSE (2.916). The prediction accuracy of the model was increased through hyperparameter optimization, which got a higher with a R2 (0.975) and a RMSE (2.863). Meanwhile, the SHapley Additive exPlanations (SHAP) method was used to explain the output results of the optimal model (CatBoost), which generated explainable insights that further revealed the complex relationship between the prediction model parameters. The results showed that AGE, W/B, and W/C had the most impact on high-performance concrete compressive strength (HPCCS) prediction, which was similar to the results of sensitivity analysis. This study provided a theoretical basis and technical guidance for developing the mix design of a new high-performance concrete (HPC) system. In the future, the interpretable results of the model output should be iteratively checked and validated in the actual laboratory in order to provide guidance for engineering practice. [ABSTRACT FROM AUTHOR]

Whistler‐mode chorus waves play an essential role in the acceleration and loss of energetic electrons in the Earth's inner magnetosphere, with the more intense waves producing the most dramatic effects. However, it is challenging to predict the amplitude of strong chorus waves due to the imbalanced nature of the data set, that is, there are many more non‐chorus data points than strong chorus waves. Thus, traditional models usually underestimate chorus wave amplitudes significantly during active times. Using an imbalanced regressive (IR) method, we develop a neural network model of lower‐band (LB) chorus waves using 7‐year observations from the EMFISIS instrument onboard Van Allen Probes. The feature selection process suggests that the auroral electrojet index alone captures most of the variations of chorus waves. The large amplitude of strong chorus waves can be predicted for the first time. Furthermore, our model shows that the equatorial LB chorus's spatiotemporal evolution is similar to the drift path of substorm‐injected electrons. We also show that the chorus waves have a peak amplitude at the equator in the source MLT near midnight, but toward noon, there is a local minimum in amplitude at the equator with two off‐equator amplitude peaks in both hemispheres, likely caused by the bifurcated drift paths of substorm injections on the dayside. The IR‐based chorus model will improve radiation belt prediction by providing chorus wave distributions, especially storm‐time strong chorus. Since data imbalance is ubiquitous and inherent in space physics and other physical systems, imbalanced regressive methods deserve more attention in space physics. Plain Language Summary: Whistler‐mode chorus waves are essential in accelerating radiation belt electrons. However, predicting the amplitude of strong chorus waves is difficult because of their imbalanced nature. In other words, there are many more observations of no‐chorus waves than strong chorus waves. A consequence is that these no‐chorus wave data dominate traditional models, so these models usually predict values that are too small for strong waves. Using an imbalanced regressive method, we developed a machine learning (ML) model of lower‐band chorus wave amplitude. For the first time, the ML‐chorus model can predict the amplitude of strong chorus waves. The ML‐chorus model shows the evolution of the chorus wave at the equator, similar to the drift path of injected electrons, which brings electron anisotropy that generates chorus waves. The ML‐chorus model shows that the chorus waves are stronger at the equator near midnight, the source region of plasma injection. Away from midnight, the chorus waves have an equatorial minimum instead. Our chorus model will improve the forecast of the radiation belt environment by providing chorus wave distributions, especially large‐amplitude strong chorus during geomagnetic storms. Because data imbalance is commonly seen in space physics and other physical systems, imbalanced regressive methods require more attention. Key Points: A neural network model of lower‐band chorus wave amplitude is developed using imbalanced regressionFor the first time, a chorus model can predict the large amplitude of strong chorus wavesChorus' evolution is consistent with electron injection drift paths, peak equatorial amplitude near midnight, and off‐equator peaks at noon [ABSTRACT FROM AUTHOR]

Background: The prevalence of violence in acute psychiatric wards is a critical concern. According to a meta-analysis investigating violence in psychiatric inpatient units, researchers estimated that approximately 17% of inpatients commit one or more acts of violence during their stay. Inpatient violence negatively affects health-care providers and patients and may contribute to high staff turnover. Therefore, predicting which psychiatric inpatients will commit violence is of considerable clinical significance. Objective: The present study aimed to estimate the violence rate for psychiatric inpatients and establish a predictive model for violence in psychiatric inpatients. Methods: We collected the structured and unstructured data from Chinese nursing electronic medical records (EMRs) for the violence prediction. The data was obtained from the psychiatry department of a regional hospital in southern Taiwan, covering the period between January 2008 and December 2018. Several text mining and machine learning techniques were employed to analyze the data. Results: The results demonstrated that the rate of violence in psychiatric inpatients is 19.7%. The patients with violence in psychiatric wards were generally younger, had a more violent history, and were more likely to be unmarried. Furthermore, our study supported the feasibility of predicting aggressive incidents in psychiatric wards by using nursing EMRs and the proposed method can be incorporated into routine clinical practice to enable early prediction of inpatient violence. Conclusions: Our findings may provide clinicians with a new basis for judgment of the risk of violence in psychiatric wards. [ABSTRACT FROM AUTHOR]

