ISO25000-Related Metrics for Evaluating the Quality of Complex Information Systems

Evaluating complex information systems necessitates deep contextual knowledge of technology, user needs, and quality. The quality evaluation challenges increase with the system’s complexity, especially when multiple services supported by varied technological modules, are offered. Existing standards for software quality, such as the ISO25000 series, provide a broad framework for evaluation. Broadness offers initial implementation ease albeit, it often lacks specificity to cater to individual system modules. This paper maps 48 data metrics and 175 software metrics on specific system modules while aligning them with ISO standard quality traits. Using the ISO25000 series as a foundation, especially ISO25010 and 25012, this research seeks to augment the applicability of these standards to multi-faceted systems, exemplified by five distinct software modules prevalent in modern information ecosystems.

特征与度量联合优化信息几何检测器

目前,信息几何检测器主要采用协方差矩阵特征模型,在矩阵流形上度量待检测单元数据与杂波数据间的差异,以区分目标与杂波,从而实现雷达目标检测.然而,在复杂杂波背景下,雷达回波信号信杂比低,杂波在其中占据主导地位.因此,含有目标回波的雷达回波信号与纯杂波具有统计相似性,该相似性使得二者在矩阵流形上较难区分,从而限制了信息几何检测器的性能优势.为突破特征表示所造成的性能增益限制,本文提出了基于特征与度量联合优化的信息几何检测器.首先设计了特征与度量可调的信息几何检测器灵活框架,并在此基础上基于纽曼-皮尔逊准则建立了特征与度量的联合优化模型,而后利用局部平坦假设与多层感知器,将联合优化模型中的决策变量参数化,并提出了双阶段优化求解方法.基于仿真数据与实测海杂波数据的实验结果表明,该方法检测性能优于现有信息几何检测器等典型目标检测方法,且在目标多普勒接近杂波谱峰时具有较大优势.

基于深度学习的视频修复方法综述

视频作为常见的媒体信息之一,目前已在各个领域得到广泛应用。尤其是以抖音等为代表的短视频软件的兴起,使得与视频相关的技术不断迭代更新。其中,视频修复技术是视频处理研究中的一个热点。视频修复技术是利用视频帧内的像素信息和帧间的时域参考信息对视频帧受损的区域进行内容推理并修复,在补全缺损视频、物体移除及视频伪造检测等场景中具有广泛应用前景。该技术可追溯到二十世纪末的老电影修复技术,该任务通常由专业的技术团队逐帧修复完成。而随着数字技术的发展,近年来已有一些人工智能技术用于视频修复,让老电影重获新生。目前,视频修复技术可分为传统方法和基于深度学习的两类方法。其中,传统方法由于缺少对高层语义信息的理解,在场景复杂、缺失区域较大的情况下修复效果不佳;而基于深度学习的方法随着算法框架的优化和图形处理器性能的提升展现了出色的效果,对修复结果的语义结构准确性和时间一致性都有明显的提升。本文在简要回顾传统视频修复方法的基础上,重点分析四类基于深度学习视频修复方法的网络结构、参数模型、性能表现与优缺点;介绍视频修复领域中常用的数据集和评价指标;最后,对视频修复领域现存的问题进行总结并展望未来可能的研究方向。

利用类内类间信息的原型补足小样本图像分类

基于度量学习的小样本学习方法中模型没有充分挖掘类内样本与类间样本的联系,将单个样本特征视作独立特征用作训练,导致模型生成的原型不准确且特征表示能力差.提出利用类内类间信息的原型补足小样本图像分类模型.首先,将支持集样本的特征送入类内信息提取分支,构造类内信息特征图,并提取特征获取类别描述信息为原型进行补足;然后,利用类间信息提取分支将不同类的查询样本融合生成新样本,并将该组查询样本的标签作为软标签;最后,使用补足后的原型对查询样本与新样本进行分类,并通过分类损失优化模型.在四个公开数据集上的实验结果表明,在MiniImageNet数据集上,准确率提升2.03%~5.48%;在TieredImageNet数据集上,准确率提升2.25%~8.55%;在CUB数据集上,准确率提升5.10%~8.82%;在CIFAR-FS数据集上,准确率提升2.61%~10.03%.证明与同类小样本图像分类方法相比,提出的模型获得了更优的分类性能与泛化性.

