关键词： 黎曼流形   黎曼几何   计算机视觉   深度学习   数学基本概念   机器学习方法  

摘要： |史泽林黎曼几何在计算机视觉中的应用研究进展从数学基本概念出发,阐述黎曼流形与图像的关系以及视觉应用的可行性,并介绍计算机视觉中具有重要应用的几种黎曼流形。对黎曼流形在计算机视觉中若干常见应用进行了概述,重点介绍了与深度学习相结合的相关进展。最后,对引入黎曼流形的机器学习方法的未来发展进行了分析和讨论。

关键词： supercomputer   high performance computing   job management system   supercomputer job scheduling   machine learning   job runtime prediction   SUPPZ   random forest  

摘要： The paper is devoted tomachine learningmethods and algorithms for the supercomputer jobs execution prediction. The supercomputers statistics shows that the actual runtime of the most of the jobs substantially diverges from the time requested by the user. This reduces the efficiency of scheduling jobs, since an inaccurate job execution time estimation leads to a suboptimal jobs schedule. The job classification is considered, it is based on the difference between the job actual and the requested execution time. Forecast was made on the base of supercomputer multiuser job management system statistics by assigning a submitted job to one of the classes. The statistics of supercomputers MVS-100K and MVS-10P in the Joint Supercomputer Center of the Russian Academy of Sciences (JSCC RAS) was used. The job flow feature ranking by importance was done on the statistical analysis results. The cross-correlation of the most important features was determined. The probability estimates of correct prediction were obtained for selected wellknown machine learning algorithms: logistic regression, decision trees, k-nearest neighbors, linear discriminant analysis, support vector machine, random forest, gradient boosting, and feedforward neural network. The best values were obtained using the random forest method.

关键词： pixel-based supervised classification   random forest   support vector machine   gray level cooccurrence matrix (GLCM)   principal component analysis (PCA)   WorldView-3  

摘要： Spatially explicit information on tree species composition is important for both the forest management and conservation sectors. In combination with machine learning algorithms, very high-resolution satellite imagery may provide an effective solution to reduce the need for labor-intensive and time-consuming field-based surveys. In this study, we evaluated the possibility of using multispectral WorldView-3 (WV-3) satellite imagery for the classification of three main tree species (Quercus robur L., Carpinus betulus L., and Alnus glutinosa (L.) Geartn.) in a lowland, mixed deciduous forest in central Croatia. The pixel-based supervised classification was performed using two machine learning algorithms: random forest (RF) and support vector machine (SVM). Additionally, the contribution of gray level cooccurrence matrix (GLCM) texture features from WV-3 imagery in tree species classification was evaluated. Principal component analysis confirmed GLCM variance to be the most significant texture feature. Of the 373 visually interpreted reference polygons, 237 were used as training polygons and 136 were used as validation polygons. The validation results show relatively high overall accuracy (85%) for tree species classification based solely on WV-3 spectral characteristics and the RF classification approach. As expected, an improvement in classification accuracy was achieved by a combination of spectral and textural features. With the additional use of GLCM variance, the overall accuracy improved by 10% and 7% for RF and SVM classification approaches, respectively.

关键词： Reference evapotranspiration   Deep learning neural network   Temporal convolution neural network   Long short-term memory neural network  

摘要： To evaluate the performance of deep learning methods (DL) for reference evapotranspiration estimation and to assess the applicability of the developed DL models beyond the study areas where they were trained, three popular DL models named deep neural network (DNN), temporal convolution neural network (TCN), and long short-term memory neural network (LSTM) were developed to estimate daily reference evapotranspiration (ETo) using incomplete meteorological data in the Northeast plain, China. The performances of the three DL models were compared to two classical machine learning models (CML)-support vector machine (SVM) and random forest (RF)-and empirical equations, including two temperature-based (Hargreaves (H) and modified Hargreaves (MH)), three radiation-based (Ritchie (R), Priestley-Talor (P), and Makkink (M)), and two humiditybased (Romanenko (ROM) and Schendel (S)) empirical models, in two strategies: (1) all proposed models were trained, tested, and compared in each single weather station, and (2) all-weather stations were split into several groups using the K-means method with their mean climatic characteristics. Then, in each group, stations took turns testing the proposed models which were trained by rest of the stations. The results showed that (1) the coefficient of determination (R-2) values of the TCN and RF were 0.048 and 0.035 significantly higher than that of MH, respectively, and the relative root mean error (RMSE) values of TCN and RF were substantially 0.096, and 0.074 mm/d lower than that of MH, indicating that TCN and RF performed better than empirical models in the first strategy, and TCN and LSTM exhibited an RMSE that was significantly decreased by 0.069 and 0.079 mm/d, showing that TCN and LSTM outperformed empirical models in the second strategy, compared with the MH method;(2) in both strategies, compared with the Ritchie (R) model, TCN, LSTM, DNN, RF, and SVM increased R-2 and decreased RMSE significantly, especially the TCN model;(3) sim