Traditional supervised learning requires ground truth labels for training, whose collection however is difficult in many cases. Recently, crowdsourcing has established itself as an efficient labeling solution by resorting to non-expert crowds. To reduce the labeling error effects, one common practice is to distribute each instance to multiple workers, whereas each worker only annotates a subset of data, resulting in the sparse annotation phenomenon. In this paper, we show that when meeting with class-imbalance, i.e., even when the groundtruth labels are slightly imbalanced, the sparse annotations are prone to be skewly distributed and would bias the learning algorithm severely. To combat this issue, we propose one Distribution Aware Self-training based Crowdsourcing learning (DASC) approach, which supplements the sparse annotations by adding confident pseudo-annotations and at the same time re-balancing the annotation distribution. Specifically, we propose one distribution aware confidence measure to select the most confident pseudo-annotations, with minority/majority classes selected more/less frequently. As a universal framework, DASC is applicable to various crowdsourcing methods for consistent performance gains. We conduct extensive experiments over real-world crowdsourcing benchmarks, from slight to heavy imbalance ratio, with various annotation sparsity levels, and show that DASC substantially improves previous crowdsourcing models by 2 % - 20 % absolute test accuracy, and yields much more balanced annotations. [ABSTRACT FROM AUTHOR]

In order to improve the accuracy of gesture recognition based on surface electromyography (sEMG), an improved deep forest combined hand motion recognition method was proposed. The extreme gradient boosting (XGBoost) tree was introduced into the deep forest model to form the cascade structure of deep forest together with the random forest and the complete random forest. The deep forest model integrates three different tree-based classifiers at each level, a total of four decision forests including a random forest, an extreme random forest and two extreme gradient boosting trees. The classification performance was improved by using the complementarity between different learning algorithms. In order to evaluate the performance of the model, the sEMG signals of 4 healthy subjects were collected for the verification experiment of hand action recognition, and compared with random forest, support vector machine, one-dimensional and two-dimensional convolutional neural networks algorithms. The result shows that the average recognition accuracy of the method for 16 commonly used hand actions is 94.14%, and the classification accuracy of sEMG signals is high. [ABSTRACT FROM AUTHOR]

Protein–protein interactions play an important role in many biological processes. However, although structure prediction for monomer proteins has achieved great progress with the advent of advanced deep learning algorithms like AlphaFold, the structure prediction for protein–protein complexes remains an open question. Taking advantage of the Transformer model of ESM-MSA, we have developed a deep learning-based model, named DeepHomo2.0, to predict protein–protein interactions of homodimeric complexes by leveraging the direct-coupling analysis (DCA) and Transformer features of sequences and the structure features of monomers. DeepHomo2.0 was extensively evaluated on diverse test sets and compared with eight state-of-the-art methods including protein language model-based, DCA-based and machine learning-based methods. It was shown that DeepHomo2.0 achieved a high precision of >70% with experimental monomer structures and >60% with predicted monomer structures for the top 10 predicted contacts on the test sets and outperformed the other eight methods. Moreover, even the version without using structure information, named DeepHomoSeq, still achieved a good precision of >55% for the top 10 predicted contacts. Integrating the predicted contacts into protein docking significantly improved the structure prediction of realistic Critical Assessment of Protein Structure Prediction homodimeric complexes. DeepHomo2.0 and DeepHomoSeq are available at http://huanglab.phys.hust.edu.cn/DeepHomo2/. [ABSTRACT FROM AUTHOR]

Rivers play an important role in water supply, irrigation, navigation, and ecological maintenance. Forecasting the river hydrodynamic changes is critical for flood management under climate change and intensified human activities. However, efficient and accurate river modeling is challenging, especially with complex lake boundary conditions and uncontrolled downstream boundary conditions. Here, we proposed a coupled framework by taking the advantages of interpretability of physical hydrodynamic modeling and the adaptability of machine learning. Specifically, we coupled the Gated Recurrent Unit (GRU) with a 1‐D HydroDynamic model (GRU‐HD) and applied it to the middle and lower reaches of the Yangtze River, the longest river in China. We show that the GRU‐HD model could quickly and accurately simulate the water levels, streamflow, and water exchange rates between the Yangtze River and two important lakes (Poyang and Dongting), with most of the Kling‐Gupta efficiency coefficient (KGE $\mathrm{K}\mathrm{G}\mathrm{E}$) above 0.90. Using machine learning‐based predicted water levels, instead of the rating curve approach, as the downstream boundary conditions could improve the accuracy of modeling the downstream water levels of the lake‐connected river system. The GRU‐HD model is dedicated to the synergy of physical modeling and machine learning, providing a powerful avenue for modeling rivers with complex boundary conditions. Key Points: Coupling machine learning into physical hydrodynamics for river modeling with complex boundary conditionsStreamflow and water levels were modeled with Kling‐Gupta efficiency coefficient above 0.90 at most hydrologic stations and 38 times faster than traditional 1‐D/2‐D coupled modelsThe proposed machine learning‐based downstream boundary condition showed better performance than the rating curve method [ABSTRACT FROM AUTHOR]

• Coupling water quality models with machine learning to model daily pollution loads. • The relative contributions of natural and anthropogenic factors were quantified. • Forces driving the anthropogenic pollution sources into the river were deciphered. Pollution control and environmental protection of the Yangtze River have received major attention in China. However, modeling the river's pollution load remains challenging due to limited monitoring and unclear spatiotemporal distribution of pollution sources. Specifically, anthropogenic activities' contribution to the pollution have been underestimated in previous research. Here, we coupled a hydrodynamic-based water quality (HWQ) model with a machine learning (ML) model, namely attention-based Gated Recurrent Unit, to decipher the daily pollution loads (i.e., chemical oxygen demand, COD; total phosphorus, TP) and their sources in the Middle-Lower Yangtze River from 2014 to 2018. The coupled HWQ-ML model outperformed the standalone ML model with KGE values ranging 0.77–0.91 for COD and 0.47–0.64 for TP, while also reducing parameter uncertainty. When examining the relative contributions at the Middle Yangtze River Hankou cross-section, we observed that the main stream and tributaries, lateral anthropogenic discharges, and parameter uncertainty contributed 15, 66, and 19% to COD, and 58, 35, and 7% to TP, respectively. For the Lower Yangtze River Datong cross-section, the contributions were 6, 69, and 25% for COD and 41, 42, and 17% for TP. According to the attention weights of the coupled model, the primary drivers of lateral anthropogenic pollution sources, in descending order of importance, were temperature, date, and precipitation, reflecting seasonal pollution discharge, industrial effluent, and first flush effect and combined sewer overflows, respectively. This study emphasizes the synergy between physical modeling and machine learning, offering new insights into pollution load dynamics in the Yangtze River. [Display omitted] [ABSTRACT FROM AUTHOR]

