多示例学习问题研究进展综述

多示例学习是一种特殊的机器学习问题,近年来得到了广泛的关注和研究,许多不同类型的多示例学习算法被提出,用以处理各个领域中的实际问题. 针对多示例学习的算法研究和应用进行了较为详细的综述, 介绍了多示例学习的各种背景假设, 从基于示例水平、包水平、嵌入空间三个方面对多示例学习的常见算法进行了描述, 并给出了多示例学习的算法拓展和若干领域的主要应用.     

One- and two-channel Kondo model with logarithmic Van Hove singularity: A numerical renormalization group solution

Simple scaling consideration and NRG solution of the one- and two-channel Kondo model in the presence of a logarithmic Van Hove singularity at the Fermi level is given. The temperature dependences of local and impurity magnetic susceptibility and impurity entropy are calculated. The low-temperature behavior of the impurity susceptibility and impurity entropy turns out to be non-universal in the Kondo sense and independent of the s–d coupling J. The resonant level model solution in the strong coupling regime confirms the NRG results. In the two-channel case the local susceptibility demonstrates a non-Fermi-liquid power-law behavior.     

共振线极化光实现原子矢量磁力仪的理论研究

共振线偏振光激发原子张量磁矩,本文理论研究在矢量磁场和射频场的共同作用下,张量磁矩进动的模型,求解刘维尔方程获得透射光时域完整解析解,包括直流、一次和二次谐波分量.研究发现:当进动的拉比频率Ω1/(22~(1/2))时,两谐波间的干涉效应使直流分量和一次谐波对称成分的单吸收峰劈裂成双峰,裂距((Ω~2+Ω~4-Ω~2-1)~(3/2))~(1/2),一次谐波反对称成分在共振处产生干涉条纹.研究结果显示,谐波间的干涉也可导致直流分量和二次谐波线宽仅为一次谐波线宽的38%,且存在磁场取向临界点,在不同的取向区间分别利用直流及两谐波共振信号辨析磁场变化,可获得最佳测磁灵敏度;同时还可通过共振时直流分量及两谐波对称成分振幅来确定磁场与激光极化方向的夹角,利用两谐波反对称成分相移的差值来确定待测磁场在垂直光极化方向投影与射频场方向的夹角,进而实现结构简单的张量磁矩进动型矢量磁力仪.这种磁力仪适合构成磁力仪阵列,可用于磁定位、水下磁异常源的检测和地磁导航等领域.     

基于炉口辐射光谱支持向量机回归的转炉终点碳含量检测

转炉终点碳含量在线检测对实现炼钢终点准确控制,提高钢铁产品质量,降低能耗,减少废气排放具有重要意义。针对转炉冶炼终点控制及碳含量检测的难题,研究了一种新的基于炉口辐射光谱分析的非接触式在线碳含量检测方法。方法基于辐射光谱的支持向量机回归(SVR)实现转炉终点前过程的碳含量预测。通过远距离光谱采集系统获取炉口火焰光谱信息,基于冶炼过程炉口火焰辐射光谱变化规律的分析,分别提取了表征辐射光谱整体特征的两个参数即总谱宽和辐射峰值、以及表征发射谱的三个特征波长600,630和775 nm处的幅值作为支持向量机的输入,结合脱碳理论和实测碳值拟合重构的脱碳函数曲线作为支持向量机的输出,利用支持向量机回归方法建立光谱分布与碳含量的关系模型。通过训练样本集和测试集循环优化确定模型最佳参数。设计的仪器和优选的模型已安装在转炉生产现场长时间运行,现场实验结果表明,终点碳含量检测准确率为90.2%,测量时间小于0.3 s,可实时在线检测,能够满足生产需求,为转炉冶炼终点的精确控制提供了重要依据。     

DPLS和SVM的掺假花椒粉近红外光谱定性鉴别

花椒是我国的八大调味料之一。目前花椒市场掺假现象较为多见,为实现掺假花椒粉的快速定性鉴别,采用了近红外光谱结合化学计量学方法进行了探讨。将麦麸粉、稻糠粉、玉米粉和松香粉以1 Wt/Wt.%的递增梯度分别掺入红花椒粉和青花椒粉中,制备掺假浓度范围为1~54 Wt/Wt.%的掺假花椒粉样品,以掺假花椒粉和纯花椒粉共462份样品依次采集其800~2 500 nm范围的漫反射近红外光谱。采用主成分分析法(PCA)对光谱数据进行分析,前3个主成分累计贡献率达98.72%,做出的得分图表明PCA法对掺假的花椒粉具有较好的区域划分。347份样本作为校正集,以特征谱区2 000~2 200 nm范围的257个采样点的光谱信号作为输入,采用判别偏最小二乘法(DPLS)和支持向量机(SVM)建立定性鉴别模型,经不同光谱预处理,对115份验证集样本进行预测,总体鉴别正确率在97.39%~100%之间,表明该方法是快速定性鉴别掺假花椒粉的一个有效手段。     