关键词： Bayesian model averaging   Ensemble method   Solar radiation   Wavelet   Artificial neural networks   Extreme learning machines   Radial basis function  

摘要： The present study investigated the potential of new ensemble method, Bayesian model averaging (BMA), in modeling monthly solar radiation based on climatic data. Data records covered monthly maximum temperature (T-max), minimum temperature (T-min), sunshine hours (H-s), wind speed (W-s), relative humidity (RH), and solar radiation values obtained from two weather stations of Turkey. The BMA estimates were compared with the artificial neural networks (ANN), extreme learning machines (ELM), radial basis function (RBF), and their hybrid versions with wavelet transform technique (wavelet-ANN or WANN, wavelet-ELM or WELM, and wavelet-RBF or WRBF). Three evaluation criteria e.g., root mean square error (RMSE), Nash-Sutcliffe efficiency, and determination coefficient (R-2), were applied to measure the accuracy of the employed methods. The results indicated the superior accuracy of the BMA4 models over six machine learning models for estimating monthly solar radiation;improvements in accuracy of ANN4, ELM4, RBF4, WANN4, WELM4, and WRBF4 models comprising T-max, T-min, H-s, W(s)and RH input variables were about 56-41%, 44-31%, 57-46%, 35-26%, 27-16%, and 43-28% in terms of RMSE reduction in both stations. While the hybrid models (i.e., WANN4, WELM4, and WRBF4) increased the accuracy of the single models about 31-21%, 23-18%, and 26-25% for ANN4, ELM4, and RBF4, respectively.

关键词： DERA   Aldolase   Protein engineering   Machine learning   Crystal structure determination   C–   C bond formation   Biocatalysis  

摘要： In this work, deoxyribose-5-phosphate aldolase (Ec DERA, EC 4.***.2.4) from Escherichia coli was chosen as the protein engineering target for improving the substrate preference towards smaller, non-phosphorylated aldehyde donor substrates, in particular towards acetaldehyde. The initial broad set of mutations was directed to 24 amino acid positions in the active site or in the close vicinity, based on the 3D complex structure of the E. coli DERA wild-type aldolase. The specific activity of the DERA variants containing one to three amino acid mutations was characterised using three different substrates. A novel machine learning (ML) model utilising Gaussian processes and feature learning was applied for the 3rd mutagenesis round to predict new beneficial mutant combinations. This led to the most clear-cut (two- to threefold) improvement in acetaldehyde (C2) addition capability with the concomitant abolishment of the activity towards the natural donor molecule glyceraldehyde-3-phosphate (C3P) as well as the non-phosphorylated equivalent (C3). The Ec DERA variants were also tested on aldol reaction utilising formaldehyde (C1) as the donor. Ec DERA wild-type was shown to be able to carry out this reaction, and furthermore, some of the improved variants on acetaldehyde addition reaction turned out to have also improved activity on formaldehyde.

关键词： Shock-shock interaction   Machine learning   Hypersonic flow   Impinging jet   Genetic programming   Triple point  

摘要： To predict the maximum heating load induced by shock-shock interaction more reliably and accurately, the geometrical scale of the overall wave configuration of shock-shock interaction is very useful. However, it is hard to be solved with traditional shock theory due to its complexity. The results of numerical and experimental studies are case-by-case. Concise formulas correlating the geometrical scales of shock-shock interaction with the given flow parameters are desired but still unavailable. In the present work, a set of correlative formulas for the triple-points' coordinates of type IVa, IV, and III shock-shock interaction are derived by multilevel block building algorithm, a functional machine learning method. The key flow structure of shock-shock interaction, i.e., the supersonic impinging jet, can be determined with the help of shock theories and the formulas. In addition, the transition criteria respectively for the overall wave configuration transitions of type IVa <-> type IV and type IV <-> type III shock-shock interaction can be obtained by the machine learning based method. (c) 2020 Elsevier Masson SAS. All rights reserved.

