基于动态卷积的体素内不相干运动成像参数估计

体素内不相干运动(IVIM)磁共振成像是一种能够表征生物组织内水分子扩散和灌注的无创技术. 传统IVIM参数估计方法受到图像噪声的影响, 参数估计效果不佳. 为了准确、快速地确定组织区域的扩散和灌注参数信息, 本文充分考虑来自于体素信号之间的上下文信息和b值对于IVIM参数的贡献程度, 提出一种基于动态卷积模块的一维卷积神经网络(dynamic convolutional neural network, DCNN)估计IVIM参数. 在具有不同噪声水平的测试仿真数据和真实采集图像上, 与传统的IVIM参数估计方法进行了比较. 实验结果表明, 本文提出的DCNN方法能够降低IVIM参数的变异系数、偏差和相对均方根误差, 提高了参数一致性和鲁棒性的同时, 仍具有较高的IVIM参数视觉质量.     

基于增强稀疏性特征选择的网络图像标注

面对网络图像的爆炸性增长,网络图像标注成为近年来一个热点研究内容,稀疏特征选择在提升网络图像标注效率和性能方面发挥着重要的作用.提出了一种增强稀疏性特征选择算法,即,基于l_2,1/2矩阵范数和共享子空间的半监督稀疏特征选择算法(semi-supervised sparse feature selection based on l_2,1/2-matix norm with shared subspace learning,简称SFSLS)进行网络图像标注.在SFSLS算法中,应用l_2,1/2矩阵范数来选取最稀疏和最具判别性的特征,通过共享子空间学习,考虑不同特征之间的关联信息.另外,基于图拉普拉斯的半监督学习,使SFSLS算法同时利用了有标签数据和无标签数据.设计了一种有效的迭代算法来最优化目标函数.SFSLS算法与其他稀疏特征选择算法在两个大规模网络图像数据库上进行了比较,结果表明,SFSLS算法更适合于大规模网络图像的标注.     

取定s的严格d-正则随机(3,2s)-SAT问题的可满足临界

3-CNF公式的随机难解实例生成对于揭示3-SAT问题的难解实质和设计满足性测试的有效算法有着重要意义.对于整数k>2和s>0,如果在一个k-CNF公式中每个变量正负出现次数均为s,则称该公式是严格正则（k,2s）-CNF公式.受严格正则（k,2s）-CNF公式的结构特征启发,提出每个变量正负出现次数之差的绝对值均为d的严格d-正则（k,2s）-CNF公式,并使用新提出的SDRRK2S模型生成严格d-正则随机（k,2s）-CNF公式.取定整数5<s<11,模拟实验显示,严格d-正则随机（3,2s）-SAT问题存在SAT-UNSAT相变现象和HARD-EASY相变现象.因此,立足于3-CNF公式的随机难解实例生成,研究了严格d-正则随机（3,2s）-SAT问题在s取定时的可满足临界.通过构造一个特殊随机实验和使用一阶矩方法,得到了严格d-正则随机（3,2s）-SAT问题在s取定时可满足临界值的一个下界.模拟实验结果验证了理论证明所得下界的正确性.     

基于深度稀疏自动编码器的社区发现算法

社区结构是复杂网络的一个重要特征,社区发现对研究网络结构有重要的应用价值.k-均值等经典聚类算法是解决社区发现问题的一类基本方法.然而,在处理网络的高维矩阵时,使用这些经典聚类方法得到的社区往往不够准确.提出一种基于深度稀疏自动编码器的社区发现算法CoDDA,尝试提高使用这些经典方法处理高维邻接矩阵进行社区发现的准确性.首先,提出基于跳数的处理方法,对稀疏的邻接矩阵进行优化处理.得到的相似度矩阵不仅能反映网络拓扑结构中相连节点间的相似关系,同时能反映不相连节点间的相似关系.接着,基于无监督深度学习方法,构建深度稀疏自动编码器,对相似度矩阵进行特征提取,得到低维的特征矩阵.与邻接矩阵相比,特征矩阵对网络拓扑结构有更强的特征表达能力.最后,使用k-均值算法对低维特征矩阵聚类得到社区结构.实验结果显示,与6种典型的社区发现算法相比,CoDDA算法能够发现更准确的社区结构.同时,参数实验结果显示,CoDDA算法发现的社区结构比直接使用高维邻接矩阵的基本k-均值算法发现的社区结构更为准确.     

