面向MPEG-21 DIA的适配决策模型

通用多媒体内容访问是未来多媒体服务的常用模式。针对这一访问模式,MPEG-21提出了DIA技术框架,以期实现相关技术标准化。但是,DIA现在只对上下文描述等元数据制定了标准,而适配决策引擎仍然是一个开放问题。针对已有适配决策模型存在的问题,提出了一种适配决策树模型,该模型将适配决策视为序列决策,并用决策树对其建模。为满足多媒体服务的个性化需求,利用用户偏好信息建立适配决策树。此外,利用优化模型构建决策空间,使参与决策的对象数缩小为模态数,从而减少了适配决策时间,满足了多媒体服务的实时性。通过仿真实验证明,该适配决策树模型在决策误差和平均迭代次数两个指标上都优于或相近于已有的适配决策模型。     

领域自适应任务中的动态参数调整方法

领域自适应方法在特征变换过程中对多个度量大多采取静态权重设置,导致方法在不同任务上效果差异较大.为此,文中提出领域自适应任务中的动态参数调整方法.基于再生希尔伯特空间模型,最小化域间可区分性联合概率分布差异,求解域间不变特征空间.在此过程中,依据A-距离计算域间差异中同类标签和不同类标签分布差异的占比,并以此动态调整可区分性和可迁移性的权重参数,从而达到最优的自适应效果.在3个图像分类数据集上的实验表明文中方法的有效性.     

移动计算环境中自适应体系结构的研究

移动计算是当前Internet和移动通信两大重要技术发展和融合的结果。介绍了移动计算的发展、概念和应用。在分析了实现移动计算的客户 代理 服务器模型的基础上 ,提出了一种新的促进移动计算发展的自适应体系结构 ,并对其进行了初步分析与设计     

基于无监督域适应的可区分联合匹配算法

当域之间差异较大时,域适应的迁移效果较差.缩小域差可改善迁移效果,但却忽略后期分类时的可区分性.因此,文中提出基于无监督域适应的可区分联合匹配算法,根据域间类别的不同进行差异化处理,并结合特征匹配和实例重加权提高迁移效果.使用联合概率分布作为域之间数据分布差异的度量,缩小相同类域之间的距离,提高迁移性;扩大不同类域之间的距离,提高区分性.在特征降维的过程中联合特征匹配和实例重加权,共同构造特征变换矩阵.实验表明,文中算法在18组任务上的分类效果较优.     

内容适配系统控制模型研究

针对内容适配系统建模问题,采用反馈控制理论设计内容适配系统控制模型,采用人工神经网络（ANN）设计系统控制器,提出基于QBP算法的控制器学习算法,并分析内容适配控制系统的运行机制。     

Emergence and Adaptation

I investigate the relationship between adaptation, as defined in evolutionary theory through natural selection, and the concept of emergence. I argue that there is an essential correlation between the former, and “emergence” defined in the field of algorithmic simulations. I first show that the computational concept of emergence (in terms of incompressible simulation) can be correlated with a causal criterion of emergence (in terms of the specificity of the explanation of global patterns). On this ground, I argue that emergence in general involves some sort of selective processes. Finally, I show that a second criterion, concerning novel explanatory regularities following the emergence of a pattern, captures the robustness of emergence displayed by some cases of emergence (according to the first criterion). Emergent processes fulfilling both criteria are therefore exemplified in evolutionary biology by some so-called “innovations”, and mostly by the new units of fitness or new kinds of adaptations (like sexual reproduction, multicellular organisms, cells, societies) sometimes called “major transitions in evolution”, that recent research programs (Maynard-Smith and Szathmary 1995; Michod 1999) aims at explaining.

Dynamic Contextual Adaptation

When developing systems based on COTS, components need to be adapted in most of the occasions to work under certain conditions which were not initially predicted by their developers. Thus, it is very important to provide systems with the skill to dynamically alter their behaviour while running, depending on the changing conditions of the environment. In this work we describe a context-dependent, dynamic mapping between the interfaces of the components being adapted, overcoming some of the limitations of the static mappings presented in previous works. This is achieved by means of contextual environments, which define flexible adaptation policies. We also present a case study, illustrating the proposal, and discuss the improvements these mappings represent in comparison with previous works, as well as some open issues.     