This paper proposes an Adaptive Fuzzy Neural Agent (AFNA) with a Patch Learning Mechanism and IEEE 1855 Fuzzy Markup Language (FML) for human and machine co-learning. There are three phases of patch learning mechanism embedded in AFNA, including (1) training an initial global model, (2) training a patch model for each identified patch, and (3) updating the global model using the training data that do not fall into any patch. The AFNA can be applied to construct the student and robot co-learning regression model, as well as the regression model for the dataset retrieved from the game of Go. First, students generate human learning data through interactions with handheld devices or robots based on the AFNA in Taiwan and Japan. Then, the AFNA utilizes the student learning data collected in the classroom and the Go game data provided by both Google DeepMind and Facebook AI Research open-source OpenGo to train the Fuzzy Machine-Learning Model. In addition, the trained Fuzzy Machine-Learning Model of AFNA is deployed to the robots to make students and machines co-learn together based on IEEE 1855 FML. The experiments show that the AFNA with Patch Learning Mechanism and Fuzzy Machine-Learning Model can improve the performance of regression model based on the datasets of student learning and Go game. In the future, we hope to apply the AFNA with robots to the other domain areas, embed it with the Artificial Intelligence of Things devices, and introduce it to more teaching fields in various countries. [ABSTRACT FROM AUTHOR]

Dengue virus, a positive-sense single-stranded RNA virus, continuously threatens human health. Although several criteria for evaluation of severe dengue have been recently established, the ability to prognose the risk of severe outcomes for dengue patients remains limited. Mutant spectra of RNA viruses, including single nucleotide variants (SNVs) and defective virus genomes (DVGs), contribute to viral virulence and growth. Here, we determine the potency of intrahost viral population in dengue patients with primary infection that progresses into severe dengue. A total of 65 dengue virus serotype 2 infected patients in primary infection including 17 severe cases were enrolled. We utilized deep sequencing to directly define the frequency of SNVs and detection times of DVGs in sera of dengue patients and analyzed their associations with severe dengue. Among the detected SNVs and DVGs, the frequencies of 9 SNVs and the detection time of 1 DVG exhibited statistically significant differences between patients with dengue fever and those with severe dengue. By utilizing the detected frequencies/times of the selected SNVs/DVG as features, the machine learning model showed high average with a value of area under the receiver operating characteristic curve (AUROC, 0.966 ± 0.064). The elevation of the frequency of SNVs at E (nucleotide position 995 and 2216), NS2A (nucleotide position 4105), NS3 (nucleotide position 4536, 4606), and NS5 protein (nucleotide position 7643 and 10067) and the detection times of the selected DVG that had a deletion junction in the E protein region (nucleotide positions of the junction: between 969 and 1022) increased the possibility of dengue patients for severe dengue. In summary, we demonstrated the detected frequencies/times of SNVs/DVG in dengue patients associated with severe disease and successfully utilized them to discriminate severe patients using machine learning algorithm. The identified SNVs and DVGs that are associated with severe dengue will expand our understanding of intrahost viral population in dengue pathogenesis. [ABSTRACT FROM AUTHOR]

Pavement distress classification is a vital step for automatic pavement inspection and maintenance. Recently, patch‐based approaches have achieved promising performances and thus extensive attention in this field. However, these methods simply assume that all patches contribute equally to the distress classification, leading to weakly discriminating abilities of models. Moreover, their tedious processes also lead to low efficiency in inference. The authors present a novel patch‐based pavement distress classification approach named Deep Patch Soft Selective Learning, which addresses these issues. Similar to other patch‐based approaches, Deep Patch Soft Selective Learning partitions the pavement images into patches and aggregates the patch features to accomplish the task. To address the first issue, a succinct Soft Patch Feature Selection Network is introduced to assess the importance of each patch to the distress classification with a score based on its feature. These scores will be considered as patch‐wise weights for feature aggregation. In such a manner, the most discriminative patches are selected in a soft way, and thereby benefit the final classification. To address the inference efficiency issue, knowledge distillation is leveraged to transfer the classification knowledge from Deep Patch Soft Selective Learning to the image‐based approaches, such as EfficientNet‐B3. This distilled model enables incorporating both the advantages of patch‐based approaches in classification performance and the advantages of image‐based approaches in inference efficiency. Extensive experiments on a large‐scale pavement image dataset named CQU‐BPDD demonstrate that Deep Patch Soft Selective Learning outperforms the baselines by a large margin of +7.4% in P@R on the detection task and +3.1% in F1 on the recognition task while enjoying 2.1x throughput of baselines. [ABSTRACT FROM AUTHOR]

