On a problem of Hornik

In 1991,Hornik proved that the collection of single hidden layer feedforward neural networks(SLFNs)with continuous,bounded,and non-constant activation functionσis dense in C(K)where K is a compact set in R~s(see Neural Networks,4(2),251-257(1991)).Meanwhile,he pointed out"Whether or not the continuity assumption can entirely be dropped is still an open quite challenging problem".This paper replies in the affirmative to the problem and proves that for bounded and continuous almost everywhere(a.e.)activation functionσon R,the collection of SLFNs is dense in C(K)if and only ifσis un-constant a.e..     

Unified flexible total synthesis of chlorofusin and artificial Click mimics as antagonists against p53–HDM2 interactions

A practical HPLC-free total synthesis of chlorofusin has been successfully accomplished in this work. The new synthesis showed great flexibility and convenience in generating artificial natural product-like mimics through Click chemistry, and enabled us to further investigate the biological importance of the chromophore and the spiro-aminal functionality of the natural product. The entire skeleton is believed to be essential for satisfactory biological activities of both natural chlorofusin and unnatural mimics. Two artificial Click hybrids were found to exhibit improved inhibitory activity against p53–HDM2 bindings over the natural product.     

基于布谷鸟算法与BP神经网络的煤灰变形温度预测

以120种煤样为数据基础,采用布谷鸟算法(CS)优化BP(Back Propagation)神经网络,建立了CSBP模型对单煤、煤掺添加剂和配煤等3类样本的煤灰变形温度(DT)样本进行预测。模型以煤灰化学成分及其组合参数等13个变量作为输入量,以变形温度(DT)作为输出量。CSBP模型预测结果与BP神经网络模型预测结果进行对比发现,无论是单煤、煤掺添加剂还是配煤,CSBP模型较BP模型对煤灰变形温度(DT)的预测都更加精准,平均相对误差分别达到了3.11%、4.08%和4.22%。另外,对比3类样本预测结果发现,无论是CSBP模型还是BP模型,相比单煤预测而言,煤掺添加剂及配煤的预测误差都有明显的增加。     

A Combined Spectroscopic and Theoretical Study on a Ruthenium Complex Featuring a π-Extended dppz Ligand for Light-Driven Accumulation of Multiple Reducing Equivalents

The design of photoactive systems capable of storing and relaying multiple electrons is highly demanded in the field of artificial photosynthesis, where transformations of interest rely on multielectronic redox processes. The photophysical properties of the ruthenium photosensitizer [(bpy)₂Ru( oxim-dppqp )]²⁺ ( Ru ), storing two electrons coupled to two protons on the π-extended oxim-dppqp ligand under light-driven conditions, are investigated by means of excitation wavelength-dependent resonance Raman and transient absorption spectroscopies, in combination with time-dependent density functional theory; the results are discussed in comparison to the parent [(bpy)₂Ru(dppz)]²⁺ and [(bpy)₂Ru( oxo-dppqp )]²⁺ complexes. In addition, this study provides in-depth insights on the impact of protonation or of accumulation of multiple reducing equivalents on the reactive excited states.     

Process prediction of Ni–SiC coatings based on RBF-BP model

The preparation of Ni–SiC coatings using magnetic field-assisted jet electrodeposition under various plating settings is described in this study. A RBF-BP composite neural network with 4 × 4 × 4 × 7 × 10 × 1 was used to predict the corrosion resistance of Ni–SiC coatings prepared by employing different plating parameters. The results show that the fitting degree between the expected value and the actual value of the RBF-BP composite neural network is 0.97497. Moreover, the hybrid neural network can accurately predict the corrosion resistance of Ni–SiC coatings prepared under different process parameters. The corrosion weight loss of the coating is the lowest at the current density of 4 A/dm², a jet rate of 3 m/s, a SiC particle concentration of 8 g/L, and at a magnetic field intensity of 0.8 T, demonstrating its corrosion resistance under these conditions. According to the coating characterization analysis, the coating's grain size was significantly refined, and the surface was smoother with a high amount of uniformly sized SiC nanoparticles.     

Computational modeling of Hg/Ni ions separation via MOF/LDH nanocomposite: Machine learning based modeling

Nowadays, sustainable supplement of water has recently been identified as a vital necessity due to the existence of limited drinkable water sources. To do this, various techniques are being developed to remove various types of pollutants from water/wastewater sources. Adsorption of common water pollutants using nanocomposite materials has been of great popularity in recent years due to its high efficiency. This paper aims to develop various models based on machine learning approach to study their efficiency on predicting the experimentally measured results of Hg/Ni ions removal from water sources. To do this, this study attempts regression on a small data set using two parameters as inputs and two parameters as outputs. In this dataset, the inputs are Ion and C0, and the outputs are Ce and Qe. AdaBoost (Adaptive Boosting), a well-known ensemble method, was applied on top of three different models, including Decision Tree Regression (DT), Gaussian Process Regression (GPR), and Linear Regression (LR). After fine-tuning their hyper-parameters, the optimized model was evaluated through various metrics. For example, the R² for ADA + GPR model has a score of 0.998 for Ce and 0.999 for Qe as the best model among these three models. This model in RMSE is the best and illustrates 0.1512 and 1.490 for Ce and Qe as error. Eventually, ADA + GPR has been selected as the optimized model with optimized dataset: (Ion = Ni, C0 = 250, Ce = 206.0). But for Qe, different amounts are illustrated: (Ion = Hg, C0 = 106.7, Ce = 577.35)     

Artificial intelligence-based microfluidic platforms for the sensitive detection of environmental pollutants: Recent advances and prospects

Environmental pollution and its drastic effects on human and animal health have urged governments to implement strict policies to minimize damage. The first step in applying such policies is to find reliable methods to detect pollution in various media, including water, food, soil, and air. In this regard, various approaches such as spectrophotometric, chromatographic, and electrochemical techniques have been proposed. To overcome the limitations associated with conventional analytical methods, microfluidic devices have emerged as sensitive technologies capable of generating high content information during the past few years. The passage of contaminant samples through the microfluidic channels provides essential details about the whole environment after detection by the detector. In the meantime, artificial intelligence is an ideal means to identify, classify, characterize, and even predict the data obtained from microfluidic systems. The development of microfluidic devices with integrated machine learning and artificial intelligence is promising for the development of next-generation monitoring systems. Combination of the two systems ensures time efficient setups with easy operation. This review article is dedicated to the recent developments in microfluidic chips coupled with artificial intelligence technology for the evolution of more convenient pollution monitoring systems.     

Discrimination of Beers by Cyclic Voltammetry Using a Single Carbon Screen-printed Electrode

A fast, simple and costless methodology without sample pre-treatment is proposed for the discrimination of beers. It is based on cyclic voltammetry (CV) using commercial carbon screen-printed electrodes (SPCE) and includes a correction of the signals measured with different SPCE units. Data are submitted to partial least squares discriminant analysis (PLS−DA) and support vector machine discriminant analysis (SVM−DA), which allow a reasonable classification of the beers. Also, CV data from beers can be used to predict their alcoholic degree by partial least squares (PLS) and artificial neural networks (ANN). In general, non-linear methods provide better results than linear ones.     

Incompatibility networks as models of scale-free small-world graphs

We make a mapping from Sierpinski fractals to a new class of networks, the incompatibility networks, which are scale-free, small-world, disassortative, and maximal planar graphs. Some relevant characteristics of the networks such as degree distribution, clustering coefficient, average path length, and degree correlations are computed analytically and found to be peculiarly rich. The method of network representation can be applied to some real-life systems making it possible to study the complexity of real networked systems within the framework of complex network theory.