改进的跨语种语音合成模型自适应方法

统计参数语音合成中的跨语种模型自适应主要应用于目标说话人语种与源模型语种不同时,使用目标发音人少量语音数据快速构建具有其音色特征的源模型语种合成系统。本文对传统的基于音素映射和三音素模型的跨语种自适应方法进行改进,一方面通过结合数据挑选的音素映射方法以提高音素映射的可靠性,另一方面引入跨语种的韵律信息映射以弥补原有方法中三音素模型在韵律表征上的不足。在中英文跨语种模型自适应系统上的实验结果表明,改进后系统合成语音的自然度与相似度相对传统方法都有了明显提升。     

一种面向部件标注的领域自适应算法

自动部件标注是一项复杂的视觉识别任务,但传统训练算法不适用于分布差异下的参数学习。为此,将部件标注描述为基于结构化输出的分类问题,提出一种支持结构化模型的自适应学习算法。通过引入基于相似度的正则算子,重新定义结构化支持向量机的损失函数,使训练损失度和源-目标参数差异度同时最小化。实验结果表明,与传统监督学习算法相比,该算法可使标注准确率提升2%~4%,同时指出部件位置特征的分布差异相比外观特征对自适应学习性能的影响更大。     

Adaptation and regulation with signal detection implies internal model

This note provides a simple result showing, under suitable technical assumptions, that if a system Σ adapts to a class of external signals , in the sense of regulation against disturbances or tracking signals in , then Σ must necessarily contain a subsystem which is capable of generating all the signals in . It is not assumed that regulation is robust, nor is there a prior requirement for the system to be partitioned into separate plant and controller components. Instead, one assumes that a “signal detection” property holds.     

基于EDFE的顽健语音识别方法中测试环境特征的自适应

作者曾提出一种基于环境特征判别学习的顽健语音识别方法EDFE,该方法基于最小分类错误准则迭代地学习环境特征。从训练数据中学得的环境特征不可能完全反映测试环境的特点,因此研究测试环境的自适应方法,有利于提高语音识别系统性能。文章讨论EDFE中一种测试环境的自适应方法。     

结合自集成和对抗学习的域自适应城市场景语义分割

针对城市场景标签获取的高额成本问题,文中提出结合自集成和对抗学习的域自适应城市场景语义分割方法.对于源域和目标域的较大域间差异问题,采用风格转换的方法将源域数据集合成具有目标域风格的新数据集,作为新的源域数据集,从而有效减少源域与目标域的域间差异.对于目标域的域内差异问题,引入自集成方法,构造教师网络,利用教师网络在目...     

Adaptation of performance during a week of simulated night work

This study aimed to provide a comparative index of the performance impairment associated with the fatigue levels frequently experienced in workplaces that require night work. To do this, we equated fatigue-related impairment with the impairment resulting from varying levels of alcohol intoxication. Fifteen young individuals participated in two counterbalanced conditions which required them to (1) ‘work’ seven consecutive 8-h night shifts, and (2) consume an alcoholic beverage at hourly intervals until their blood alcohol concentration (BAC) reached 0.10%. In each condition, performance was measured at hourly intervals using a 10-min psychomotor vigilance task (PVT). Analysis indicated that as BAC increased, performance impairment significantly increased. Similarly, response times significantly increased during the first six simulated night-shifts, and lapse frequency significantly increased during the first two shifts. Equating the two conditions indicated that the first simulated night shift was associated with the greatest degree of performance impairment. In general, the impairment at the end of this shift was greater than that observed at a BAC of 0.10%. During the second and third simulated night shifts, the performance impairment was less than on the first night, but greater than that observed at a BAC of 0.05%. For the final four nights, the performance decrements generally did not exceed those observed at a BAC of 0.05%. This suggests that during a week of consecutive night shifts, adaptation of performance occurs.     

Adaptation methodology of CBR for environmental emergency preparedness system based on an Improved Genetic Algorithm