In this study, eggplant seeds of fifteen different varieties were selected for discriminant analyses with a multispectral imaging technique. Seventy-eight features acquired with the multispectral images were extracted from individual eggplant seeds, which were then classified using SVM and a one-dimensional convolutional neural network (1D-CNN), and the overall accuracy was 90.12% and 94.80%, respectively. A two-dimensional convolutional neural network (2D-CNN) was also adopted for discrimination of seed varieties, and an accuracy of 90.67% was achieved. This study not only demonstrated that multispectral imaging combining machine learning techniques could be used as a high-throughput and nondestructive tool to discriminate seed varieties but also revealed that the shape of the seed shell may not be exactly the same as the female parents due to the genetic and environmental factors. [ABSTRACT FROM AUTHOR]

Antimicrobial peptides (AMPs) are considered as potential substitutes of antibiotics in the field of new anti-infective drug design. There have been several machine learning algorithms and web servers in identifying AMPs and their functional activities. However, there is still room for improvement in prediction algorithms and feature extraction methods. The reduced amino acid (RAA) alphabet effectively solved the problems of simplifying protein complexity and recognizing the structure conservative region. This article goes into details about evaluating the performances of more than 5,000 amino acid reduced descriptors generated from 74 types of amino acid reduced alphabet in the first stage and the second stage to construct an excellent two-stage classifier, Identification of Antimicrobial Peptides by Reduced Amino Acid Cluster (iAMP-RAAC), for identifying AMPs and their functional activities, respectively. The results show that the first stage AMP classifier is able to achieve the accuracy of 97.21 and 97.11% for the training data set and independent test dataset. In the second stage, our classifier still shows good performance. At least three of the four metrics, sensitivity (SN), specificity (SP), accuracy (ACC), and Matthews correlation coefficient (MCC), exceed the calculation results in the literature. Further, the ANOVA with incremental feature selection (IFS) is used for feature selection to further improve prediction performance. The prediction performance is further improved after the feature selection of each stage. At last, a user-friendly web server, iAMP-RAAC, is established at http://bioinfor.imu.edu.cn/iampraac. [ABSTRACT FROM AUTHOR]

Label distribution learning (LDL) is proposed for solving the label ambiguity problem in recent years, which can be seen as an extension of multi-label learning. To improve the performance of label distribution learning, some existing algorithms exploit label correlations in a global manner that assumes the label correlations are shared by all instances. However, the instances in different groups may share different label correlations, and few label correlations are globally applicable in real-world tasks. In this paper, two novel label distribution learning algorithms are proposed by exploiting label correlations on local samples, which are called GD-LDL-SCL and Adam-LDL-SCL, respectively. To utilize the label correlations on local samples, the influence of local samples is encoded, and a local correlation vector is designed as the additional features for each instance, which is based on the different clustered local samples. Then, the label distribution for an unseen instance can be predicted by exploiting the original features and the additional features simultaneously. Extensive experiments on some real-world data sets validate that our proposed methods can address the label distribution problems effectively and outperform state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Reliable prediction model of CO 2 corrosion is essential for the oil-well cement sheath integrity. This study constructed a complete database containing 301 records of published literature for machine learning modeling. By comparing four standard algorithms (artificial neural networks (ANN), radial product function neural network (RBF), random forest (RF), and support vector machine (SVM)), an effective method is proposed to predict the carbonation depth of oil-well cement under different conditions. The results showed that the coefficient of determination was 0.9732, the root mean square error was 0.6973, the mean absolute error was 0.4231, and the mean square error was 0.0765, indicating that the SVM model has excellent prediction ability. In addition, the RF model characteristic importance analysis reveals that corrosion time, corrosion-resistant additives, and partial pressure of CO 2 are the most important factors for the corrosion depth. This work provides a reliable prediction method for the CO 2 corrosion rate of oil-well cement. [Display omitted] • A Machine Learning (ML) algorithm to interpret corrosion depth is presented. • Results of four machine learning models, ANN, RF, RBF, and SVM, were compared. • SHAP visualizes the process in which all features affect the output results. • SVM achieved high prediction ability in contrast to ANN, RF and RBF. • Corrosion-resistant additive, time and CO 2 pressure are the most important factors. [ABSTRACT FROM AUTHOR]

In multi-label learning, it is rather expensive to label instances since they are simultaneously associated with multiple labels. Therefore, active learning, which reduces the labeling cost by actively querying the labels of the most valuable data, becomes particularly important for multi-label learning. A good multi-label active learning algorithm usually consists of two crucial elements: a reasonable criterion to evaluate the gain of querying the label for an instance, and an effective classification model, based on whose prediction the criterion can be accurately computed. In this paper, we first introduce an effective multi-label classification model by combining label ranking with threshold learning, which is incrementally trained to avoid retraining from scratch after every query. Based on this model, we then propose to exploit both uncertainty and diversity in the instance space as well as the label space, and actively query the instance-label pairs which can improve the classification model most. Extensive experiments on 20 datasets demonstrate the superiority of the proposed approach to state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Traditionally, the classification of seed defects mainly relies on the characteristics of color, shape, and texture. This method requires repeated extraction of a large amount of feature information, which is not efficiently used in detection. In recent years, deep learning has performed well in the field of image recognition. We introduced convolutional neural networks (CNNs) and transfer learning into the quality classification of seeds and compared them with traditional machine learning algorithms. Experiments showed that deep learning algorithm was significantly better than the machine learning algorithm with an accuracy of 95% (GoogLeNet) vs. 79.2% (SURF+SVM). We used three classifiers in GoogLeNet to demonstrate that network accuracy increases as the depth of the network increases. We used the visualization technology to obtain the feature map of each layer of the network in CNNs and used the heat map to represent the probability distribution of the inference results. As an end-to-end network, CNNs can be easily applied for automated seed manufacturing. [ABSTRACT FROM AUTHOR]

