基于双重权重偏差建模的无监督域适应

无监督域适应(Unsupervised Domain Adaptation,UDA)是一类新兴的机器学习范式,其通过对源域知识在无标记目标域上的迁移利用,来促进目标域模型的训练。为建模源域与目标域之间的域分布差异,最大均值差异(Maximum Mean Discrepancy,MMD)建模被广泛应用,其对UDA的性能提升起到了有效的促进作用。然而,这些方法通常忽视了领域之间对应类规模与类分布等结构信息,因为目标域与源域的数据类规模与数据分布通常并非一致。为此,文中提出了一种基于跨域类和数据样本双重加权的无监督域适应模型(Sample weighted and Class weighted based Unsupervised Domain Adaptation Network,SCUDAN)。具体而言,一方面,通过源域类层面的适应性加权来调整源域类权重,以实现源域与目标域之间的类分布对齐;另一方面,通过目标域样本层面的适应性加权来调整目标域样本权重,以实现目标域与源域类中心的对齐。此外,文中还提出了一种CEM(Classification Expectation Maximization)优化算法,以实现对SCUDAN的优化求解。最后,通过对比实验和分析,验证了所提模型和算法的有效性。     

动态迁移实体块信息的跨领域中文实体识别模型

由于中文文本之间没有分隔符,难以识别中文命名实体的边界.此外,在垂直领域中难以获取充足的标记完整的语料,例如医疗领域和金融领域等垂直领域.为解决上述不足,提出一种动态迁移实体块信息的跨领域中文实体识别模型(TES-NER),将跨领域共享的实体块信息(entity span)通过基于门机制(gate mechanism)的动态融合层,从语料充足的通用领域(源领域)动态迁移到垂直领域(目标领域)上的中文命名实体模型,其中,实体块信息用于表示中文命名实体的范围.TES-NER模型首先通过双向长短期记忆神经网络(BiLSTM)和全连接网络(FCN)构建跨领域共享实体块识别模块,用于识别跨领域共享的实体块信息以确定中文命名实体的边界;然后,通过独立的基于字的双向长短期记忆神经网络和条件随机场(BiLSTM-CRF)构建中文命名实体识别模块,用于识别领域指定的中文命名实体;最后构建动态融合层,将实体块识别模块抽取得到的跨领域共享实体块信息通过门机制动态决定迁移到领域指定的命名实体识别模型上的量.设置通用领域(源领域)数据集为标记语料充足的新闻领域数据集(MSRA),垂直领域(目标领域)数据集为混合领域(OntoNotes 5.0)、金融领域(Resume)和医学领域(CCKS 2017)这3个数据集,其中,混合领域数据集(OntoNotes 5.0)是融合了6个不同垂直领域的数据集.实验结果表明,提出的模型在OntoNotes 5.0、Resume和CCKS 2017这3个垂直领域数据集上的F1值相比于双向长短期记忆和条件随机场模型(BiLSTM-CRF)分别高出2.18%、1.68%和0.99%.     

Box-Jenkins identification revisited—Part I: Theory

In classical time domain Box-Jenkins identification discrete-time plant and noise models are estimated using sampled input/output signals. The frequency content of the input/output samples covers uniformly the whole unit circle in a natural way, even in case of prefiltering. Recently, the classical time domain Box-Jenkins framework has been extended to frequency domain data captured in open loop. The proposed frequency domain maximum likelihood (ML) solution can handle (i) discrete-time models using data that only covers a part of the unit circle, and (ii) continuous-time models. Part I of this series of two papers (i) generalizes the frequency domain ML solution to the closed loop case, and (ii) proves the properties of the ML estimator under non-standard conditions. Contrary to the classical time domain case it is shown that the controller should be either known or estimated. The proposed ML estimators are applicable to frequency domain data as well as time domain data.     

Modeling application domains

Modeling the interactions among the objects in an application domain and the activities in the domain are necessary extensions of modeling the structure and behaviour of objects in the domain. Object models capture the structure and behaviour of components in the application domains. Interaction models describe how the components in the domain interact to realize the activities in the domain. Activity models are used to describe the functionality of the domain. The integration of object, interaction and activity models is essential for producing complete domain models which we refer to as Component-Interaction-Activity (CIA) models. The CIA model of a domain simplifies the development of applications in that domain because it makes extensive reuse of the elements of the domain possible. Concepts for developing the CIA models of application domains are described. A Domain Modeling Language (DML) which provides constructs for expressing the concepts in CIA models is introduced. The DML provides: (i) a class construct to define components, (ii) rule, trigger, event and relationship constructs to define interactions among components, and (iii) process and task constructs to define the activities in the domain. A Domain Modeling Tool which provides a graphic interface to DML and makes the CIA domain models expressed in DML executable is introduced. An example of using CIA models in developing applications is presented.     