多任务的高光谱图像卷积稀疏编码去噪网络

目的 高光谱图像由于其成像机理、设备误差和成像环境等因素导致采集到的数据存在噪声。传统稀疏表示方法需要把高光谱图像划分为一系列的重叠局部图像块进行表示,通过对重叠图像块去噪结果进行平均,实现整体图像去噪。这种局部—整体去噪方法不可避免地会破坏高光谱图像空间关系,产生较差的去噪效果和视觉瑕疵。本文利用卷积算子的平移不变性,采用卷积稀疏编码（convolutional sparse coding,CSC）对高光谱图像进行整体表示,保留不同图像块之间的空间关系,提升高光谱图像去噪性能。方法 将每个波段去噪看做单任务,采用卷积稀疏编码描述单波段的局部空间结构关系。通过共享稀疏编码系数,实现不同波段之间的全局光谱关联关系建模,形成多任务卷积稀疏编码模型。多任务卷积稀疏编码模型一方面可以实现高光谱图像的空间—光谱关系联合建模;另一方面,对高光谱图像进行整体处理,有效地利用图像块之间的关系,因此具有很强的去噪能力。借鉴深度学习强大的表征能力,将多任务卷积稀疏编码模型的算法迭代过程通过深度展开（deep unfolding）方式转化为端到端可学习深度神经网络,即多任务卷积稀疏编码网络（multitas...     

K-means聚类方法下的复数图像群组稀疏编码降噪算法

稀疏编码已经广泛应用于复数图像的降噪问题,其中,近些年提出的分组稀疏编码由于能够充分利用同一分组图像块的相似性,在滤除噪声和提高降噪信噪比方面具有更大的优势.研究了一种基于K-means聚类方法的复数图像分组稀疏降噪算法,通过改进聚类算法,验证了K-means算法对分组稀疏编码算法的分组有效性.采用在线复数词典训练算法快速获取编码字典,并运用分组正交匹配追踪算法,实现了分组图像块的稀疏编码.通过限制每一分组图像块中编码的相似性,有效抑制了对图像块中噪声的编码,提高了对复数图像的降噪效果.为验证算法的有效性,对模拟和真实的干涉合成孔径雷达图像的仿真噪声进行了定量分析,证明了所提算法相对于以前的分组稀疏编码算法在峰值信噪比指标上有一定的提升.最后对真实的干涉合成孔径雷达图像进行了降噪,进一步验证了所提降噪算法对于真实噪声的降噪能力.     

深度学习的快速磁共振成像及欠采样轨迹设计

目的 快速成像一直是磁共振成像（MRI）技术中的焦点之一,现有多通道并行成像和部分k空间数据重建都是通过减少梯度编码步数来降低数据的获取时间,两者结合起来更能有效地提高扫描速度。然而,在欠采样倍数加高的情况下,依然有很严重的混叠伪影,因此研究一种在保证成像精度的前提下加快成像速度的方法尤为重要。方法 基于卷积神经网络的磁共振成像（CNN-MRI）方法利用大量现有的全采样多通道数据的先验信息,设计并线下训练一个深度卷积神经网络,学习待重建图像与全采样图像之间的映射关系,从而在线上成像时,欠采样所丢失数据能被训练好的网络进行预测。本文探讨了对于深度学习磁共振成像的可选择性欠采样方式,提出了一种新的欠采样轨迹方案。为了判断本文方法的性能,用峰值信噪比（PSNR）、结构相似度（SSIM）以及均方根差（RMSE）来作为衡量的指标。结果 实验结果表明,所提出欠采样方案的综合性能要优于传统欠采样轨迹,PSNR要高出12 dB,SSIM高出近0.1,RMSE要降低0.020.04左右。此外重建结果还与经典的并行重建方法GRAPPA（geneRalized autocalibrating partially parallel acquisitions）、SPIRiT（iterative self-consistent parallel imaging reconstruction from arbitrary k-space）以及SAKE（simultaneous autocalibrating and k-space estimation）作比较,从视觉效果以及各项量化指标得出本文方法能重建出更准确的结果,并且重建速度要快5倍以上。结论 深度学习方法能很好地在线下训练时从大量数据集中提取并学习到有价值的先验信息,所以在线上测试时能在较短时间内重建出优于经典算法的高质量结果;提出的1维低频汉明滤波欠采样方案则有利于提升该网络的性能。     