Traditional Chinese Medicine (TCM) is an experiential form of medicine with a history dating back thousands of years. The present study aimed to utilize neural network analysis to examine specific prescriptions for colorectal cancer (CRC) in clinical practice to arrive at the most effective prescription strategy. The study analyzed the data of 261 CRC cases recruited from a total of 141,962 cases of renowned veteran TCM doctors collected from datasets of both the DeepMedic software and TCM cancer treatment books. The DeepMedic software was applied to normalize the symptoms/signs and Chinese herbal medicine (CHM) prescriptions using standardized terminologies. Over 20 percent of CRC patients demonstrated symptoms of poor appetite, fatigue, loose stool, and abdominal pain. By analyzing the prescription patterns of CHM, we found that Atractylodes macrocephala (Bai-zhu) and Poria (Fu-ling) were the most commonly prescribed single herbs identified through analysis of medical records, and supported by the neural network analysis; although there was a slight difference in the sequential order. The study revealed an 81.9% degree of similarity of CHM prescriptions between the medical records and the neural network suggestions. The patterns of nourishing Qi and eliminating dampness were the most common goals of clinical prescriptions, which corresponds with treatments of CRC patients in clinical practice. This is the first study to employ machine learning, specifically neural network analytics to support TCM clinical diagnoses and prescriptions. The DeepMedic software may be used to deliver accurate TCM diagnoses and suggest prescriptions to treat CRC. [ABSTRACT FROM AUTHOR]

• Anchor-based discriminative dual distribution calibrated feature generative network for TZSL. • Use real sample anchors to calibrate the distribution of generated and real samples. • Distribution calibrations in visual and semantic spaces improve discriminating powers. • Online classifier benefits the abilities of the classifier and the generator. Zero-shot learning (ZSL) is machine learning task to recognize samples from classes that are not observed during training. Transductive ZSL (TZSL) is a more realistic and effective paradigm that leverages unlabeled unseen data during training to reduce the bias towards seen classes. However, most existing TZSL methods neglect the information gap between visual and semantic spaces, and thus fail to generate distribution-consistent unseen features. To address this issue, we propose a novel TZSL approach named Anchor-based Discriminative Dual Distribution Calibrated Feature Generative Network (AD3C-FGN), which performs anchor-based distribution calibration in both visual and semantic spaces for improving generalization ability of the model. In AD3C-FGN, we adopt conditional generative adversarial network with an unseen discriminator to construct a Y-shape generation model that mitigates the domain shift problem. Moreover, an AD3C module is elaborated for calibrating the distribution of generated and real samples in both visual and semantic spaces with real sample anchors, and also enhance the discriminability of the generated samples. AD3C enforces the generated sample to be closer to its homogenous anchor but farther away from inhomogeneous anchors in both spaces. Extensive experimental results on six popular ZSL benchmarks demonstrate that our method achieves promising performances in different settings. The source codes of our model have been released on https://github.com/ZYi-CQU/AD3C-FGN. [ABSTRACT FROM AUTHOR]

An integrated cellular-resolution optical coherence tomography (OCT) module with near-infrared Raman spectroscopy was developed on the discrimination of various skin cancer cells and normal cells. Micron-level three-dimensional (3D) spatial resolution and the spectroscopic capability on chemical component determination can be obtained simultaneously. We experimentally verified the effectiveness of morphology, intensity, and spectroscopy features for discriminating skin cells. Both spatial and spectroscopic features were employed for the discrimination of five types of skin cells, including keratinocytes (HaCaT), the cell line of squamous cell carcinoma (A431), the cell line of basal cell carcinoma (BCC-1/KMC), primary melanocytes, and the cell line of melanoma (A375). The cell volume, compactness, surface roughness, average intensity, and internal intensity standard deviation were extracted from the 3D OCT images. After removing the fluorescence components from the acquired Raman spectra, the entire spectra (600 to 2100 cm − 1) were used. An accuracy of 85% in classifying five types of skin cells was achieved. The cellular-resolution OCT images effectively differentiate cancer and normal cells, whereas Raman spectroscopy can distinguish the cancer cells with nearly 100% accuracy. Among the OCT image features, cell surface roughness, internal average intensity, and standard deviation of internal intensity distribution effectively differentiate the cancerous and normal cells. The three features also worked well in sorting the keratinocyte and melanocyte. Using the full Raman spectra, the melanoma and keratinocyte-based cell carcinoma cancer cells can be discriminated effectively. [ABSTRACT FROM AUTHOR]

