人工智能追求的长期目标是使机器能像人一样学习和思考。由于人类面临的许多问题具有不确定性、脆弱性和开放性，任何智能程度的机器都无法完全取代人类，这就需要将人的作用或人的认知模型引入到人工智能系统中，形成混合-增强智能的形态，这种形态是人工智能或机器智能的可行的、重要的成长模式。混合-增强智能可以分为两类基本形式：一类是人在回路的人机协同混合增强智能，另一类是将认知模型嵌入机器学习系统中，形成基于认知计算的混合智能。本文讨论人机协同的混合-增强智能的基本框架，以及基于认知计算的混合-增强智能的基本要素：直觉推理与因果模型、记忆和知识演化；特别论述了直觉推理在复杂问题求解中的作用和基本原理，以及基于记忆与推理的视觉场景理解的认知学习网络；阐述了竞争-对抗式认知学习方法，并讨论了其在自动驾驶方面的应用；最后给出混合-增强智能在相关领域的典型应用。

鸡尾酒会问题即在多人同时说话的场景下追踪并识别某一个特定说话人的语音。在自动语音识别技术大规模推广应用中，鸡尾酒会问题是亟待解决的关键问题之一。本文回顾了在过去20多年中针对鸡尾酒会问题提出的相关技术。主要讨论在鸡尾酒会问题中扮演中心角色的语音分离问题。介绍了以下内容：传统的单通道情况下的技术，如计算听觉场景分析（computational auditory scene analysis, CASA）、非负矩阵分解（non-negative matrix factorization, NMF）以及生成式模型建模；传统的多通道情况下的技术，如波束成形和多通道盲源分离；一些基于深度学习的最新技术，如深度聚类（deep clustering, DPCL）、深度吸引网络（deep attractor network, DANet）以及排列不变性训练（permutation invariant training, PIT）。此外，介绍了在鸡尾酒会环境下针对改善多说话人语音识别和说话人识别精度的相关技术。笔者认为，利用一个更加强大的模型来有效地开发和利用来自麦克风阵列、声学训练集合以及语言本身的知识非常重要。更好的优化策略和技术的提出会逐步解决鸡尾酒会问题。

In some image classification tasks, similarities among different categories are different and the samples are usually misclassified as highly similar categories. To distinguish highly similar categories, more specific features are required so that the classifier can improve the classification performance. In this paper, we propose a novel two-level hierarchical feature learning framework based on the deep convolutional neural network (CNN), which is simple and effective. First, the deep feature extractors of different levels are trained using the transfer learning method that fine-tunes the pre-trained deep CNN model toward the new target dataset. Second, the general feature extracted from all the categories and the specific feature extracted from highly similar categories are fused into a feature vector. Then the final feature representation is fed into a linear classifier. Finally, experiments using the Caltech-256, Oxford Flower-102, and Tasmania Coral Point Count (CPC) datasets demonstrate that the expression ability of the deep features resulting from two-level hierarchical feature learning is powerful. Our proposed method effectively increases the classification accuracy in comparison with flat multiple classification methods.

We propose a self-adaptive process (SAP) that maintains the software architecture quality using the MAPE-K standard model. The proposed process can be plugged into various software development processes and service-oriented methodologies due to its explicitly defined inputs and outputs. To this aim, the proposed SAP is integrated with the service-oriented modeling and application (SOMA) methodology in a two-layered structure to create a novel methodology, named self-adaptive service-oriented architecture methodology (SASOAM), which provides a semi-automatic self-aware method by the composition of architectural tactics. Moreover, the maintenance activity of SOMA is improved using architectural and adaptive patterns, which results in controlling the software architecture quality. The improvement in the maintainability of SOMA is demonstrated by an analytic hierarchy process (AHP) based evaluation method. Furthermore, the proposed method is applied to a case study to represent the feasibility and practicality of SASOAM.

