关键词： PM2.5 prediction   Environmental exposure   Random forest   Ensemble methods   London case study  

摘要： The adverse health effects of air pollutants, especially those of PM2.5, are well documented. However, a lack of adequate monitoring and weaknesses in modelling approaches do not allow a good assessment of health effects in many areas of the World. Advances in computational methods and the availability of new data sets, e.g. satellite remote observations, have enlarged the possibilities of modelling for application in large scale health effects studies. However, PM2.5 monitoring is very recent in most of the World and more limited compared to other pollutants, and understanding how to use PM(10 )monitors to estimate PM2.5 exposure is therefore important. Since interest in these methods is relatively recent, there is a need for testing their performance against ambient measurements, but long term PM2.5 datasets are less readily available than PM10 in many regions. In the present study we report the methodology and results of using regression modelling and a machine learning method (Random Forest-RF), as well as a combination of the two, to enhance a PM2.5 measurement data base in London using PM10 and NOx measurements as well as other predictors and compare the relative performance of each method. We found that the combination of predictions by the regression model and the RF performs best and we obtain a cross-validation R-2 of 99.29% and 98.22% for the 5-year periods 2004-2008 and 2009-2013, respectively, and a Mean Square Error near 1. Our enhanced data base for PM2.5 is available for use by other researchers.

关键词： Turbulence  

摘要： Deep artificial neural networks (ANNs) are used for modeling sub-grid scale mixing quantities such as the filtered density function (FDF) of the mixture fraction and the conditional scalar dissipation rate. A deep ANN with four hidden layers is trained with carrier-phase direct numerical simulations (CP-DNS) of turbulent spray combustion. A priori validation corroborates that ANN predictions of the mixture fraction FDF and the conditional scalar dissipation rate are in very good agreement with CP-DNS data. ANN modeled solutions show much better performance with a mean error of around 1%, which is one order of magnitude smaller than that of standard modeling approaches such as the beta -FDF and its modified version. The predicted conditional scalar dissipation rate agrees very well with CP-DNS data over the entire mixture fraction space, whereas conventional models derived for pure gas phase combustion fail to describe N|xi = eta in regions with higher mixture fraction and low probability. In the second part of this paper, uncertainties associated with ANN predictions are analyzed. It is shown that a suitable selection of training sets can reduce the size of the necessary test database by similar to 50% without compromising the accuracy. Feature importance analysis is used to analyze the importance of different combustion model parameters. While the droplet evaporating rate, the droplet number density, and the mixture fraction remain the dominant features, the influence of turbulence related parameters only becomes important if turbulence levels are sufficiently high.

关键词： Artificial neural network   Fuzzy regression function   Outlier   Robustness   Support vector machine  

摘要： The presence of outliers in the dependent and/or independent features distorts predictions with machine learning techniques and may lead to erroneous conclusions. It is important to implement methods that are robust against the outliers to make reliable predictions and to know the accuracy of the existing methods when data is contaminated with outliers. The focus of this study is to propose a robust fuzzy regression functions (FRFN) approach against the outliers and evaluate the performance of the proposed and several mainstream machine learning approaches in the presence of outliers for the regression problem. The proposed FRFN approach is based on fuzzy k-means clustering with a noise cluster. We compare the accuracy of Artificial Neural Networks (ANN), Support Vector Machines (SVM) and the proposed FRFN approaches with different training algorithms kernel functions via simulated and real benchmark datasets. In total, accuracies of 36 ANN, SVM, and FRNF implementations with training algorithms and kernel and loss functions have been evaluated and compared to each other with samples containing outliers via a Monte Carlo simulation setting. It is observed in both Monte Carlo simulations and applications with benchmark dataset that FRFN with ANN trained with Bayes regularization algorithm and FRFN with SVM with Gaussian kernel outperforms the classical implementations of ANN and SVMs under the existence of outliers. The proposed noise cluster implementation considerably increases the robustness of fuzzy regression functions against outliers. (C) 2020 Elsevier B.V. All rights reserved.