• Firstly apply artificial intelligence (AI) methods to the research on pipe jacking. • An intelligent framework for predicting the jacking force is proposed. • The predictive performance of the PSO-BP and PSO-SVR models is compared. • The theoretical bases of algorithms may lead to the performance differences. • The generalization ability of the two ML-based models was proved with a new dataset. Jacking force prediction is essential for pipe jacking construction, especially when facing randomness, uncertainty, and nonlinear conditions, which is a challenge for most analysis methods. To address this issue, machine learning (ML), a data analysis method that has been successfully applied in other construction fields such as TBM thrust force forecast, was introduced in this study. Specifically, the PSO-BPNN and PSO-SVR models were developed for jacking force prediction, involving raw data preprocessing, sensitivity analysis, hyper-parameter selection, and prediction accuracy evaluation. To validate and compare the two ML-based models, a curved pipe roof project was employed, which included three geometric, five geological, and eight operational parameters, totaling 4398 data points. The results show that both models exhibit good performance in predicting jacking force, and the PSO-SVR model outperforms the PSO-BPNN model; on the other hand, both models demonstrate better prediction accuracy in the frontal resistance, grouting pressure, and rotational speed of the cutterhead. Additionally, the generalization ability of the jacking force prediction models was examined using a new dataset, and the PSO-SVR model exhibited superior performance over the PSO-BPNN model. The study highlights the unique features of the two models in their application, and the advantages of the ML models provide valuable insights for jacking force prediction in pipe jacking construction. Thus, this study serves as a reference for future research in this area. [ABSTRACT FROM AUTHOR]

Hypergraph is a powerful representation for several computer vision, machine learning, and pattern recognition problems. In the last decade, many researchers have been keen to develop different hypergraph models. In contrast, no much attention has been paid to the design of hyperedge weighting schemes. However, many studies on pairwise graphs showed that the choice of edge weight can significantly influence the performances of such graph algorithms. We argue that this also applies to hypergraphs. In this paper, we empirically study the influence of hyperedge weights on hypergraph learning via proposing three novel hyperedge weighting schemes from the perspectives of geometry, multivariate statistical analysis, and linear regression. Extensive experiments on ORL, COIL20, JAFFE, Sheffield, Scene15 and Caltech256 datasets verified our hypothesis for both classification and clustering problems. For each of these classes of problems, our empirical study concludes with suggesting a suitable hypergraph weighting scheme. Moreover, the experiments also demonstrate that the combinations of such weighting schemes and conventional hypergraph models can achieve competitive classification and clustering performances in comparison with some recent state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Collaborative representation (CR), known as an effective way to address the signal representation (regression) problem, has achieved remarkable success in visual classification. According to our theoretical analysis, the subspace learning issue can also be deemed as a signal representation problem. Therefore, we extend the graph embedding (GE) framework as a CR model to improve the discriminating power of the subspace learning algorithm. The new GE framework, which is named collaborative GE (CGE) framework, enjoys many desirable properties of CR. From theoretical analysis, CGE is robust to the noise and has the same computational complexity as GE. From experimental analysis, CGE can generally enhance the subspace learning algorithms and a reasonable regularization parameter can be inferred from its intrinsic graph. Several state-of-the-art subspace learning algorithms are plugged into our framework to produce their collaborative versions. Meanwhile, by exploring the intrinsic relation among GE methods, we present a new collaborative method named collaborative class-scattering locality preserving projections (CCSLPPs). The results of extensive experiments on ORL, AR, Scene15, Caltech256, LFW-A, and OU-ISIR-A databases demonstrate that the collaborative versions consistently outperform their original algorithms with a remarkable improvement and CCSLPP gets the best performance compared with all used methods. [ABSTRACT FROM AUTHOR]

Discriminant Locality Preserving Projections (DLPP) is one of the most influential supervised subspace learning algorithms that consider both discriminative and geometric (manifold) information. There is an obvious drawback of DLPP that it only considers the pairwise geometric relationship of samples. However, in many real-world issues, relationships among the samples are often more complex than pairwise. Naively squeezing the complex into pairwise ones will inevitably lead to loss of some information, which are crucial for classification and clustering. We address this issue via using the Hyper-Laplacian instead of the regular Laplacian in DLPP, which only can depict the pairwise relationship. This new DLPP algorithm is exactly a generalization of DLPP and we name it Discriminant Hyper-Laplacian Projection (DHLP). In order to make DHLP can be feasibly applied to big data dimensionality reduction, we adopt the spectral regression framework to reduce the computational complexity of DHLP from cubic-time to linear-time. We named this new DHLP algorithm Scalable Discriminant Hyper-Laplacian Projections (SDHLP). Six popular visual databases are adopted for validating our work. The results not only demonstrate the superiorities of DHLP and SDHLP in comparison with the state-of-the-art algorithms but also demonstrate the efficiency improvement of SDHLP over DHLP. [ABSTRACT FROM AUTHOR]

Motivated by the fact that the signals tend to have a representation biased towards their own classes, we propose a novel Sparse Representation-based Classifier (SRC) named Class Specific Sparse Representation-based Classifier (CSSRC), which incorporates the class information in the representation learning. Unlike the conventional SRC algorithms, CSSRC defines each class as a group and then impels these groups to compete for representing the test sample. To achieve such property, CSSRC imposes a L 1 -norm constraint to the classes for compulsively selecting the most relevant classes and introduces a L 2 -norm constraint to the samples belonging to the same class for making sure that all homogeneous samples can be sufficiently exploited for representation. Since CSSRC is a typical structure sparse representation issue, it can be efficiently solved by the convex optimization. Seven popular visual and audio signal databases are employed for evaluation. The results demonstrate its effectiveness in comparison with the state-of-the-art classifiers. [ABSTRACT FROM AUTHOR]