Nondestructive testing of external defects on Nanguo pear

In view of the common pests and diseases and irregularly shaped fruits of Nanguo pear, this paper fused the spectral information and image features to realize the rapid nondestructive testing and recognition of the external defects on Nanguo pear by hyperspectral imaging technology. Backpropagation neural network and support vector machine model was established to identify external defects, which are commonly used in classification and pattern recognition. The testing results show that recognition effect of support vector machine is better than backpropagation neural network. Among them, the recognition accuracy of fruits damaged by insects and rotten fruits of Nanguo pears reaches 100%. This study provides a theoretical basis for developing online grading system and quality detection of Nanguo pear based on multispectral imaging technique.     

An energy sum constraint of event mixing for Bose-Einstein correlations in reactions with ππX final states around 1 GeV

We present a new event mixing technique for measuring two-pion Bose-Einstein correlations(BEC)in reactions with only two identical bosons among three final state particles.This new mixing method contains a missing mass consistency(MMC)cut and an energy sum order(ESO)cut.Unlike the previous proposed pion energy cut,which abandons nearly half the original events,the ESO cut does not eliminate any original events and hence improves the statistics of both original events and mixed events.Numerical tests using theγp→π~0π~0p events around1 Ge V are carried out to verify the validity of the ESO cut.This cut is able to reproduce the relative momentum distribution of the original events in the absence of BEC effects.In addition,its ability to observe BEC effects is tested by an event sample in the presence of BEC effects.Simulation results show the BEC effects can be observed clearly as an enhancement in the correlation function,and the BEC parameters extracted by this event mixing cut are consistent with the input BEC parameters.     

互连网络的向量图模型

n-超立方体,环网,k元n超立方体,Star网络,煎饼(pancake)网络,冒泡排序(bubble sort)网络,对换树的Cayley图,De Bruijn图,Kautz图,Consecutive-d有向图,循环图以及有向环图等已被广泛的应用做处理机或通信互连网络.这些网络的性能通常通过它们的度,直径,连通度,hamiltonian性,容错度以及路由选择算法等来度量.在本文中,首先,我们提出了有向向量图和向量图的概念;其次,我们开发了有向向量图模型和向量图模型来更好地设计,分析,改良互连网络;我们进一步证明了上述各类著名互连网络都可表示为有向向量图模型或向量图模型;更重要的是该模型能够使我们设计出了新的互连网络---双星网络和三角形网络.     

Flexible Manipulation of the Polarization Conversions in a Structured Vector Field in Free Space

The conversion between different polarizations generally requires an optical element or an optical system. We demonstrate, theoretically and experimentally, that the conversion between different states of polarization (SoP) can be flexibly manipulated in a structured optical field by itself. During the structured optical field propagation in free space, any two orthogonal polarizations can be generated and auto‐focus at any desired propagation positions by purposely designing the initial SoP distribution and a caustic phase in the field cross‐section. The experiment arrangement combines a caustic‐based approach and a 4f optical system. This approach provides a way to flexibly manipulate the polarization conversions in free space propagation by the optical field itself (without any physical element along the propagation path), may lead to more complex and flexible vectorial manipulations of an optical field scenarios and potential applications in corresponding areas.     

反应物分子初始振动激发对H+CH~+→C~++H_2反应的影响

在CH_2~+体系的电子基态势能面上运用准经典轨线方法,研究了当碰撞能E=500 me V时,反应物分子的振动激发对H(~2S)+CH~+(X~1Σ~+)→C+(~2P)+H_2(X1_gΣ~+)反应的反应概率、反应截面和立体动力学性质的影响.分别计算了两矢量相关k-j′的P(θ_r)分布,三矢量相关k-k′-j′的P(φr)分布以及反应产物的四个极化微分截面.结果表明,产物分子转动角动量不仅在Y轴方向有取向效应,还定于Y轴的正方向.并且发现,随着振动量子数的增加,对反应体系产物分布的影响就越明显.