随着互联网的负载高速增长和各种应用的日益丰富，为提供方便、快捷的互联网接入服务，现有的网络结构已经变得相当复杂和多样，这给网络管理提出了新的挑战。为了应对这一挑战，软件定义网络的概念被提出，并被认为是一种有前景的解决方案。在软件定义网络结构中，网络控制域和数据传输域是分开的；所有的网络设备被抽象化，通过一个集中式的控制器来进行管理。通过使用软件定义网络的结构，网络管理变得灵活和高效，因此可以潜在地提升网络性能。为了充分利用软件定义网络的优势，最近又提出了软件定义无线网络结构。本文首先分析在不同无线网络中软件定义网络结构的应用。然后讨论在基于软件定义网络结构的无线网络中使性能提升的几个关键技术环节。最后，提出一些关于软件定义无线网络可能的未来研究方向。

在过去几年，一些智能制造概念相继被提出，如云制造、工业4.0以及工业互联网。它们共同目的之一是通过构建汇聚资源的平台，实现跨企业资源协同优化配置。在所有这些概念中，一个完整制造系统包括分布式物理制造系统和一个包含从物理制造系统映射的虚拟制造系统平台。我们称这样的制造系统为平台型智能制造系统。因此，一个平台型智能制造系统可看作一个规模巨大的信息物理系统，其中信息部分是平台而物理部分是相应物理制造系统。对一个平台型智能制造系统而言，一个重要问题是如何实现汇聚资源的优化调度。多智能体技术为该问题的解决提供了一种有效方法。本文提出一个面向平台型智能制造系统调度的多智能体架构，包括平台层次的调度多智能体系统和企业层次的调度多智能体系统。提出平台型智能制造系统调度的流程、特征和需求。基于上述架构，提出一个面向平台型智能制造系统的调度模型。通过案例，验证了所提架构和模型的有效性。

针对固有时间尺度分解算法的模态混叠问题和最小二乘支持向量机的参数优化问题，本文提出了一种新的基于完备集合固有时间尺度分解和混合差分进化和粒子群算法优化最小二乘支持向量机的柴油机故障诊断方法。该方法主要包括以下几个步骤：首先，为解决固有时间尺度分解算法的模态混叠问题，提出了一种完备集合固有时间尺度分解算法。随后，利用完备集合固有时间尺度分解算法将非平稳的柴油机振动信号分解为一系列平稳的旋转分量和残差信号。然后，提取了前几阶旋转分量的三类典型的时频特征，包括奇异值、旋转分量能量和能量熵、AR模型参数，作为故障特征。最后，提出了混合差分进化和粒子群算法对最小二乘支持向量机的参数进行优化的方法，并通过将故障特征输入训练好的最小二乘支持向量机模型实现故障诊断。仿真和实验结果表明提出的故障诊断方法可以克服固有时间尺度分解的模态混叠问题，而且能够准确的识别柴油机故障。

Controller area networks (CANs) have been designed for multiplexing communication between electronic control units (ECUs) in vehicles and many high-level industrial control applications. When a CAN bus is overloaded by a large number of ECUs connected to it, both the waiting time and the error probability of the data transmission are increased. Thus, it is desirable to reduce the CAN frame length, since the duration of data transmission is proportional to the frame length. In this paper, we present a CAN message compression method to reduce the CAN frame length. Experimental results indicate that CAN transmission data can be compressed by up to 81.06% with the proposed method. By using an embedded test board, we show that 64-bit engine management system (EMS) CAN data compression can be performed within 0.16 ms; consequently, the proposed algorithm can be successfully used in automobile applications.

The objective of this research is the rapid reconstruction of ancient buildings of historical importance using a single image. The key idea of our approach is to reduce the infinite solutions that might otherwise arise when recovering a 3D geometry from 2D photographs. The main outcome of our research shows that the proposed methodology can be used to reconstruct ancient monuments for use as proxies for digital effects in applications such as tourism, games, and entertainment, which do not require very accurate modeling. In this article, we consider the reconstruction of ancient Mughal architecture including the Taj Mahal. We propose a modeling pipeline that makes an easy reconstruction possible using a single photograph taken from a single view, without the need to create complex point clouds from multiple images or the use of laser scanners. First, an initial model is automatically reconstructed using locally fitted planar primitives along with their boundary polygons and the adjacency relation among parts of the polygons. This approach is faster and more accurate than creating a model from scratch because the initial reconstruction phase provides a set of structural information together with the adjacency relation, which makes it possible to estimate the approximate depth of the entire structural monument. Next, we use manual extrapolation and editing techniques with modeling software to assemble and adjust different 3D components of the model. Thus, this research opens up the opportunity for the present generation to experience remote sites of architectural and cultural importance through virtual worlds and real-time mobile applications. Variations of a recreated 3D monument to represent an amalgam of various cultures are targeted for future work.