基于数据特征相关性和自适应差分隐私的深度学习方法研究

基于差分隐私的深度学习隐私保护方法中,训练周期的长度以及隐私预算的分配方式直接制约着深度学习模型的效用.针对现有深度学习结合差分隐私的方法中模型训练周期有限、隐私预算分配不合理导致模型安全性与可用性差的问题,提出一种基于数据特征相关性和自适应差分隐私的深度学习方法(deep learning methods based on data feature Relevance and Adaptive Differential Privacy,RADP).首先,该方法利用逐层相关性传播算法在预训练模型上计算出原始数据集上每个特征的平均相关性;然后,使用基于信息熵的方法计算每个特征平均相关性的隐私度量,根据隐私度量对特征平均相关性自适应地添加拉普拉斯噪声;在此基础上,根据加噪保护后的每个特征平均相关性,合理分配隐私预算,自适应地对特征添加拉普拉斯噪声;最后,理论分析该方法(RADP)满足ε-差分隐私,并且兼顾安全性与可用性.同时,在三个真实数据集(MNIST,Fashion-MNIST,CIFAR-10)上的实验结果表明,RADP方法的准确率以及平均损失均优于AdLM(Adaptive Laplace Mechanism)方法、DPSGD(Differential Privacy with Stochastic Gradient Descent)方法和DPDLIGDO(Differentially Private Deep Learning with Iterative Gradient Descent Optimization)方法,并且RADP方法的稳定性仍能保持良好.

一种基于数据增强的科技文献关键词提取模型

[研究目的]科技文献关键词提取研究具有重要价值,目前研究中关键词提取方法存在较大误差且只能提取文本中的关键词,难以根据深层语义信息提炼出更符合文本核心主旨的词语。本研究针对关键词提取对上下文隐含语义挖掘不足导致的局限性和重点信息关注不足问题开展研究。[研究方法]提出一种基于数据增强的关键词提取模型(GPT-2 BiLSTM Mul-Attention,GPBA),通过语言模型进行数据增强,并结合BiLSTM+Mul-Attention提取模型进行多特征语义信息融合理解。[研究结论]实验结果表明,基于数据增强的关键词提取模型GPBA总体表现优于其他基线模型,并且能更精确地凝练和提取文本中的关键词。

安全多方计算应用的隐私度量方法

安全多方计算应用对输入信息的隐私保护能力,一方面依靠底层的安全机制,另一方面依靠具体的目标函数.目前对安全多方计算的研究主要集中于防止计算过程泄露信息的安全机制;而对部署安全多方计算的目标函数对参与者的输入信息的隐私保护能力的度量或评估方法研究较少.目标函数的各参与者通过合法的输入和输出推导其他参与者的输入信息的问题不能由安全多方计算的安全机制阻止,因此对目标函数的隐私保护强度的度量关乎安全多方计算方案的具体实施应用.根据信息熵模型,从攻击者的角度定义平均熵和特定熵的概念,提出计算信息收益的方法.进而,通过计算目标函数的理想隐私损耗和实际安全多方计算应用中的实际隐私损耗,衡量安全多方计算具体应用方案的隐私保护强度.

基于复数向量余弦相似度KNN和深度度量学习的高精度无源室内定位

AI及深度学习的发展为未来6G实现高精度室内定位系统提供了新思路。通过分析CSI数据的物理特性,提出了一种基于复数向量余弦相似度的改进KNN算法,显著提升了无源定位的性能。进一步地,采用度量学习方法,设计了Structured Embedding Loss损失函数,并引入Softmax Structure Loss优化神经网络模型,实现了端到端的高效训练和推理。实验结果表明,这些创新方法显著提高了无源定位的精度和鲁棒性,定位准确率和Macro-F1评分分别达到99.15%和99.1%,为无源定位、无线信号处理等领域提供了新的研究视角和技术路径。

基于节点中心性和热度的数据布局方法

随着5G和大数据的飞速发展,海量数据的快速、持久存储需求对存储设备和网络性能提出了极大挑战。信息中心网络以及边缘存储等技术的出现,使数据能够在网络边缘就近存储,减少了数据传输时延。但数据长期存储在网络边缘势必会引发边缘节点空间不足的问题,迫使数据传向更远方,降低存储效率。针对上述问题,提出一种基于节点中心性和热度的数据布局方法,利用节点空闲时间将高空间负载节点的数据迁移至低空间负载的节点,确保边缘高热度的节点空间富余。实验结果表明,相比较Random和一致性Hash等存储布局方案,在进行长期存储任务后,该方法能在边缘空间相对不足的情况下,减少存储数据时的传输开销,使数据写入时间减少30%以上。

Mechanism of Universal Quantum Computation in the Brain

In this paper, the authors extend [1] and provide more details of how the brain may act like a quantum computer. In particular, positing the difference between voltages on two axons as the environment for ions undergoing spatial superposition, we argue that evolution in the presence of metric perturbations will differ from that in the absence of these waves. This differential state evolution will then encode the information being processed by the tract due to the interaction of the quantum state of the ions at the nodes with the “controlling’ potential. Upon decoherence, which is equal to a measurement, the final spatial state of the ions is decided and it also gets reset by the next impulse initiation time. Under synchronization, several tracts undergo such processes in synchrony and therefore the picture of a quantum computing circuit is complete. Under this model, based on the number of axons in the corpus callosum alone, we estimate that upwards of 50 million quantum states might be prepared and evolved every second in this white matter tract, far greater processing than any present quantum computer can accomplish.