面向磁共振图像重建的k空间降采样优化

成像速度是关系磁共振临床应用效能的重要因素,在k空间中降采样,再配合图像重建,可有效加快成像速度.因此,文中考虑降采样方式对磁共振图像重建质量的影响,在训练深度学习网络进行磁共振图像重建的情况下,提出联合优化k空间降采样方式与重建模型的方法.从k空间全采样入手,逐步删除次要的相位编码,直到针对相位编码的采样满足稀疏性要求为止.同时,采样方式的优化是和深度学习图像重建模型参数优化交替进行,即赋予每个相位编码一个权重,通过权重大小确定相位编码的重要性,在优化重建网络参数的同时,完成对k空间降采样方式的优化.实验表明文中方法可提升磁共振图像重建质量.     

光RP(k)网络上Hypercube通信模式的波长指派算法

波长指派是光网络设计的基本问题,设计波长指派算法是洞察光网络通信能力的基本方法.基于光RP(k)网络,讨论了其波长指派问题. 含有N=2ⁿ个节点的Hypercube通信模式,构造了节点间的一种排列次序Xn,并设计了RP(k)网络上的波长指派算法.在构造该算法的过程中,得到了在环网络上实现n维Hypercube通信模式的波长指派算法.这两个算法具有较高的嵌入效率.在RP(k)网络上,实现Hypercube通信模式需要max{2,「5(2^n-5/3」}个波长.而在环网络上,实现该通信模式需要复用(N/3+N/12(个波长,比已有算法需要复用「N/3+N/4」个波长有较大的改进.这两个算法对于光网络的设计具有较大的指导价值.     

面向移动对象连续k近邻查询的双层索引结构

移动对象连续k近邻（CKNN）查询是指给定一个连续移动的对象集合,对于任意一个k近邻查询q,实时计算查询q的k近邻并在查询有效时间内对查询结果进行实时更新.现实生活中,交通出行、社交网络、电子商务等领域许多基于位置的应用服务都涉及移动对象连续k近邻查询这一基础问题.已有研究工作解决连续k近邻查询问题时,大多需要通过多次迭代确定一个包含k近邻的查询范围,而每次迭代需要根据移动对象的位置计算当前查询范围内移动对象的数量,整个迭代过程的计算代价占查询代价的很大部分.为此,提出了一种基于网络索引和混合高斯函数移动对象分布密度的双重索引结构（grid GMM index,GGI）,并设计了移动对象连续k近邻增量查询算法（incremental search for continuous k nearest neighbors,IS-CKNN）.GGI索引结构的底层采用网格索引对海量移动对象进行维护,上层构建混合高斯模型模拟移动对象在二维空间中的分布.对于给定的k近邻查询q,IS-CKNN算法能够基于混合高斯模型直接确定一个包含q的k近邻的查询区域,减少了已有算法求解该区域的多次迭代过程;当移动对象和查询q位置发生变化时,进一步提出一种高效的增量查询策略,能够最大限度地利用已有查询结果减少当前查询的计算量.最后,在滴滴成都网约车数据集以及两个模拟数据集上进行大量实验,充分验证了算法的性能.     

Chemical Genetic Programming – evolutionary optimization of the genotype-to-phenotype translation set

A new method of genetic programming, named chemical genetic programming (CGP), which enables evolutionary optimization of the mapping from genotypic strings to phenotypic trees is proposed. A cell is evolved which includes a DNA string that codes the fundamental mapping from the DNA code to computational functionality. Genetic modification of a cell's DNA allows the DNA code and the genotype-to-phenotype translation to coevolve. Building an optimal translation table enhances evolution within a population while maintaining the necessary diversity to explore the entire search space. This work was presented in part at the 9th International Symposium on Artificial Life and Robotics, Oita, Japan, January 28–30, 2004     

On the status and prospects of systems analysis methods in ukraine

The paper deals with the role and place of systems analysis methods in the solution of complex problems in various spheres of human activity in Ukraine. The social and socio-economic prerequisites for development of theory and applied methods are being considered. The main characteristics of the problems to which these methods are applied are formulated. The hierarchy of problems and methods used for solution of complex problems based on systems analysis methods is presented, and the main spheres of their application in Ukraine are presented. Translated from Kibernetika i Sistemnyi Analiz, No. 1, pp. 101–109, January–February, 2000.     