Global illumination is the core part of photo-realistic rendering. The photon mapping algorithm is an effective method for computing global illumination with its obvious advantage of caustic and color bleeding rendering. It is an active research field that has been developed over the past two decades. The deficiency of precise details and efficient rendering are still the main challenges of photon mapping. This report reviews recent work and classifies it into a set of categories including radiance estimation, photon relaxation, photon tracing, progressive photon mapping, and parallel methods. The goals of our report are giving readers an overall introduction to photon mapping and motivating further research to address the limitations of existing methods.

为了针对不同的应用和流量提供服务质量保障和差异化服务，负载均衡和多优先级队列技术被广泛地应用于网络中。在传统网络中，网络管理员经常使用“traceroute”和“ping”工具来检测负载均衡机制或者服务质量策略是否正常工作。然而，由于这些工具并不被现有的OpenFlow交换机所支持，所以还不能够应用于软件定义网络中。此外，traceroute和ping依靠主动发送探测包来探测路径。然而，当负载均衡机制把探测包和所需追踪流的数据包均衡到不同路径时，这些工具将无法探测出流的真实转发路径，更无法测量出真实的路径时延。因此，为了准确的测量链路时延，测量工具必须能够提前找出数据包的真实转发路径。基于此发现，我们提出了一套新的软件定义网络中的流追踪机制“FlowTrace”，利用它来追踪任意流量的转发路径以及测量数据流所经历的链路时延。该工具通过收集交换机的流表来计算流的转发路径。然而，如果直接从交换机中查询流表会产生很大的数据平面流量，从而带来巨大的开销。因此，我们提出了一种被动的零开销的流表收集方法来解决该问题。在获得流的真实转发路径后，我们提出了一种新的测量方法来测量不同流的网络时延。最后，实验结果显示我们设计的方法可以准确的找出流的真实转发路径并测量出不同种类流所经历的时延。

We present a high-definition (HD) 3D laparoscopic system including a dual channel optical system, two cameras, a camera control unit (CCU), and an HD 3D monitor. This laparoscopic system is capable of outputting dual high-definition videos and providing vivid 3D images. A modified pinhole camera model is used for camera calibration and a new method of depth measurement to improve precision. The average error of depth measurement measured by experiment (about 1.13 mm) was small in proportion to the large range in distance of the system (10−150 mm). The new method is applicable to any calibrated binocular vision system.

Traditional CMOS technology faces some fundamental physical limitations. Therefore, it has become very important for the integrated circuit industry to continue to develop modern devices and new design methods. The threshold logic gate has attracted much attention because of its powerful logic function. The resonant tunneling diode (RTD) is well suited for implementing the threshold logic gate because of its high-speed switching capability, negative differential resistance (NDR) characteristic, and functional versatility. In this paper, based on the Reed-Muller (RM) algebraic system, a novel method is proposed to convert three-variable non-threshold functions to the XOR of multiple threshold functions, which is simple and has a programmable implementation. With this approach, all three-variable non-threshold functions can be presented by the XOR of two threshold functions, except for two special functions. On this basis, a novel three-variable universal logic gate (ULG3) is proposed, composed of two RTD-based universal threshold logic gates (UTLG) and an RTD-based three-variable XOR gate (XOR3). The ULG3 has a simple structure, and a simple method is presented to implement all three-variable functions using one ULG3. Thus, the proposed ULG3 provides a new efficient universal logic gate to implement RTD-based arbitrary n -variable functions.