基于人工智能的电力控制系统运行状态自动化识别模型

针对电力控制系统当中逻辑拓扑复杂、数据参数繁多、运行状态无法准确判断的问题,该文提出一种以图模型为依据的电力控制系统运行状态自动化识别模型。该模型以图模型为基础依据,对电力控制系统当中不同数据特征的时间序列进行聚类,离散化之后构建相应的差异矩阵。通过识别模型的建立,对电力控制系统的运行状态、是否故障报警等进行判断识别。经过算例验证,该方法在召回率、nDCG以及误警率等方面相比其他算法更具优势。

Improved Information Set Decoding Algorithms over Galois Ring in the Lee Metric

The security of most code-based cryptosystems relies on the hardness of the syndrome decoding(SD) problem.The best solvers of the SD problem are known as information set,decoding(ISD) algorithms.Recently,Weger,et al.(2020) described Stern’s ISD algorithm,s-blocks algorithm and partial Gaussian elimination algorithms in the Lee metric over an integer residue ring Z_(pm),where p is a prime number and m is a positive integer,and analyzed the time complexity.In this paper,the authors apply a binary ISD algorithm in the Hamming metric proposed by May,et al.(2011)to solve the SD problem over the Galois ring GR(p^(m),k) endowed with the Lee metric and provide a detailed complexity analysis.Compared with Stern’s algorithm over Zpmin the Lee metric,the proposed algorithm has a significant improvement in the time complexity.

Online RGB-D person re-identification based on metric model update

Person re-identification （re-id） on robot platform is an important application for human-robot- interaction （HRI）, which aims at making the robot recognize the around persons in varying scenes. Although many effective methods have been proposed for surveillance re-id in recent years, re-id on robot platform is still a novel unsolved problem. Most existing methods adapt the supervised metric learning offline to improve the accuracy. However, these methods can not adapt to unknown scenes. To solve this problem, an online re-id framework is proposed. Considering that robotics can afford to use high-resolution RGB-D sensors and clear human face may be captured, face information is used to update the metric model. Firstly, the metric model is pre-trained offline using labeled data. Then during the online stage, we use face information to mine incorrect body matching pairs which are collected to update the metric model online. In addition, to make full use of both appearance and skeleton information provided by RGB-D sensors, a novel feature funnel model （FFM） is proposed. Comparison studies show our approach is more effective and adaptable to varying environments.

Metric of States

Metric of quantum states plays an important role in quantum information theory. In this letter, we find the deep connection between quantum logic theory and quantum information theory. Using the method of quantum logic, we can get a famous inequality in quantum information theory, and we answer a question raised by S. Gudder.

Single Nucleotide Polymorphisms: A Window into the Informatics of the Living Genome

Nested in the environment of the nucleus of the cell, the 23 sets of chromosomes that comprise the human genome function as one integrated whole system, orchestrating the expression of thousands of genes underlying the biological characteristics of the cell, individual and the species. The extraction of meaningful information from this complex data set depends crucially upon the lens through which the data are examined. We present a biophysical perspective on genomic information encoded in single nucleotide polymorphisms (SNPs), and introduce metrics for modeling information encoded in the genome. Information, like energy, is considered to be a conserved physical property of the universe. The information structured in SNPs describes the adaptation of a human population to a given environment. The maintained order measured by the information content is associated with entropies, energies, and other state variables for a dynamic system in homeostasis. “Genodynamics” characterizes the state variables for genomic populations that are stable under stochastic environmental stresses. The determination of allelic energies allows the parameterization of specific environmental influences upon individual alleles across populations. The environment drives population-based genome variation. From this vantage point, the genome is modeled as a complex, dynamic information system defined by patterns of SNP alleles and SNP haplotypes.