Multimedia object modeling and storage allocation strategies

The improvements in disk speeds have not kept up with improvements in processor and memory speeds. Many techniques have been proposed and utilized to maximize the bandwidths of storage devices. These techniques have proven useful for conventional data, but when applied to multimedia data, they tend to be insufficient or inefficient due to the diversified data types, bandwidth requirements, file sizes and structures of complex objects of multimedia data. In this paper, we discuss the design of an efficient multimedia object allocation strategy that strives to achieve the expected retrieval rates and I/O computational requirements of objects; and also effectively balances the loads on the storage devices. We define a multimedia object model, describe the multimedia object and storage device characteristics, the classification of the multimedia objects according to their I/O requirements, and the fragmentation strategies. We use a bipartite graph model for mapping of fragments to storage devices. A cost function based on a disk utilization per allocated space, the amount of free space, and the bandwidth of a storage device is used to determine the optimal allocation for an object's data.     

A Generalization of Local Fuzzy Structures

The class of bounded residuated lattice ordered monoids Rl-monoids) contains as proper subclasses the class of pseudo BL-algebras (and consequently those of pseudo MV-algebras, BL-algebras and MV-algebras) and of Heyting algebras. In the paper we introduce and investigate local bounded Rl-monoids which generalize local algebras from the above mentioned classes of fuzzy structures. Moreover, we study and characterize perfect bounded Rl-monoids.     

Models with conjugation conditions and high-accuracy methods of their discretization

New classes of problems with discontinuous solutions are considered. Corresponding generalized problems are obtained. Increased-accuracy computational schemes for their discretization are proposed. Numerical schemes are proposed with asymptotic accuracy not worse than that of similar schemes for problems with smooth solutions. Translated from Kibernetika i Sistemnyi Analiz, No. 1, pp. 110–131, January–February, 2000.     

Optimal (N,T)-policy for M/G/1 system with cost structures

In this paper, the optimal (N,T)-policy for M/G/1 system with cost structure is studied. The system operates only intermittently. It is shut down when no customers are present. A fixed set-up cost of K>0 is incurred each time the system is reopened. Also, a holding cost of h>0 per unit time is incurred for each customer present. The (N,T)-policy studied for this system is as follows: the system reactivates as soon as N customers are present or the waiting time of the leading customer reaches a predefined time T (see A.S. Alfa, I. Frigui, Eur. J. Oper. Res. 88 (1996) 599-613; Y.N. Doganata, in: E. Arikan (Ed.), Communication, Control, and Signal Processing, 1990, pp. 1663–1669). Later on, as a comparison, the start of the timer count is relaxed as follows: the system reactivates as soon as N customers are present or the time units after the end of the last busy period reaches a predefined time T. For both cases, the explicit optimal policy (N^*,T^*) for minimizing the long-run average cost per unit time are obtained. As extreme cases, we include the simple optimal policies for N-and T-polices. Several counter-intuitive results are obtained about the optimal T-policies for both types of models.     

A Fully Automatic hp-Adaptivity

We present an algorithm, and a 2D implementation for a fully automatic hp-adaptive strategy for elliptic problems. Given a mesh, the next, optimally refined mesh, is determined by maximizing the rate of decrease of the hp-interpolation error for a reference solution. Numerical results confirm optimal, exponential convergence rates predicted by the theory of hp methods.     

Bessel function of the first kind with complex argument

A new method of computing integral order Bessel functions of the first kind J_n(z) when either the absolute value of the real part or the imaginary part of the argument z = x + iy is small, is described. This method is based on computing the Bessel functions from asymptotic expressions when x∼ 0 (or y ∼ 0). These expansions are derived from the integral definition of Bessel functions. This method is necessary because some existing algorithms and methods fail to give correct results for small x small y. In addition, our overall method of computing Bessel functions of any order and argument is discussed and the logarithmic derivative is used in computing these functions. The starting point of the backward recurrence relations needed to evaluate the Bessel function and their logarithmic derivatives are investigated in order to obtain accurate numerical results. Our numerical method, together with established techniques of computing the Bessel functions, is easy to implement, efficient, and produces reliable results for all z.