Active learning reduces the labeling cost by iteratively selecting the most valuable data to query their labels. It has attracted a lot of interests given the abundance of unlabeled data and the high cost of labeling. Most active learning approaches select either informative or representative unlabeled instances to query their labels, which could significantly limit their performance. Although several active learning algorithms were proposed to combine the two query selection criteria, they are usually ad hoc in finding unlabeled instances that are both informative and representative. We address this limitation by developing a principled approach, termed QUIRE, based on the min-max view of active learning. The proposed approach provides a systematic way for measuring and combining the informativeness and representativeness of an unlabeled instance. Further, by incorporating the correlation among labels, we extend the QUIRE approach to multi-label learning by actively querying instance-label pairs. Extensive experimental results show that the proposed QUIRE approach outperforms several state-of-the-art active learning approaches in both single-label and multi-label learning. [ABSTRACT FROM AUTHOR]

Room-temperature gas-sensitive materials are urgently needed for lithium-ion battery monitoring to ensure the safety of battery. In this work, we proposed a strategy for predicting gas-sensitive materials to sense gas in lithium-ion batteries by the combination of machine learning and ab initio calculations. Copper acetylacetonate functionalized perovskite CsPbBr 3 , an excellent room-temperature gas-sensitive material, was chosen as an example to demonstrate the correctness and extensibility of the strategy. It is found that the variation of adsorption characteristic parameters of different gases determines the uniqueness of the electrical response behavior by calculations. However, it is difficult to obtain the correlation between characteristic data and gas-sensitive performance directly. Therefore, using machine learning, combining multiple features with algorithmic voting classifiers can achieve a prediction accuracy of 85.71%. The results show that the features of adsorption energy, band structure and state of density play a major role in the prediction, making a great influence on carrier transport and further affecting the gas-sensitive performance. This work provides a theoretical framework and a new perspective for the follow-up work. [Display omitted] • A strategy to explore gas monitoring inside lithium-ion batteries is proposed. • Adding metal-organic ligands to perovskite materials to enhance gas-sensitive characteristics. • DFT calculations were used to study the mechanism of gas adsorption. • Machine learning was used to create a model for gas identification. • Determination of gas-sensitive behavior using changes in electrical properties. [ABSTRACT FROM AUTHOR]

A machine learning (ML)–based framework has been developed to optimize the process parameters and unravel the paramount process–microstructure–property (PMP) relationships rapidly and precisely, which is demonstrated using electron beam melting (EBM®)-processed Ti–6Al–4V alloy. The process maps are constructed using ML to discover the optimal process window based on the three criteria: fully dense (≥ 99.5% in relative build density), flat fusion surface and highest possible strength, without which the optimization outcome will be deficient. By using the ML-optimized process parameters, Ti–6Al–4V parts were fabricated with ultimate tensile strength ∼5% higher and ∼80% faster than the classical parameter setting developed by the original machine manufacturer. The α/β dual-phase microstructure and tensile deformation mechanism are further studied to understand the superior mechanical properties obtained in the ML-optimized sample. In addition, the information-rich top fusion surface is found to be an alternative route to implicitly materialize the microstructure-to-property relationships with a prediction accuracy of ∼98%. The full-scope surface and microstructure maps are then generated from the process parameters. Surface features that reflect superior and poor mechanical properties are particularly identified. Finally, the implications of the ML-based PMP framework with surface feature correlation are discussed in the context of improving resources utilization efficiency. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, we propose the MIML (Multi-Instance Multi-Label learning) framework where an example is described by multiple instances and associated with multiple class labels. Compared to traditional learning frameworks, the MIML framework is more convenient and natural for representing complicated objects which have multiple semantic meanings. To learn from MIML examples, we propose the MimlBoost and MimlSvm algorithms based on a simple degeneration strategy, and experiments show that solving problems involving complicated objects with multiple semantic meanings in the MIML framework can lead to good performance. Considering that the degeneration process may lose information, we propose the D-MimlSvm algorithm which tackles MIML problems directly in a regularization framework. Moreover, we show that even when we do not have access to the real objects and thus cannot capture more information from real objects by using the MIML representation, MIML is still useful. We propose the InsDif and SubCod algorithms. InsDif works by transforming single-instances into the MIML representation for learning, while SubCod works by transforming single-label examples into the MIML representation for learning. Experiments show that in some tasks they are able to achieve better performance than learning the single-instances or single-label examples directly. [Copyright &y& Elsevier]

Machine learning (ML) has been used for many clinical decision-making processes and diagnostic procedures in bioinformatics applications. We examined eight algorithms, including linear discriminant analysis (LDA), logistic regression (LR), k -nearest neighbor (KNN), random forest (RF), gradient boosting machine (GBM), support vector machine (SVM), Naïve Bayes classifier (NB), and artificial neural network (ANN) models, to evaluate their classification and prediction capabilities for four tissue types in Wolfiporia extensa using their monosaccharide composition profiles. All 8 ML-based models were assessed as exemplary models with AUC exceeding 0.8. Five models, namely LDA, KNN, RF, GBM, and ANN, performed excellently in the four-tissue-type classification (AUC > 0.9). Additionally, all eight models were evaluated as good predictive models with AUC value > 0.8 in the three-tissue-type classification. Notably, all 8 ML-based methods outperformed the single linear discriminant analysis (LDA) plotting method. For large sample sizes, the ML-based methods perform better than traditional regression techniques and could potentially increase the accuracy in identifying tissue samples of W. extensa. [Display omitted] [ABSTRACT FROM AUTHOR]

