Universality of deep convolutional neural networks

Deep learning has been widely applied and brought breakthroughs in speech recognition, computer vision, and many other domains. Deep neural network architectures and computational issues have been well studied in machine learning. But there lacks a theoretical foundation for understanding the approximation or generalization ability of deep learning methods generated by the network architectures such as deep convolutional neural networks. Here we show that a deep convolutional neural network (CNN) is universal, meaning that it can be used to approximate any continuous function to an arbitrary accuracy when the depth of the neural network is large enough. This answers an open question in learning theory. Our quantitative estimate, given tightly in terms of the number of free parameters to be computed, verifies the efficiency of deep CNNs in dealing with large dimensional data. Our study also demonstrates the role of convolutions in deep CNNs.     

基于双向循环神经网络的汉语语音识别*

当前基于深度神经网络模型中,虽然其隐含层可设置多层,对复杂问题适应能力强,但每层之间的节点连接是相互独立的,这种结构特性导致了在语音序列中无法利用上下文相关信息来提高识别效果,而传统的循环神经网络虽然做出了改进,但是只能对上文信息进行利用。针对以上问题,该文采用可以同时利用语音序列中上下文相关信息的双向循环神经网络模型与深度神经网络模型相结合,并应用于语音识别。构建具有5层隐含层的模型,其中第3层为双向循环神经网络结构,其他层采用深度神经网络结构。实验结果表明:加入了双向循环神经网络结构的模型与其他模型相比,较好地提高了识别正确率;噪声对双向循环神经网络汉语识别有重要影响,尤其是训练集和测试集附加噪声类型不同时,单一的含噪声语音的训练模型无法适应不同噪声类型的语音识别;调整神经网络模型中隐含层神经元数量后,识别正确率并不是一直随着隐含层中神经元数量的增加而增加,神经元数量数目增加到一定程度后正确率出现了降低的趋势。     

基于深度学习的ELM实时识别研究

基于深度学习的方法，在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习，得到一个深度为22层的卷积神经网络。为衡量算法的识别能力，识别了HL-2A装置自2009年实现稳定ELMy H模放电以来所有历史数据(约26000次放电数据)，共识别出1665次H模放电，其中误识别35次，误报率为2.10%。在实际的1634次H模放电中，漏识别4次，漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下，对单个时间点的平均计算时间为0.46ms，可以满足实时控制的计算速度要求。     

SE-MCNN-CTC的中文语音识别声学模型

针对国内外缺少对振动轮噪声预估的问题,以某型振动轮为研究对象,首先基于动力学有限元理论对振动轮进行频率响应分析,其次采用声学边界元技术对振动轮辐射噪声进行了数值模拟,并通过实验验证了仿真结果的准确性,然后比较了垂直振动与圆周振动两种不同激振形式对辐射噪声的影响,得出垂直振动辐射噪声低的结论,最后对驾驶室声腔模态进行了仿真,与振动轮激振频率相近发生共振。通过调整激振频率,降低了司机耳旁噪声。所得研究成果可为振动轮辐射噪声的预估与改进提供一种切实可行的参考依据。     

Nanocomposite hydrogel consisting of Na‐montmorillonite with enhanced mechanical properties

A new kind of nanocomposite (NC) hydrogel with Na‐montmorillonite (MMT) is presented in this article. The NC hydrogels were synthesized by free radical copolymerization of acrylamide and (3‐acrylamidopropyl) trimethylammonium chloride (ATC) in the presence of MMT and N,N′‐methylene‐bis‐acrylamide used as chemical cross‐linker. Due to the cation‐exchange reaction between MMT and ATC (cationic monomer) during the synthesis of NC hydrogels, MMT platelets were considered chemical “plane” cross‐linkers, different from “point” cross‐linkers. With increasing amount of MMT, the crosslinking degree enhanced, causing a decrease of the swelling degree at equilibrium. Investigations of mechanical properties indicated that NC hydrogels exhibited enhanced strength and toughness, which resulted from chemical interaction between exfoliated MMT platelets and polymer chains in hydrogels. Dynamic shear measurements showed that both storage modulus and loss modulus increased with increasing MMT content. The idea described here provided a new route to prepare hydrogels with high mechanical properties by using alternative natural Na‐MMT. © 2015 Wiley Periodicals, Inc. J. Polym. Sci., Part B: Polym. Phys. 2015 , 53, 1020–1026     

Pd(0) nanoparticles immobilized on multinitrogen functionalized halloysite for promoting Sonogashira reaction: studying the role of the number of surface nitrogens in catalytic performance

Halloysite nanoclay, Hal, was amine-functionalized and subsequently reacted with 2,4,6-trichloro-1,3,5-triazine, TCT, and ethylenediamine, EDA, to provide multinitrogen containing functionality on the surface of Hal. The resulting surface-modified Hal, Hal-2N-TCT-EDA, was then used for immobilization of Pd nanoparticles and affording a heterogeneous catalyst, Pd@Hal-2N-TCT-EDA, with utility for copper and ligand-free Sonogashira coupling of alkynes and aryl halides. The results established the efficiency of this protocol in terms of product yield, ecofriendly nature, and reaction time. Study of the reusability of the catalyst confirmed that the catalyst could be recovered and recycled up to seven times with slight loss of catalytic activity and Pd leaching, indicating the efficiency of Hal-2N-TCT-EDA for embedding Pd nanoparticles. To elucidate the role of the number of surface nitrogens on the catalytic performance, the catalytic activity, and recyclability of the catalyst was compared with those of Hal-2N and Hal-2N-TCT. It was found that more surface nitrogen atoms gave higher loading of Pd and lower Pd leaching. This result confirms the contribution of surface nitrogens to anchor the Pd species and suppress leaching.

