基于改进CGANs的入侵检测方法研究

近年来,机器学习算法在入侵检测系统(IDS)中的应用获得越来越多的关注。然而,传统的机器学习算法更多的依赖于已知样本,因此需要尽可能多的数据样本来对模型进行训练。遗憾地是,随着越来越多未知攻击的出现,且用于训练的攻击样本具有不平衡性,传统的机器学习模型会遇到瓶颈。文章提出一种将改进后的条件生成对抗网络(CGANs)与深度神经网络(DNN)相结合的入侵检测模型(CGANs-DNN),通过解决样本不平衡性问题来提高检测模型对未知攻击类型或只有少数攻击样本类型的检测率。深度神经网络(DNN)具有表征数据潜在特征的能力,而经过改进后的条件CGANs,能够通过学习已知攻击样本潜在数据特征分布,来根据指定类型生成新的攻击样本。此外,与生成对抗网络(GANs)和变分自编码器(VAE)等无监督生成模型相比,CGANsDNN经过改进后加入梯度惩罚项,在训练的稳定性上有了很大地提升。通过NSL-KDD数据集对模型进行评估,与传统算法相比CGANs-DNN不仅在整体准确率、召回率和误报率等方面有更好的性能,而且对未知攻击和只有少数样本的攻击类型具有较高的检测率。     

视频侦查中人像智能分析应用及算法优化

人像智能分析指的是对视频或录像中的人像进行结构化和可视化分析,对目标人物进行性别、年龄、发型等特征的智能识别,这项技术在视频侦查中有极高的应用价值。人像识别早期的算法是通过人工提取特征,通过学习低级视觉特征来针对不同属性进行分类学习,这种基于传统方法的模型表现常常不尽如人意。在计算机视觉领域,通过海量图像数据学习的神经网络比传统方法有更丰富的信息量和特征可以被提取。文章尝试通过深度学习技术训练神经网络模型对行人进行检测和识别,对于衣着不同的行人进行智能识别,具有更好的鲁棒性,提升了视频人像识别的准确率,拓展了人工智能技术在身份识别领域的应用。     

基于深度学习的传感器优化布置方法

针对传感器优化布置(optimal sensor placement,简称OSP)问题,提出了一种新的使用深度神经网络的解决方案,并以简化的桥梁形状的桁架结构中的振动测试传感器优化为例进行了验证。首先,选择一种传统的传感器优化布置方法,对自动化生成的大量不同的桁架结构分别进行传感器优化布置计算,将所得优化布置结果在进行数据预处理后构建出深度学习方法所需要的训练集与验证集;其次,使用Python语言和深度学习框架TensorFlow设计实现与本研究问题适配的深度神经网络模型并训练;然后,随机生成了新的桁架结构参数;最后,将深度神经网络输出的传感器布置结果和传统方法的计算结果进行了比较,验证了本研究方法的有效性以及在速度上、可移植性与可扩展性方面的性能优势。     

基于卷积神经网络的葡萄叶片氮含量识别方法

为实现对葡萄叶片氮素含量快速、便捷的识别，在卷积神经网络VGG-16网络结构基础上，将数据增广后的图像按不同梯度划分进行模型训练，通过十折交叉验证法探究最佳的训练集与验证集分配比例，并构建4个不同深度的网络模型进行训练对比，采用全局平均池化代替全连接层约简网络参数量。训练结果表明，氮含量梯度设为0.70%、0.35%和0.175%时，室内简单背景识别准确率分别为85.9%、76.2%和71.1%；晴天室外复杂背景下识别准确率分别为44.6%、35.0%和30.4%。研究结果表明利用VGG-16建立的网络学习模型对葡萄叶片氮含量识别提供了一种新的便捷方法，对农业信息化和智能化技术应用具有一定促进作用。     

Learning bipartite Bayesian networks under monotonicity restrictions

ABSTRACT

Learning parameters of a probabilistic model is a necessary step in machine learning tasks. We present a method to improve learning from small datasets by using monotonicity conditions. Monotonicity simplifies the learning and it is often required by users. We present an algorithm for Bayesian Networks parameter learning. The algorithm and monotonicity conditions are described, and it is shown that with the monotonicity conditions we can better fit underlying data. Our algorithm is tested on artificial and empiric datasets. We use different methods satisfying monotonicity conditions: the proposed gradient descent, isotonic regression EM, and non-linear optimization. We also provide results of unrestricted EM and gradient descent methods. Learned models are compared with respect to their ability to fit data in terms of log-likelihood and their fit of parameters of the generating model. Our proposed method outperforms other methods for small sets, and provides better or comparable results for larger sets.     