VPN中的分布式访问控制

针对VPN系统中的分布式访问控制及其管理问题，分析了现有的IETF模型的不足，提出一种分布式管理的访问控制模型。该模型将全局策略分解为局部(SubDomain，每个SubDomain对应一个网关)策略数据库的集合，同时在IETF的模型中增加策略判决点，以转发用户认证请求。并通过VPN的加密隧道来保护VPN系统安全策略传输过程和用户认证数据的完整性、机密性。本文最后给出了基于CORBA的原型来说明该模型的工作模式，实战证明，该模型能有效解决安全VPN中的策略的分布式管理和用户的漫游问题。     

Decision support for the software product line domain engineering lifecycle

Software product line engineering is a paradigm that advocates the reusability of software engineering assets and the rapid development of new applications for a target domain. These objectives are achieved by capturing the commonalities and variabilities between the applications of the target domain and through the development of comprehensive and variability-covering feature models. The feature models developed within the software product line development process need to cover the relevant features and aspects of the target domain. In other words, the feature models should be elaborate representations of the feature space of that domain. Given that feature models, i.e., software product line feature models, are developed mostly by domain analysts by sifting through domain documentation, corporate records and transcribed interviews, the process is a cumbersome and error-prone one. In this paper, we propose a decision support platform that assists domain analysts throughout the domain engineering lifecycle by: (1) automatically performing natural language processing tasks over domain documents and identifying important information for the domain analysts such as the features and integrity constraints that exist in the domain documents; (2) providing a collaboration platform around the domain documents such that multiple domain analysts can collaborate with each other during the process using a Wiki; (3) formulating semantic links between domain terminology with external widely used ontologies such as WordNet in order to disambiguate the terms used in domain documents; and (4) developing traceability links between the unstructured information available in the domain documents and their formal counterparts within the formal feature model representations. Results obtained from our controlled experimentations show that the decision support platform is effective in increasing the performance of the domain analysts during the domain engineering lifecycle in terms of both the coverage and accuracy measures.     

多核局部领域适应学习

领域适应(或跨领域)学习旨在利用源领域(或辅助领域)中带标签样本来学习一种鲁棒的目标分类器,其关键问题在于如何最大化地减小领域间的分布差异.为了有效解决领域间特征分布的变化问题,提出一种三段式多核局部领域适应学习(multiple kernel local leaning-based domain adaptation,简称MKLDA)方法:1)基于最大均值差(maximum mean discrepancy,简称MMD)度量准则和结构风险最小化模型,同时,学习一个再生多核Hilbert空间和一个初始的支持向量机(support vector machine,简称SVM),对目标领域数据进行初始划分;2)在习得的多核Hilbert空间,对目标领域数据的类别信息进行局部重构学习;3)最后,利用学习获得的类别信息,在目标领域训练学习一个鲁棒的目标分类器.实验结果显示,所提方法具有优化或可比较的领域适应学习性能.     

跨域引用监视器及其以数据为中心的多级安全模型

为基于不可信计算机系统来构建一个可信的多级安全(MLS)大系统,提出一种新型的跨域引用监视器及其多级安全模型。该跨域引用监视器采用现有的商业现货(COTS)产品,使用一个或多个独立的计算机,在两个或多个不同的网络之间,通过满足EAL7的单向传输硬件装置来连接。基于该跨域监视器实现了以数据为中心的多级安全模型。该模型允许信息从低密级网络流向高密级网络,也允许高密级网络把低密级数据发布给低密级网络,禁止高密级网络的高密级信息和无密级标记信息流向低密级网络,并已在分级保护的网络系统中成功应用。通过安全模型和安全策略的形式化描述和证明表明,基于该安全模型构建可信MLS大系统是可行的。     

领域适应核支持向量机

领域适应学习是一种新颖的解决先验信息缺少的模式分类问题的有效方法, 最大化地缩小领域间样本分布差是领域适应学习成功的关键因素之一,而仅考虑领域间分布均值差最小化, 使得在具体领域适应学习问题上存在一定的局限性.对此,在某个再生核Hilbert空间, 在充分考虑领域间分布的均值差和散度差最小化的基础上,基于结构风险最小化模型, 提出一种领域适应核支持向量学习机(Kernel support vector machine for domain adaptation, DAKSVM)及其最小平方范式,人造和实际数据集实验结果显示,所提方法具有优化或可比较的模式分类性能.     

