Density Functional Theory as a Data Science

The development of density functional theory (DFT) functionals and physical corrections are reviewed focusing on the physical meanings and the semiempirical parameters from the viewpoint of data science. This review shows that DFT exchange‐correlation functionals have been developed under many strict physical conditions with minimizing the number of the semiempirical parameters, except for some recent functionals. Major physical corrections for exchange‐correlation function‐ als are also shown to have clear physical meanings independent of the functionals, though they inevitably require minimum semiempirical parameters dependent on the functionals combined. We, therefore, interpret that DFT functionals with physical corrections are the most sophisticated target functions that are physically legitimated, even from the viewpoint of data science.     

Combining convolutional neural networks and in‐line near‐infrared spectroscopy for real‐time monitoring of the chromatographic elution process in commercial production of notoginseng total saponins

The chromatographic elution process is a key step in the production of notoginseng total saponins. Due to quality variability of loading samples and resin capacity decreasing over cycle time, saponins, especially the five main saponins of notoginseng total saponins, need to be monitored in real time during the elution process. In this study, convolutional neural networks, one of the most popular deep learning methods, were used to develop quantitative calibration models based on in‐line near‐infrared spectroscopy for notoginsenoside R₁, ginsenosides Rg₁, Re, Rb₁ and Rd, and their sum concentration, with root mean square error of prediction values of 0.87, 2.76, 0.60, 1.57, 0.28, and 4.99 mg/mL, respectively. Partial least squares calibration models were also developed for model performance comparison. Results show predicted concentration profiles outputted by both the convolutional neural network models and partial least squares models show agreements with the real trends defined by reference measurements, and can be used for elution process monitoring and endpoint determination. To the best of our knowledge, this is the first reported case study of combining convolutional neural networks and in‐line near‐infrared spectroscopy for monitoring of the chromatographic elution process in commercial production of botanical drug products.     

Machine-Learning Adsorption on Binary Alloy Surfaces for Catalyst Screening

Over the last few years, machine learning is gradually becoming an essential approach for the investigation of heterogeneous catalysis. As one of the important catalysts, binary alloys have attracted extensive attention for the screening of bifunctional catalysts. Here we present a holistic framework for machine learning approach to rapidly predict adsorption energies on the surfaces of metals and binary alloys. We evaluate different machine-learning methods to understand their applicability to the problem and combine a tree-ensemble method with a compressed-sensing method to construct decision trees for about 60000 adsorption data. Compared to linear scaling relations, our approach enables to make more accurate predictions lowering predictive root-mean-square error by a factor of two and more general to predict adsorption energies of various adsorbates on thousands of binary alloys surfaces, thus paving the way for the discovery of novel bimetallic catalysts.     

医学影像处理的深度学习可解释性研究进展

随着医学影像数据的迅速增长,传统的影像分析方法给医生带来巨大挑战。利用计算机视觉技术提供自动或半自动辅助诊断,可大大缓解人工阅片压力,提高诊断的准确性,促进医疗流程的标准化建设等。目前,深度学习卷积神经网络在医学影像处理中已取得不俗表现,但深度学习“黑匣子”的不可解释性阻碍了智能医疗诊断的发展。为增强对医学影像数据处理的深度学习可解释性的了解,对近几年相关研究进展进行了综述。首先,综述了深度学习在医学领域的应用现状及面临的问题,对神经网络的可解释性内涵进行了讨论;然后,从现有深度学习可解释性的常见方法出发,重点讨论了医学影像处理的深度学习可解释性研究进展;最后,探讨了医学影像处理的深度学习可解释性的发展趋势。     

基于联合神经网络学习的中文电力计量命名实体识别

随着电力计量业务的不断扩展,迫切需要由业务信息、技术知识、行业标准及其内在联系所组成的电力计量知识图谱,为电网的决策和发展提供更为全面有效的支持。命名实体识别是构建知识图谱的基础。针对电力计量领域需要,结合中文分词技术特点,基于联合学习思想,提出了一种基于联合学习的中文电力计量命名实体识别技术。该技术联合CNN-BLSTM-CRF模型与整合词典知识的分词模型,使其共享实体类别和置信度;同时将2个模型的先后计算顺序改为并行计算,减少了识别误差累积。结果表明,在不需要人工构建特征的情况下,方法的正确率、召回率、F值等均显著优于以往方法。     