神经网络全局势函数在多相催化中的应用

当今的多相催化研究需要新的技术和方法从原子尺度上表征活性中心结构和反应中间体。本文作者课题组近期开发了理论模拟新技术来探索催化剂活性位点结构，即基于神经网络势函数的大规模原子模拟（LASP）软件中实现的全局神经网络势函数计算方法。本文介绍了该方法可以显著降低催化体系的计算代价，而维持与密度泛函理论同一级别的计算精度，从而解决多相催化中的许多复杂问题。本文对神经网络势函数方法的实现细节和目前已实现的应用场景进行了详细介绍。神经网络势函数可以用来预测材料晶体结构，理解高压氢化条件下TiO₂表面的结构演化和确定三元氧化物ZnCrO晶相中合成气制甲醇活性位点。最后文章分析了神经网络势函数的局限性和今后可能的三个研究方向，即材料性质预测、多元素体系神经网络势函数构造和化学反应拟合。     

3D Microstructures of Liquid Crystal Networks with Programmed Voxelated Director Fields

The shape-shifting behavior of liquid crystal networks (LCNs) and elastomers (LCEs) is a result of an interplay between their initial geometrical shape and their molecular alignment. For years, reliance on either one-step in situ or two-step film processing techniques has limited the shape-change transformations from 2D to 3D geometries. The combination of various fabrication techniques, alignment methods, and chemical formulations developed in recent years has introduced new opportunities to achieve 3D-to-3D shape-transformations in large scales, albeit the precise control of local molecular alignment in microscale 3D constructs remains a challenge. Here, the voxel-by-voxel encoding of nematic alignment in 3D microstructures of LCNs produced by two-photon polymerization using high-resolution topographical features is demonstrated. 3D LCN microstructures (suspended films, coils, and rings) with designable 2D and 3D director fields with a resolution of 5 µm are achieved. Different shape transformations of LCN microstructures with the same geometry but dissimilar molecular alignments upon actuation are elicited. This strategy offers higher freedom in the shape-change programming of 3D LCN microstructures and expands their applicability in emerging technologies, such as small-scale soft robots and devices and responsive surfaces.     

A benchmark data set for two-phase Coriolis metering

For more than a decade there has been growing interest in the use of Coriolis mass flow metering applied to two-phase (gas/liquid) and multiphase (oil/water/gas) conditions. It is well-established that the mass flow and density measurements generated from multiphase flows are subject to large errors, and a variety of physical models and correction techniques have been proposed to explain and/or to compensate for these errors. One difficulty is the absence of a common basis for comparing correction techniques, because different flowtube designs and configurations, as well as liquid and gas properties, may result in quite different error curves. Furthermore, some researchers with interests in the modelling aspects of the field may not have suitable multiphase laboratory facilities to generate their own data sets. This paper offers a small data set that may be used by researchers as a benchmark i.e. a common data set for comparing correction techniques. The data set was collected at the UK National Flow Laboratory TUV-NEL, using air and a viscous oil, and provides experimental points over a wide flow range (8:1 turndown) and with Gas Volume Fraction (GVF) values up to 60%. As a first investigation using the benchmark data set, we consider how data sparsity (i.e. the flow rate and GVF spacing in the experimental grid) affects the accuracy of a correction model. A range of neural network models are evaluated, based on different subsets of the benchmark data set. The data set and some exemplary code are provided with the paper. Additional data sets are available on a web site created to support this initiative.     

改进的卷积神经网络单幅图像超分辨率重建

针对经典的基于卷积神经网络的单幅图像超分辨率重建方法网络较浅、提取的特征少、重建图像模糊等问题，提出了一种改进的卷积神经网络的单幅图像超分辨率重建方法，设计了由密集残差网络和反卷积网络组成的新型深度卷积神经网络结构。原始低分辨率图像输入网络，利用密集残差学习网络获取更丰富的有效特征并加快特征梯度流动，其次通过反卷积层将图像特征上采样到目标图像大小，再利用密集残差学习高维特征，最后融合不同卷积核提取的特征得到最终的重建图像。在Set5和Set14数据集上进行了实验，并和Bicubic、K-SVD、SelfEx、SRCNN等经典重建方法进行了对比，重建出的图像在整体清晰度和边缘锐度方面更好，另外峰值信噪比（PSNR）平均分别提高了2.69?dB、1.68?dB、0.74?dB和0.61?dB。实验结果表明，该方法能够获取更丰富的细节信息，得到更好的视觉效果，达到了图像超分辨率的增强任务。     

智融标识网络基础研究

现有互联网因其原始设计不足已在服务质量、资源利用率、管控能力等诸多方面暴露出严重弊端，难以支撑起"互联网+"新业态高速发展对通信网络与服务日趋复杂化、多元化的需求.因此，如何设计全新的互联网已成为信息领域最为迫切与核心的研究内容之一，受到学术界与产业界的高度重视.本文基于前期"标识网络"与"智慧标识网络"研究，创造性提出"智融标识网络"体系及其关键机制，通过全网多空间、多维度资源的智慧融合，实现个性化服务的按需供给与灵活化组网的有效支撑，为不同行业与用户提供高效的差异化、定制化通信网络服务，以普遍适用于高铁网络、工业互联网、车联网等各类迥异应用情景.