主题方面共享的领域主题层次模型

层次主题模型是构建主题层次的重要工具. 现有的层次主题模型大多通过在主题模型中引入nCRP构造方法, 为文档主题提供树形结构的先验分布, 但无法生成具有明确领域涵义的主题层次结构, 即领域主题层次. 同时, 领域主题不仅存在层次关系, 而且不同父主题下的子主题之间还存在子领域方面共享的关联关系, 在现有主题关系研究中没有合适的模型来生成这种领域主题层次. 为了从领域文本中自动、有效地挖掘出领域主题的层次关系和关联关系, 在4个方面进行创新研究. 首先, 通过主题共享机制改进nCRP构造方法, 提出nCRP+层次构造方法, 为主题模型中的主题提供具有分层主题方面共享的树形先验分布; 其次, 结合nCRP+和HDP模型构建重分层的Dirichlet过程, 提出rHDP (reallocated hierarchical Dirichlet processes)层次主题模型; 第三, 结合领域分类信息、词语语义和主题词的领域代表性, 定义领域知识, 包括基于投票机制的领域隶属度、词语与领域主题的语义相关度和层次化的主题-词语贡献度; 最后, 通过领域知识改进rHDP主题模型中领域主题和主题词的分配过程, 提出结合领域知识的层次主题模型rHDP_DK (rHDP with domain knowledge), 并改进采样过程. 实验结果表明, 基于nCRP+的层次主题模型在评价指标方面均优于基于nCRP的层次主题模型(hLDA, nHDP)和神经主题模型(TSNTM); 通过rHDP_DK模型生成的主题层次结构具有领域主题层次清晰、关联子主题的主题词领域差异明确的特点. 此外, 该模型将为领域主题层次提供一个通用的自动挖掘框架.     

A model-based computer-aided prototyping system

Computer aided prototyping systems which allow application domain experts to compose prototypes using domain specific components, semantics, and primitives can reduce the time and effort needed for prototyping considerably.The notion of domain model is introduced as a vehicle for defining, generating, and categorizing domain specific components. Facilities to compose the components generated from the domain model to implement complex functionality are presented.A rapid prototyping system which provides a domain modeling tool and a visual direct manipulation composition tool is described. The main element of the domain modeling tool is a domain modeling language which provides: (i) a class construct to define object, (ii) rule, trigger, event, and relationship constructs to define interactions among objects, and (iii) process, task, and signal constructs to define the activities in the domain.The composition tool is the extension of a visual tool called IntelligentPad (IP) which supports unified data and control flows through a simple paste and connect metaphor. Domain models are translated to generate components, which are then composed using the extended IP.     

Optimal control theory: A generic tool for identification and control of (bio-)chemical reactors

In this review paper the potential of optimal control theory for optimization in the time as well as in the space domain is highlighted. Various case studies in the area of (bio-)chemical reactors are discussed ranging from the dual problem of performance optimization and accurate parameter identification (time domain) to plug flow reactor optimization (space domain). Furthermore, it is illustrated that application of the Minimum Principle of Pontryagin to distributed parameter systems leads to extremal control profile structures (in the space domain) which are very similar to those obtained during optimization (in the time domain) of well mixed bioreactors. The analogy is reflected at various levels during analytical optimal control computations.     

基于混合灵敏度优化的鲁棒特征结构配置

将H_∞理论的混合灵敏度指标融入到特征结构配置中, 将混合灵敏度指标转换为特征结构设计参数的函数关系, 建立了特征结构配置的频率域指标约束. 这两种方法的结合各自利用了其在时域和频率域的设计优势. 通过优化算法所获得的鲁棒特征结构配置, 既保证了闭环特征值/特征向量, 又使反馈回路具有一定的频率域特性. 计算实例表明这种综合设计方法的有效性.     

基于本体的领域需求分析方法与模型研究

领域需求分析是实施领域工程的重要内容,由于领域需求分析参与各方对领域知识理解的差异是领域需求分析的主要困难。该文提出了一种基于本体的领域需求分析方法,试图利用本体在知识表达方面形式化、规范化的优势统一领域内各方对知识的理解,最后给出了UML(统一建模语言)建立领域本体的实例与基于本体的领域需求分析的过程。     

基于频域、时域相结合的自适应图像超分辨率重建

基于最大后验概率(MAP)的超分辨率(SR)重建的研究重点是规则化项的选择,且其大都在频域中实现,为此提出一种基于频域、时域相结合的图像SR重建方法.首先,根据不同图像的特点,定义了频域规则化项(FR)和时域规则化项(TR);然后,给出了图像重建模型,引入频域、时域自适应权值来加强算法的自适应性;最后,运用共轭梯度法推导出重建迭代计算公式.实验表明,所提出的算法具有良好的收敛性和精确性.     