基于快速神经网络架构搜索的鲁棒图像水印网络算法

为解决深度学习在图像水印算法中计算量大且模型冗余的问题,提高图像水印算法在抵抗噪声、旋转和剪裁等攻击时的鲁棒性,提出基于快速神经网络架构搜索（neural architecture search,NAS）的鲁棒图像水印网络算法。通过多项式分布学习快速神经网络架构搜索算法,在预设的搜索空间中搜索最优网络结构,进行图像水印的高效嵌入与鲁棒提取。首先,将子网络中线性连接的全卷积层设置为独立的神经单元结构,并参数化表示结构单元内节点的连接,预先设定结构单元内每个神经元操作的搜索空间;其次,在完成一个批次的数据集训练后,依据神经元操作中的被采样次数和平均损失函数值动态更新概率;最后,重新训练搜索完成的网络。水印网络模型的参数量较原始网络模型缩减了92%以上,大大缩短了模型训练时间。由于搜索得到的网络结构更为紧凑,本文算法具有较高的时间性能和较好的实验效果,在隐藏图像时,对空域信息的依赖比原始网络更少。对改进前后的2个网络进行了大量鲁棒性实验,对比发现,本文算法在CIFAR-10数据集上对抵抗椒盐噪声和旋转、移除像素行（列）等攻击优势显著;在ImageNet数据集上对抵抗椒盐高斯噪声、旋转、中值滤波、高斯滤波、JPEG压缩、裁剪等攻击优势显著,特别是对随机移除行（列）和椒盐噪声有较强的鲁棒性。     

游客情感计算的文本大数据挖掘方法比较研究

旅游文本大数据以其方便、快捷和低门槛的特点为游客情感计算提供了极大便利,已经成为旅游大数据的主要来源之一。基于大数据理论和情感理论,以文本大数据为数据源,在全面梳理国内外情感计算相关成果的基础上,利用人工智能中的逻辑/算法编程方法、机器学习方法、深度学习方法对旅游文本大数据进行挖掘,探索最佳的基于文本大数据的游客情感计算方法。研究发现：（1）基于情感词典的游客情感计算模型,其核心是构建情感词典和设计情感计算规则,方法简单,容易实现,适用语料范围广。（2）机器学习,用统计学方法抽取文本中的特征项,具有非线性特征,可靠性较线性特征的情感词典方法高。（3）基于深度学习技术的游客情感计算,效果良好,准确率在85%以上。训练多领域的文本语料易于移植,实用性强,且泛化能力好,较适合大数据时代游客情感计算研究。     

融合图卷积神经网络和注意力机制的PM2.5小时浓度多步预测

PM_2.5小时浓度多为单步预测。为实现PM_2.5小时浓度的多步预测,基于“编码器-解码器”的序列-序列预测（Seq2Seq）模型,集合图卷积神经网络提取非欧式空间数据特征的能力以及注意力机制自适应关注特征的能力,提出了融合图卷积神经网络和注意力机制的PM_2.5小时浓度多步预测（GCN_Attention_Seq2Seq）模型。并与Seq2Seq模型和使用了图卷积神经网络、未使用注意力机制的GCN_Seq2Seq模型进行了对照,以2015—2016年北京市22个空气质量监测站点的空气质量数据为样本进行实例验证,结果表明,Seq2Seq模型和图卷积神经网络（GCN）可对PM_2.5小时浓度数据的时空依赖进行有效建模,注意力机制有助于减缓多步预测中的预测精度衰减,提升PM_2.5小时浓度多步预测的精度。GCN_Attention_Seq2Seq模型可有效应用于多种长度的PM_2.5浓度预测窗口。     

化学知识类型与学习方式选择的探讨

基础教育新一轮课程改革是教育领域里的一场深刻革命,它呼唤着学生学习方式的根本转变。根据现代认知心理学的研究成果与化学学科内容的特点,我们将化学知识分为系统-演绎化学知识与经验-缄默化学知识2大类,每一类又可划分为不同的类型,化学知识的类型影响着化学学习方式的选择。     

平面波超声成像中的波束合成方法研究进展

平面波成像通过单次全孔径发射-接收即可获取整幅图像,将成像帧频显著地提升至1000帧/秒以上.然而,平面波成像过程中发射的非聚焦波束将导致回波信号信噪比降低,进而使图像的分辨率和对比度变差.通过多角度相干复合成像技术可以改善平面波成像的图像质量,但是会以牺牲帧频为代价.因此研究人员们开始将新型波束合成技术引入平面波成像...