关键词： Structural dynamics   Physics-guided machine learning   Data-driven machine learning   Recurrent neural network  

摘要： Physics-guided machine learning (ML) is an emerging paradigm that combines both data-driven ML models and physics-based models together to fully take advantage of the data discovery ability of ML without losing the valuable physics/domain knowledge. This paper proposes a novel physics-guided ML method based on recurrent neural network (RNN) and multilayer perceptron (MLP) for structural dynamics simulation. The key idea is to integrate the underlying physics of structural dynamics into data-enabled ML models to `guide' the training and prediction of ML models. First, structural dynamics formulation and the use of data-driven RNN and MLP for modeling dynamical systems are briefly reviewed, which leads to the development of the proposed physics-guided ML model. Physics-guided ML model contains physics-based layers to encode the known physics and data-driven layers to approximate the unknown relationships. Thus, the data-driven RNN and MLP are augmented with existing physics knowledge for better performance in simulations. Following this, several numerical case studies for structural dynamics are presented to demonstrate the proposed methodology. It is observed that: (1) compared with purely data-driven ML method, the proposed physics-guided ML method has better generalization ability and reduced training costs;(2) compared with physics-based modeling, the proposed method has improved computational efficiency and can handle partially unknown physics. Finally, conclusions and future works are presented based on the proposed study.

关键词： image classification   feature extraction   biological tissues   support vector machines   medical image processing   learning (artificial intelligence)   pattern classification   cancer   permittivity   normal tissues   malignant colorectal tissues   dielectric properties   machine learning methods   malignant tissues   computer-aided diagnostic technologies   neighbour   support vector machine classifiers  

摘要： Numerous researchers approved discrepancies in dielectric properties between malignant and normal tissues. Such discrepancies serve as a foundation for the development of computer-aided diagnostic technologies. In this study, machine learning methods were proposed for discrimination between normal and malignant colorectal tissues based on discrepancies in their dielectric properties. To do so, first, two independent-sample t-tests and receiver operating characteristic curve analysis were utilised to examine discrimination power with respect to three types of features, namely, permittivity, conductivity and Cole-Cole fitting parameters. K-nearest neighbour and support vector machine classifiers were used to assess the possibility of combining these features for better classification accuracy. Obtained k-fold cross-validation accuracy reached 88.2%. The obtained accuracy indicated the potential capability of discrimination between normal and malignant colorectal tissues based on discrepancies in their dielectric properties.

关键词： Diffusion   Activation energy   Machine learning   Support vector machine  

摘要： We evaluate the performance of a popular machine learning (ML) method support vector machine (SVM) for modeling and predicting the solute diffusion activation energies in fcc, bcc, and hcp metallic hosts. The diffusion activation energies of 408 host-solute systems from ab-initio calculations are made as our dataset. We obtain an optimal set of features by combining prior physics knowledge and combination ranking based on the LeaveGroup-Out (LOG) cross-validation (CV) method, including solute migration barrier, atomic volume of host, elastic modulus of host, melting point of host, unpaired d electrons of host, and the corresponding parameters of solute. We present the results of LOG/10-fold/5-fold/3-fold CV, with the corresponding root mean squared error (RMSE) of 0.128/0.106 +/- 0.014/0.107 +/- 0.004/0.110 +/- 0.005 eV. SVM gives an about 0.1 eV errors when extrapolating to new host-solute systems for main hosts. We further make predictions on the activation energies of thousands of new systems with quite small computational cost. Our work demonstrates that the ML method is a promising method to accelerate materials science researches.

关键词： Streamflow forecast   Artificial neural network   Long short term memory   Long lead-time   Machine learning  

摘要： Long lead-time streamflow forecasting is of great significance for water resources planning and management in both the short and long terms. Despite of some studies using machine learning methods in streamflow forecasting, only few studies have been conducted to explore long lead-time forecasting capabilities of these methods, and gain an insight into systematic comparison of model forecasting performance in both the short and long terms. In this work, an artificial neural network (ANN) and a long short term memory (LSTM), a powerful tool for learning long-term temporal dependencies and capturing nonlinear relationship, have been adopted to forecast streamflow at daily and monthly scales for a long lead-time period. For long lead-time streamflow forecasting, a recursive forecasting procedure, which takes the last one-step-ahead forecast as a new input for the next-step-ahead forecast, is used in the ANN and LSTM forecasting systems. Two models are trained and validated for streamflow forecasting using the rainfall and runoff datasets collected from the Nan River Basin and Ping River Basin, Thailand, covering the period 1974 to 2014. To further explore the impact of parameter settings on model performance, two parameters, i.e. the length of time lag and the number of maximum epochs, are examined in the ANN and LSTM models. The main findings are highlighted here. First, with an optimal setting up of model parameters, both the ANN and LSTM model can provide accurate daily forecasting (up to 20 days ahead). Second, in comparison to the ANN model, the LSTM model exhibits better model performance in long lead-time daily forecasting, but less satisfactory in multi-monthly forecasting due to lack of large monthly training dataset. Third, the selection of the length of the time lag and number of maximum epochs used in both ANN and LSTM modelling are the key for long lead-time streamflow forecasting at daily and monthly scales. These findings suggest that the LSTM could be ad