Ethanol and water are sufficiently miscible, so it is challenging to distinguish ethanol from a mixed atmosphere of ethanol and water. In this work, gas-sensitive materials were prepared with a high response to ethanol, and combined with intelligent algorithms, which can effectively identify ethanol from the mixture of ethanol and water. Specifically, a route was developed to improve the stability of perovskite CsPbBr 3 quantum dots (QDs) by passivating its surface with ligands, leading to excellent gas sensing performance at room temperature (RT). Zinc acetylacetonate (Zn(acac) 2) was introduced into the reaction mixture to stabilize the surface of thin QDs and purify them without losing their perovskite structure or quantum confinement effect. The as-obtained perovskite QDs with capping Zn(acac) 2 displayed an average size of about 9.35 nm, as well as a gas sensing response of 0.275 at 1300 ppm towards ethanol, response/recovery time of 8.5/13.9 s. Such an excellent gas sensing performance was attributed to an interaction of the ethanol molecules with CsPbBr 3 QDs through a metal-organic Zn(acac) 2 molecule, leading to the enhancement of both the sensitivity and response towards ethanol. Furthermore, the accurate recognition of low-concentration ethanol under humidity conditions was also realized with an aid of machine learning methods. Totally, this work is not only helpful for the synthesis of perovskite QDs, but also develop a method for identifying low-concentration ethanol through machine learning under a wet environment. Schematic model of a gas sensor based on perovskite CsPbBr 3 quantum dots. [Display omitted] • Perovskite CsPbBr 3 QDs are synthesized through a solution route, followed by ligand-exchange for improving its moisture stability. • Gas sensor fabricated with CsPbBr 3 QDs displays an excellent gas sensing performance at RT. • As-prepared QDs with capping Zn(acac) 2 displays average size of ∼9.35 nm, gas sensing response of 0.275 at 1300 ppm and response/recovery time of 8.5/13.9 s. • Accurate recognition of trace ethanol under different humidity is realized with an aid of machine learning. [ABSTRACT FROM AUTHOR]

• We used UML to classify AS patients based on similar clinical characteristics. We then used the clinical data to construct a novel subtype predictive model based on the two UML clusters. • To determine the AS pathogenesis process, we analyzed the difference in the immune microenvironment between AS patients and healthy control subjects by immune cell infiltration analysis and verified the results with clinical data and immunohistochemistry. • A novel subtype of AS was identified in UML clustering. CRP, NEU, and MONO were the independent predictive factors for the novel subtype of AS. • FTO expression was correlated with immune cell infiltration in AS patients. Accurate classification of patients with ankylosing spondylitis (AS) is the premise of precision medicine so as to perform different medical interventions for different patient types. AS pathology is closely related to the changes in the immune microenvironment. In this study, we used unsupervised machine learning (UML) to classify patients with AS based on clinical characteristics. We then constructed a novel subtype predictive model for AS based on the clinical classification, after which we investigated the difference in the immune microenvironment to unravel the AS pathogenesis. Overall, 196 patients with AS were enrolled. UML was used to cluster AS patients by similar clinical characteristics. Functional ability, disease status, and grading of radiologic features were assessed to verify the accuracy and heterogeneity of UML clustering. Least Absolute Shrinkage and Selection Operator (LASSO) regression and Random Forest algorithm were used to screen and identify predictive factors for the novel subtype of AS. Logistic regression was also performed to construct a predictive model of this novel subtype. Datasets were downloaded from the Gene Expression Omnibus database to assess immune cell infiltration, and the results were validated using data of routine blood tests from 3671 AS patients and 5720 non-AS patients. The differential expression of Fat Mass and Obesity-Associated Protein (FTO) , an m6A regulator, between AS patients and healthy control subjects was confirmed using immunohistochemistry. UML clustering identified two clusters. The clinical characteristics of the two clusters were significantly heterogeneous. For the novel subtype of AS identified in UML clustering, a predictive model was built using three predictive factors, namely, C-reactive protein (CRP), absolute value of neutrophils (NEU), and absolute value of monocytes (MONO). The area under the curve of the predictive model was 0.983. Heterogeneity in the neutrophil and monocyte counts in AS was verified through immune cell infiltration analysis. Data from routine blood tests revealed that NEU and MONO were significantly higher in AS patients than in non-AS patients (p < 0.001). FTO expression was negatively correlated with both NEU and MONO. Immunohistochemistry analysis confirmed the downregulated expression of FTO. UML provides an explicable and remarkable classification of a heterogeneous cohort of AS patients. A novel subtype of AS was identified in UML clustering. CRP, NEU, and MONO were the independent predictive factors for the novel subtype of AS. FTO expression was correlated with immune cell infiltration in AS patients. [ABSTRACT FROM AUTHOR]

Cascaded regression approaches have been widely applied to computer vision tasks recently, and achieve state-of-the-art performance. In this paper, we consider the problem of face alignment model fitting using cascaded regression and propose a new face alignment approach named Joint Local Regressors Learning (JLRL). The main novelty of our learning framework lies in two following aspects: (1) shape constraints among facial landmarks are considered by jointly learning local regressors; (2) the contribution to face alignment errors for each facial landmark is explored in the training phase. Compared with the previous face alignment methods that have shown state-of-the-art performances, our JLRL approach performed best on the LFPW, Helen and 300-W datasets which are the most challenging datasets today. [ABSTRACT FROM AUTHOR]