     

Nano-liquid chromatography combined with a sustainable microextraction based on natural deep eutectic solvents for analysis of phthalate esters

The separation of 11 phthalic acid ester (PAEs) was carried out by nano-liquid chromatography coupled to ultraviolet and MS detection. Preliminary experiments were achieved in order to select suitable stationary phases and chromatographic conditions. The baseline separation was obtained, for all compounds, with an XBridge^TM C₁₈ column in less than 15 min, working in step gradient mode. The sensitivity of the method was improved by on-column focusing. PAEs were extracted from alcoholic and nonalcoholic beverages using vortex-assisted emulsification dispersive liquid–liquid microextration and natural deep eutectic solvents. The whole method was validated in terms of linearity, sensitivity, precision, recovery, and repeatability. Combination of both off-line sample preparation preconcentration and large injection volume led to obtain LOQs in the range 5–47 ng/mL. The developed nano-LC-UV method was extended to MS detection to confirm the presence of PAEs in some beverages commercialized in different types of packaging.     

ABLE: Attention based learning for enzyme classification

Classifying proteins into their respective enzyme class is an interesting question for researchers for a variety of reasons. The open source Protein Data Bank (PDB) contains more than 1,60,000 structures, with more being added everyday. This paper proposes an attention-based bidirectional-LSTM model (ABLE) trained on over sampled data generated by SMOTE to analyse and classify a protein into one of the six enzyme classes or a negative class using only the primary structure of the protein described as a string by the FASTA sequence as an input. We achieve the highest F1-score of 0.834 using our proposed model on a dataset of proteins from the PDB. We baseline our model against eighteen other machine learning and deep learning networks, including CNN, LSTM, Bi-LSTM, GRU, and the state-of-the-art DeepEC model. We conduct experiments with two different oversampling techniques, SMOTE and ADASYN. To corroborate the obtained results, we perform extensive experimentation and statistical testing.     

温度对低共熔溶剂影响的二维拉曼光谱研究

低共熔溶剂（DES）作为一种新式的绿色溶剂，在多种化学过程中表现出色，因此在诸多热门范畴都展现出良好的发展潜力。对DES的光谱分析通常局限在一维光谱技术上，但其分辨率低、谱峰重叠严重等缺点，导致光谱数据存在误差。运用二维拉曼光谱（2D Raman），能够明显提高光谱分辨率，并发现重叠峰位置，获得在外扰条件下不同谱峰的变化顺序及其相互作用等重要信息，以实现对复杂体系的精确分析。以氯化胆碱（ChCl）和ZnCl₂合成的DES为例，利用显微共焦激光拉曼光谱仪分别对ChCl和DES进行拉曼实验，发现与ChCl相比，DES中各处峰的整体强度显著下降，原有的谱峰未消失，说明Zn2+的加入没有破坏ChCl的骨架结构。287 cm-1处出现一个新的特征峰，推测有Zn-Cl配位键的伸缩振动。对DES进行升温拉曼实验，发现随着温度的升高，NC₄的不对称伸缩振动峰强度逐渐减小，峰宽变大，峰形变缓，Zn-Cl配位键伸缩振动峰强度逐渐降低，峰位置基本不变，峰形有明显重叠。运用2D Raman技术对溶液内NC₄和Zn-Cl特征峰的变化进行研究，结果表明，随着温度的升高，溶液中发生ChCl向Ch+的解离过程，Zn2+与Cl-形成了多种配合物，ZnCl-₃，Zn₂Cl-₅，Zn₃Cl-₇之间存在相互转化。不同位置的特征峰随温度的变化顺序不同，将特征峰与团簇进行一一归属，得出了各个团簇对温度的敏感程度。基于量子化学中的密度泛函理论，对推测的物质结构进行构型优化和参数计算，证实了其存在的可能性，同时也验证了2D Raman的分析结果准确可行。这些结果将为DES的后续研究提供理论参考，拓展了二维光谱技术的应用范围。     

Effect of Acid Treatment of Montmorillonite on “Support‐Activator” Performance to Support Metallocene for Propylene Polymerization Catalyst

This work is focused on montmorillonite (MMT)‐based “support‐activators” (S‐As) for the metallocene‐catalyzed propylene polymerization. This catalyst was previously industrialized; however, for further technological advances, the activation mechanism is investigated. The chemical and morphological requirements of the S‐A are surveyed using both commercially available raw clay minerals (non‐acid‐treated) and acid‐treated clay minerals. The S‐A possessing strong‐acid sites (pK a < ?8.2) gives a highly active catalyst. Acid treatment of MMT induces morphological changes as well as the formation of strong acid sites. Based on pore size distribution analysis and atomic force microscopy observations, it is concluded that the strong acid sites are located in the small pores around the edge of the clay mineral (not in the interlayer), where the structure is disordered by the acid treatment.