关键词： Partial discharges   Industries   Condition monitoring   Training   Machine learning algorithms   Focusing   Real-time systems   condition monitoring   deep learning   partial discharge   intelligent fault diagnostics   machine learning  

摘要： This paper presents a state-of-the-art review on machine learning (ML) based intelligent diagnostics that have been applied for partial discharge (PD) detection, localization, and pattern recognition. ML techniques, particularly those developed in the last five years, are examined and classified as conventional ML or deep learning (DL). Important features of each method, such as types of input signal, sampling rate, core methodology, and accuracy, are summarized and compared in detail. Advantages and disadvantages of different ML algorithms are discussed. Moreover, technical roadblocks preventing intelligent PD diagnostics from being applied to industry are identified, such as insufficient/imbalanced dataset, data inconsistency, and difficulties in cost-effective real-time deployment. Finally, potential solutions are proposed, and future research directions are suggested.

关键词： soil salinity   remote sensing   machine learning   predictive mapping  

摘要： Soil salinization, one of the most severe global land degradation problems, leads to the loss of arable land and declines in crop yields. Monitoring the distribution of salinized soil and degree of salinization is critical for management, remediation, and utilization of salinized soil;however, there is a lack of thorough assessment of various data sources including remote sensing and landscape characteristics for estimating soil salinity in arid and semi-arid areas. The overall goal of this study was to develop a framework for estimating soil salinity in diverse landscapes by fusing information from satellite images, landscape characteristics, and appropriate machine learning models. To explore the spatial distribution of soil salinity in southern Xinjiang, China, as a case study, we obtained 151 soil samples in a field campaign, which were analyzed in laboratory for soil electrical conductivity. A total of 35 indices including remote sensing classifiers (11), terrain attributes (3), vegetation spectral indices (8), and salinity spectral indices (13) were calculated or derived and correlated with soil salinity. Nine were used to model and estimate soil salinity using four predictive modelling approaches: partial least squares regression (PLSR), convolutional neural network (CNN), support vector machine (SVM) learning, and random forest (RF). Testing datasets were divided into vegetation-covered and bare soil samples and were used for accuracy assessment. The RF model was the best regression model in this study, with R-2 = 0.75, and was most effective in revealing the spatial characteristics of salt distribution. Importance analysis and path modeling of independent variables indicated that environmental factors and soil salinity indices including digital elevation model (DEM), B10, and green atmospherically resistant vegetation index (GARI) showed the strongest contribution in soil salinity estimation. This showed a great promise in the measurement and monitoring of

关键词： land subsidence   persistent scatters interferometry   remote sensing   machine learning  

摘要： In the Beijing Plain, land subsidence is one of the most prominent geological problems, which is affected by multiple factors. Groundwater exploitation, thickness of the Quaternary deposit and urban development and construction are important factors affecting the formation and development of land subsidence. Here we choose groundwater level change, thickness of the Quaternary deposit and index-based built-up index (IBI) as influencing factors, and we use the influence factors to predict the subsidence amount in the Beijing Plain. The Sentinel-1 radar images and the persistent scatters interferometry (PSI) were adopted to obtain the information of land subsidence. By using Google Earth Engine platform and Landsat8 optical images, IBI was extracted. Groundwater level change and thickness of the Quaternary deposit were obtained from hydrogeological data. Machine learning algorithms Linear Regression and Principal Component Analysis (PCA) were used to investigate the relationship between land subsidence and influencing factors. Based on the results obtained by Linear Regression and PCA, a suitable machine learning algorithm was selected to predict the subsidence amount in the Beijing Plain in 2018 through influencing factors. In this study, we found that the maximum subsidence rate in the Beijing Plain had reached 115.96 mm/y from 2016 to 2018. The land subsidence was serious in eastern Chaoyang and northwestern Tongzhou. In addition, the area where thickness of the Quaternary deposit reached 150-200 m was prone to more serious land subsidence in the Beijing Plain. In groundwater exploitation, the second confined aquifer had the greatest impact on land subsidence. Through Linear Regression and PCA, we found that the relationship between land subsidence and influencing factors was nonlinear. XGBoost was feasible to predict subsidence amount. The prediction accuracy of XGBoost on the subsidence amount reached 0.9431, and the mean square error was controlled at 15.97. By u