Emergency preparedness enables us to effectively handle sudden environmental events. In previous research, we have proposed to develop environmental emergency preparedness systems employing Case-Based Reasoning (CBR) technology, though developing such a CBR system has been stifled by a deficiency of cases and difficulties of case adaptation. In this paper, an Improved Genetic Algorithm (IGA) is put forward to resolve the issue of adaptability, and thus simultaneously overcoming the deficiency of cases.First we introduce the Frame method, which creates a case representation in accordance with the characteristics of, for instance, a sudden chemical leakage event and the system’s preparedness for treating this case. Then we present the principle of genetic algorithm (GA) for case adaptation. Next, we introduce an Improved Genetic Algorithm (IGA) that achieves case adaptation in the CBR system. The IGA overcomes simplex GA (SGA)’s defects including premature and slow convergence rate, and also enhances search efficiency for globally optimal solutions. The IGA employs technologies including the Multi-Factor Integrated Fitness Function, the Multi-Parameter Cascade Code method, the Small Section method for generation of an initial population, and Niche technology for genetic operations including selection, crossover, and mutation. The results of SGA and IGA performance testing are also presented. A prototype CBR-IGA environmental emergency preparedness system is developed and introduced, and its applicability is tested using a hypothetical ammonia leakage emergency in one district of Shanghai. The results indicate that the proposed IGA methodology can resolve the adaptation issue and expand the case base effectively in CBR systems for environmental emergency preparedness. Future research opportunities are discussed, including the potential to combine CBR, GA, and Back Propagation-Artificial Neural Network (BP-ANN) to alleviate additional challenges, such as the “speed and accuracy” of environmental emergency response.     

伪标签不确定性估计的源域无关鲁棒域自适应

无监督域自适应是解决训练集(源域)和测试集(目标域)分布不一致的有效途径之一.现有的无监督域自适应的理论和方法在相对封闭、静态的环境下取得了一定成功,但面向开放动态任务环境时,在隐私保护、数据孤岛等限制条件下,源域数据往往不可直接获取,现有无监督域自适应方法的鲁棒性将面临严峻的挑战.鉴于此,研究了一个更具挑战性却又未被...     

基于Web注释的移动客户端自适应机制的研究

由于各种移动设备在计算能力上有所区别,目前没有一种较好的机制能够使Web内容根据移动设备的性能转换成不同页面大小。该文给出一种通过注释Web页面的方法来实现自适应移动设备,在不改动原有网站的基础上根据移动设备的计算能力生成合适的Web内容。与其它方法相比,该方法在设备自适应性、用户订制和转换速度等方面有较大的提高。     

基于自监督知识的无监督新集域适应学习

无监督域适应(unsupervised domain adaptation,UDA)旨在利用带大量标注数据的源域帮助无任何标注信息的目标域学习.在UDA中,通常假设源域和目标域间的数据分布不同,但共享相同的类标签空间.但在真实开放学习场景中,域间的标签空间很可能存在差异.在极端情形下,域间的类别不存在交集,即目标域中类...     

基于特征点的人脸模型调整方法探讨

基于特征点的三维人脸模型调整．是利用标准人脸的网格模型生成特定人脸模型的重要步骤，该方法已在人脸识别、虚拟现实、表情动画等多方面的应用中取得了良好效果。本文介绍了网格模型及其调整，对利用模型特征点进行非特征点自动调整的方法进行了具体的分类和分析，并对该方法的进一步发展和应用进行了展望。     

两阶段领域自适应学习

针对领域自适应问题中源域和目标域的联合分布差异最小化问题,提出两阶段领域自适应学习方法.在第一阶段考虑样本标签和数据结构的判别信息,通过学习一个共享投影变换,使投影后的共享空间中边缘分布的差异最小.第二阶段利用源域标记数据和目标域非标记数据学习一个带结构风险的自适应分类器,不仅能最小化源域和目标域条件分布差异,还能进一步保持源域和目标域边缘分布的流形一致性.在3个基准数据集上的实验表明,文中方法在平均分类准确率和Kappa系数两项评价指标上均表现较优.     

A modified Covariance Matrix Adaptation Evolution Strategy with adaptive penalty function and restart for constrained optimization

In the last decades, a number of novel meta-heuristics and hybrid algorithms have been proposed to solve a great variety of optimization problems. Among these, constrained optimization problems are considered of particular interest in applications from many different domains. The presence of multiple constraints can make optimization problems particularly hard to solve, thus imposing the use of specific techniques to handle fitness landscapes which generally show complex properties. In this paper, we introduce a modified Covariance Matrix Adaptation Evolution Strategy (CMA-ES) specifically designed for solving constrained optimization problems. The proposed method makes use of the restart mechanism typical of most modern variants of CMA-ES, and handles constraints by means of an adaptive penalty function. This novel CMA-ES scheme presents competitive results on a broad set of benchmark functions and engineering problems, outperforming most state-of-the-art algorithms as for both efficiency and constraint handling.