Personalized microblog recommendations face challenges of user cold-start problems and the interest evolution of topics. In this paper, we propose a collaborative filtering recommendation algorithm based on a temporal interest evolution model and social tag prediction. Three matrices are first prepared to model the relationship between users, tags, and microblogs. Then the scores of the tags for each microblog are optimized according to the interest evolution model of tags. In addition, to address the user cold-start problem, a social tag prediction algorithm based on community discovery and maximum tag voting is designed to extract candidate tags for users. Finally, the joint probability of a tag for each user is calculated by integrating the Bayes probability on the set of candidate tags, and the top n microblogs with the highest joint probabilities are recommended to the user. Experiments using datasets from the microblog of Sina Weibo showed that our algorithm achieved good recall and precision in terms of both overall and temporal performances. A questionnaire survey proved user satisfaction with recommendation results when the cold-start problem occurred.

We propose a pipelined Reed-Solomon (RS) decoder for an ultra-wideband system using a modified stepby-step algorithm. To reduce the complexity, the modified step-by-step algorithm merges two cases of the original algorithm. The pipelined structure allows the decoder to work at high rates with minimum delay. Consequently, for RS(23,17) codes, the proposed architecture requires 42.5% and 24.4% less area compared with a modified Euclidean architecture and a pipelined degree-computationless modified Euclidean architecture, respectively. The area of the proposed decoder is 11.3% less than that of the previous step-by-step decoder with a lower critical path delay.

Real-time encryption and decryption of digital images stored on end-user devices is a challenging task due to the inherent features of the images. Traditional software encryption applications generally suffered from the expense of user convenience, performance efficiency, and the level of security provided. To overcome these limitations, the concept of transparent encryption has been proposed. This type of encryption mechanism can be implemented most efficiently with kernel file systems. However, this approach has some disadvantages since developing a new file system and attaching it in the kernel level requires a deep understanding of the kernel internal data structure. A filesystem in userspace (FUSE) can be used to bridge the gap. Nevertheless, current implementations of cryptographic FUSE-based file systems suffered from several weaknesses that make them less than ideal for deployment. This paper describes the design and implementation of ImgFS, a fully transparent cryptographic file system that resides on user space. ImgFS can provide a sophisticated way to access, manage, and monitor all encryption and key management operations for image files stored on the local disk without any interaction from the user. The development of ImgFS has managed to solve weaknesses that have been identified on cryptographic FUSE-based implementations. Experiments were carried out to measure the performance of ImgFS over image files’ read and write against the cryptographic service, and the results indicated that while ImgFS has managed to provide higher level of security and transparency, its performance was competitive with other established cryptographic FUSE-based schemes of high performance.

利用海量网络安全威胁情报数据，构建网络安全知识图谱实施深度关联分析和挖掘，可帮助识别安全威胁并提出相应防御措施。这已成为网络安全领域研究热点。本文针对网络安全文本数据，研究实体识别算法，为构建网络安全知识图谱奠定基础。传统方法难以识别网络安全领域的新实体、中英文混合安全实体等，且提取的特征不够充分。本文在神经网络模型基础上，提出基于特征模板的CNN-BiLSTM-CRF网络安全实体识别算法。首先构建人工特征模板，提取局部上下文特征。再利用CNN提取字符特征，与局部上下文特征结合，传入BiLSTM模型提取语义特征。最后利用CRF对安全实体进行标注。结果表明，在大规模网络安全数据集上，该方法优于其它算法，F值达到86%。

隐式反馈是指用户通过行为间接表达用户的意见。由于其在现实世界中的易得性和丰富性，越来越受到推荐系统领域的关注。使用隐式反馈的常见做法是把隐式反馈当做正面或者负面的用户偏好，并附带不同的置信度。这类方法大多认为，与隐式反馈相关的数值（例如观看的时间）反映了置信度的大小，并且数值越大代表置信度越大。这种假设只适合只有一种隐式反馈的场景。然而在现实世界中，经常同时存在多种隐式反馈，称为异构隐式反馈。现有方法无法有效地从异构隐式反馈中推测出置信度。本文提出了一种新颖的利用异构隐式反馈预测置信度的方法，把预测的置信度应用到“基于点的”和“基于对的”矩阵分解模型中，并提出了一种更通用的方法来为“基于对的”方法选择有效的训练样本。在天猫提供的真实电商数据集上的实验表明，本文方法在多个指标上的效果均要优于现有的方法。