Dual geometric perception for cross-domain road segmentation

Road segmentation plays an important role in navigation systems and autonomous driving. However, many methods in road segmentation are based on supervised learning and suffer from performance degradation in the real world. There is a certain domain gap (distribution shift problem) between the source domain (training data) and the target domain (testing data). In this paper, we propose a Dual-Geometric Perception (DGP) approach for cross-domain road segmentation, which jointly uses semantic and dual-geometric information to learn the domain-invariant feature for road segmentation. First, we propose an RGB-N dual stream network structure, which effectively fuses normal vector information and RGB information to reduce domain gap. Moreover, a dual geometric adversarial learning strategy is proposed to utilize depth-aware and normal vector features to perform better domain alignment. Furthermore, a self-training learning strategy is used to further improve the model’s generalizability in the target domain. Extensive experiments demonstrate that our proposed DGP achieves superior performance on lane-to-lane and lane-to-sidewalk road domain adaptation tasks.     

异构复合迁移学习的视频内容标注方法

针对传统的机器学习需要大量的人工标注训练模型的弊端,以及目前多数迁移学习方法只适用于同构空间的问题,提出了一种异构复合迁移学习（HCTL）的视频内容标注方法。首先,借助视频与图像的对应关系,利用典型相关性分析（CCA）来实现图像域（源域）和视频域（目标域）特征空间的同构化;然后,基于这两个特征空间向共同空间投影的代价最小化这一思想,找到源域特征空间向目标域特征空间对齐的矩阵;最后,通过对齐矩阵使得源域特征能够翻译到目标域特征空间中去,进而实现知识迁移,完成视频内容标注任务。所提方法在Kodak数据库上的平均标注准确率达到了35.81%,与标准的支持向量机（S-SVM）领域适应支持向量机（DASVM）、异构直推式迁移学习（HTTL）、跨领域的结构化模型（CDSM）、领域选择机（DSM）、异构源域下的多领域适应（MDA-HS）和判别性相关分析（DCA）方法相比分别提高了58.03%、23.06%、45.04%、6.70%、15.52%、13.07%和6.74%;而在哥伦比亚用户视频（CCV）数据库上达到了20.73%,分别相对提高了133.71%、37.28%、14.34%、24.88%、16.40%、20.73%和12.48%。实验结果表明先同构再对齐的复合迁移思想在异构领域适应问题上能够有效地提升识别准确率。     

面向服务的领域需求分析方法及模型研究

面向特征方法是当前进行领域分析的基本方法,但是由于特征的抽象性,使得用面向特征方法所生成的领域模型在领域需求描述和向领域设计、领域实现的转换方面略显不足.提出了面向服务的领域需求分析方法,将Web服务结合到领域分析过程中,通过分析不同的领域需求来源,形成共性和个性领域服务,并以领域服务为中心构建领域需求模型,并在公共出行信息服务系统中进行应用.     

Web文本中维吾尔语领域术语的自动发现

针对维吾尔语领域术语获取难度大,人工扩充领域术语工作量大、效率低等特点,利用词汇共现原理,以维吾尔语连接词和互信息(MI)为工具,快速扩充原始维吾尔语领域术语;建立了以维吾尔语领域术语为特征模板,利用条件随机场(CRF)模型实现Web文本中维吾尔语领域术语的自动发现方法,并在此基础上实现长维吾尔语领域术语的自动发现。实验表明,对短维吾尔语领域术语的自动发现准确率为 97.59%,召回率为 93.38%,对长维吾尔语领域术语的自动发现正确率达到55.72%。     

基于单点多盒检测器的全局-局部层级的域适应目标检测

在目标检测领域里通常希望在拥有大量标记的场景中训练好的模型能够应用在无标记的其他场景中,但是不同的域分布往往是不同的,这样往往导致域迁移时模型性能的急剧下降.为了提高域迁移时模型的目标检测性能,通过两个层级来解决域迁移问题,包括全局层级迁移和局部层级迁移.这两种层级迁移分别对应不同的特征对齐方式,即全局层级采用选择性对...