Remote Sensing and Landscape Metrics-Based Forest Physical Degradation: Two-Decades Assessment in Gishwati-Mukura Biological Corridor in Rwanda, East-Central Africa

The management of forest corridors and related ecology is one of the effective strategies to minimize the adverse effects of forest degradation. It controls the connectivity of inhabitant species and the connection of the isolated patches. This study analyzed spatial and temporal forest physical degradation based on forest cover change and forest fragmentation in the Gishwati-Mukura biological corridor from 1990-2019. Remotely sensed datasets, Geographical Information System (GIS) and FRAGSTATS software were used to analyze the spatial and temporal physical degradation and changes in forest cover. The results indicated that the Gishwati-Mukura corridor experienced massive deforestation where approximately 7617.1 ha (64.22%) of forest cover was completely cleared out, which implies an annual forest loss of 262.6 ha·year-1 (2.21%) during 1990-2019. The forest cover transitions patterns and geostatistical analysis indicated that extensive deforestation was associated with intensive agriculture. The results demonstrated that agriculture has dramatically increased from 29.46% in 1990 to 57.22% in 2019, with an annual increase of 1.97%. Since Gishwati-Mukura has changed to National Park (NP), it lacks diversified scientific studies addressing the analysis of the remote and spatial patterns to investigate its physical degradation and landscape dynamics. This research study will serve as remote forest analysis gap-filling and as the cornerstone of numerous other research that will contribute to the improvement of the connectivity assessments along the Gishwati-Mukura corridor and other related ecosystems.

第二届中文抽象语义表示解析评测

抽象语义表示是近年来国内外句子语义解析领域的研究热点,国际上已举办了CoNLL2019和CoNLL2020两届跨语言的评测。中文抽象语义表示评测是CoNLL2020的五大任务之一,取得了接近英语的解析效果,但是评测数据和评测指标仍有较大改进空间。为了推动中文抽象语义解析研究,该文在第二十一届中国计算语言学大会技术评测任务研讨会上组织了第二届评测,以新设计的Align-smatch指标为排名标准,采用改进的语义标注方案和标注语料库来进行评测。在基础测试集上,封闭模式的最高F1值为80.00%;盲测集上的表现则相比基础测试集下降了7个百分点左右。本次评测的最佳结果在MRP指标下比上届提高了2.66个百分点。统计发现,整体性能提升主要来源于概念之间的语义关系预测准确率的提高,而语义关系的对齐还有待提升。

基于宋词生成的大容量构造式信息隐藏算法

在基于文本生成的信息隐藏算法研究中,如何在保证生成文本质量的同时提高隐藏容量是主要存在的挑战.为此本文提出一种基于宋词生成的构造式信息隐藏算法.首先对宋词文本数据进行预训练,然后基于自回归语言模型搭建宋词生成模型;其次根据宋词词牌固有的格式信息设计格律模块,在宋词生成阶段,需要向生成模型输入该格律模块,并通过符号集设计、编码等综合作用,生成宋词诗句.在利用宋词生成模型进行秘密信息隐藏的过程中,对格律模块进行重构,通过平仄韵词牌、词牌格式模板、关键字、韵律及押韵字符的不同选择,有效实现秘密信息的隐藏.信息提取是隐藏的逆过程,且提取过程不需要利用宋词生成模型,仅需根据模板和词典库来进行索引即可,提高了信息提取的效率.实验结果表明,本文提出的算法能够生成格式严格、韵律清晰、句子完整性高的宋词,且生成的宋词文本的信息隐藏容量均值可达21比特/句、安全性高,整体性能优于已报道的主流算法.

结合时序和质量评估的视频行人重识别

针对行人重识别中出现的行人遮挡、图像质量参差不齐的情况,提出一种结合时序和局域质量评估的网络模型。从时序信息和质量评估两方面出发,利用时序信息和注意力机制弥补空间上出现的噪音或者缺失,进一步提高基础特征表达的判别能力;将行人图像按比例分割为3段,分别进行局域质量评估;将得到的分数作为权重与行人特征加权求和得到最终的特征。将ResNet-50作为主干网络,在MARS、DukeMTMC-VideoReID和PRID-2011数据集上的实验结果表明,该网络模型在行人重识别准确率上有一定提升。

基于博弈映射学习的多传感源信息融合三相电机智能故障诊断方法

利用多传感数据能够全面地诊断三相电机的各类机械或电气故障,然而现有的智能诊断方法缺乏有效的关键故障信息挖掘与跨传感源特征聚合学习机制,导致其诊断效果不佳。针对以上问题,该文提出一种基于博弈映射学习的多传感源信息融合三相电机智能故障诊断方法。首先,使用多个并行的自学习特征映射网络从各传感数据中自动地提取故障特征;然后,构建传感源鉴别器并使其与自学习特征映射网络形成博弈关系,精炼故障特征并引导其跨传感源分类聚合;之后,将样本差异度量损失函数引入到优化目标中,以确保各类故障特征之间的空间可分性;最后,利用故障模式识别器进行多传感故障特征融合诊断。实验结果表明,该方法能够基于振动、电流、声等多传感数据准确地诊断三相电机轴承故障、转子故障及电气故障,且诊断性能优于现有方法。