关键词： machine learning   drug discovery   deep learning   in silico screening  

摘要： The advancements of information technology and related processing techniques have created a fertile base for progress in many scientific fields and industries. In the fields of drug discovery and development, machine learning techniques have been used for the development of novel drug candidates. The methods for designing drug targets and novel drug discovery now routinely combine machine learning and deep learning algorithms to enhance the efficiency, efficacy, and quality of developed outputs. The generation and incorporation of big data, through technologies such as high-throughput screening and high through-put computational analysis of databases used for both lead and target discovery, has increased the reliability of the machine learning and deep learning incorporated techniques. The use of these virtual screening and encompassing online information has also been highlighted in developing lead synthesis pathways. In this review, machine learning and deep learning algorithms utilized in drug discovery and associated techniques will be discussed. The applications that produce promising results and methods will be reviewed.

关键词： family designs   gene-gene interaction   ischemic stroke   sparse factor analysis   unsupervised machine learning  

摘要： Gene-gene interaction (G x G) is thought to fill the gap between the estimated heritability of complex diseases and the limited genetic proportion explained by identified single-nucleotide polymorphisms. The current tools for exploring G x G were often developed for case-control designs with less considerations for their applications in families. Family-based studies are robust against bias led from population stratification in genetic studies and helpful in understanding G x G. We proposed a new algorithm epistasis sparse factor analysis (EPISFA) and epistasis sparse factor analysis for linkage disequilibrium (EPISFA-LD) based on unsupervised machine learning to screen G x G. Extensive simulations were performed to compare EPISFA/EPISFA-LD with a classical family-based algorithm FAM-MDR (family-based multifactor dimensionality reduction). The results showed that EPISFA/EPISFA-LD is a tool of both high power and computational efficiency that could be applied in family designs and is applicable within high-dimensionality datasets. Finally, we applied EPISFA/EPISFA-LD to a real dataset drawn from the Fangshan/family-based Ischemic Stroke Study in China. Five pairs of G x G were discovered by EPISFA/EPISFA-LD, including three pairs verified by other algorithms (FAM-MDR and logistic), and an additional two pairs uniquely identified by EPISFA/EPISFA-LD only. The results from EPISFA might offer new insights for understanding the genetic etiology of complex diseases. EPISFA/EPISFA-LD was implemented in R. All relevant source code as well as simulated data could be freely downloaded from .

关键词： Visible Light Positioning   robust   machine learning   multi layer perceptron   gaussian processes  

摘要： In this work, the use of Machine Learning methods for robust Received Signal Strength (RSS)-based Visible Light Positioning (VLP) is experimentally evaluated. The performance of Multilayer Perceptron (MLP) models and Gaussian processes (GP) is investigated when using relative RSS input features. The experimental set-up for the RSS-based VLP technology uses light-emitting diodes (LEDs) transmitting intensity modulated light and a single photodiode (PD) as a receiver. The experiments focus on achieving robustness to cope with unknown received signal strength modifications over time. Therefore, several datasets were collected, where per dataset either the LEDs transmitting power is modified or the PD aperture is partly obfuscated by dust particles. Two relative RSS schemes are investigated. The first scheme uses the maximum received light intensity to normalize the received RSS vector, while the second approach obtains RSS ratios by combining all possible unique pairs of received intensities. The Machine Learning (ML) methods are compared to a relative multilateration implementation. It is demonstrated that the adopted MLP and GP models exhibit superior performance and higher robustness when compared to the multilateration strategies. Furthermore, when comparing the investigated ML models, the GP model is proven to be more robust than the MLP for the considered scenarios.