视觉运动感知的神经网络研究

本文涉及到视觉图象信息加工与感知过程中一个十困难的问题:视觉运动信息的神经计算。文中介绍了蝇视觉系统的图形-背景相对运动分辨网络的基本框架,集中描述了图形-背景分辨系统的空间神经生理整合作用的计算机实验结果,证明在图形-背景分辨系统中的增益控制机制、突触传递的非线性、双眼之间的相互作用,以及池细胞的方向选择性等,在空间生理整合中起着重要作用。 文中以Reichardt的相关型初级运动检测器二维阵列和Kohonen的自组织特征映射网络相结合,构成了一个视觉运动感知的自组织神经网络。计算机仿真表明,通过无监督学习,网络可以学会解决局域运动检测所带来的多义性问题。并在网络的输出层自组织生成了类似灵长类视皮层MT区的运动功能柱结构。结果表明:时空互相关、映射、合作、竞争、自适应等可能是运动感知网络的几个基本的神经原理。     

复杂化学模式群的非线性映射及其应用

提出新的非线性映射算法，并分别采用传统非线性映射算法和新的非线性映射算法，将8维橄榄油样本映射于平面。其中新的非线性映射算法获得更好保留样本模式拓扑结构的映射平面，映射平面清晰地反映模式的类别关系，即同类模式都清晰地聚集在一起，实现聚类。     

神经网络与多元统计在复杂化学信息模式分类中的集成应用

天然有机物料的化学组成十分复杂，其组分通常高达几百种，从而使其组成与性质间自关系，即构效关系，难以分辨、识别和确定．例如白酒、香料的构效关系就难以用化学机理描述．通常采用数据处理方法，将其归结为模式分类问题．复杂化学信息模式分类问题有模三维数高、样本容量小等特点，采用人工神经网络（ANN）具有一定的优势和不足．文献［1，2］讨论了从网络训练算法出发，改善ANN运行性能的技术．本文拟从另一角度探讨该问题．1对输入模式的分析当物料的组分数增多时，构效关系的复杂程度将剧增．为提高神经网络的表达和处理有力，必…     

基于自组织特征映射神经网络的金银花分类研究

自组织特征映射神经网络(SOM)以无监督方式进行网络训练,具有自组织功能。网络通过自身训练,自动对输入模式进行分类。中药药用价值与其所含微量元素有直接的关系,药材分类是中药质量控制的重要方法。将金银花中微量元素含量作为网络输入,利用自组织特征映射神经网络对不同产地金银花进行分类。结果表明分类效果较好,符合生产实际。     

人工神经网络方法预测气相色谱保留指数

用误差反向传播（ＢＰ）的人工神经网络（ＡＮＮ）模型及分子结构描述码作为输入特征参数，预测气相色谱保留指数．研究了链烷烃、环脂烃、烯烃及醇、酯、醚等３００个化合物，预测结果平均相对误差不大于２．８３％．     

人工神经网络用于近红外光谱测定柴油闪点

采用主成分－人工神经网络对不同留程柴油的近红外光谱进行校正,预测其闪点。采用监控集控制网络训练过程,以避免过训练。探讨了人工神经网络（ＡＮＮ）、直接线性连接人工神经网络（ＬＡＮＮ）的校正效果,并与局部权重回归（ＬＷＲ）、主成分回归（ＰＣＲ）及偏最小二乘（ＰＬＳ）等校正方法进行了比较,认为人工神经及直接线性关联的较好手段。     

函数连接型网络应用于维生素B族四组分同时测定

报道了函数连接型网络（ＦＬＮ）用于维生素Ｂ族四组分同时测定。采用相关系数和标准偏差从原始紫外光谱数据中挑选１１个波长点供网络处理。在函数连接型网络中，非线性输入模式得到了增强，并使用了推广的δ学习规则。预测结果极行，其相关系数和标准偏差分别为０．９９９０４和０．２６８８５。     

饱和醇定量结构-保留相关研究中人工神经网络的应用

以拓扑指数（分子连接性指数）为结构描述符,用人工神经网络技术建立了醇类化合物的结构与色谱保留值的相关性模型。研究了网络构造对模型稳定性的影响,考察了模型在单一固定相上及固定相上的适应性。与多元线性回归法相比较,人工神经网络模型具有更好的预测结果,但外推能力较弱。     

基于4-咪唑苯甲酸的具有新颖贯穿结构的锌和镉配合物的合成、晶体结构和性质研究

利用刚性不对称配体4-咪唑苯甲酸（HIBA）与过渡金属Zn（Ⅱ）和Cd（Ⅱ）的硝酸盐在水热条件下进行反应制备了两个新颖的配位聚合物[Zn（IBA）2]n（1）和{[Cd（IBA）2（H2O）]·4H2O}n（2）.化合物1结晶在手性空间群P4（1）22,并且具有二重贯穿的二维（4,4）拓扑网格结构.化合物2则是个三维多孔dia结构,它具有6^6拓扑并且发生了四重贯穿.化合物1的非线性光学性质研究表明它具有SHG活性,其强度约是尿素的0.4倍,同时研究了化合物的室温荧光性质.     

人工神经网络应用于光谱分析同时测定增效联磺片三组分含量

应用人工神经网络原理，采用误差反向传播方法，对紫外吸收光谱重叠的增效联磺片中的三个组分含量不经分离可同时进行测定。磺胺嘧啶（ＳＤ）、磺胺甲恶唑（ＳＭＰ）、甲氧苄胺嘧啶（ＴＭＰ）的平均回收率分别为１０３．１％、９８．２１％、９８．７６％。探讨了人工神经网络中网络参数对分析结果的影响。     

遗传程度设计在氟化物非晶态形成的判别条件中的应用

遗传程度设计在氟化物非晶态形成的判别条件中的应用蔡煜东，许伟杰，陈念贻（中国科学院上海冶金研究所上海２０００５０）关键词氟化物，非晶态形成条件，遗传程序设计近几年来，随着遗传程序设计［１］的理论及应用的不断进步，该方法被逐渐应用于模式识别、优化控制、...     

自组织人工神经网络在氟化物非晶态形成的判别条件中的应用

本文运用Ｔ．Ｋｏｈｏｎｅｎ自组织人工神经网络，利用化学健参数法，研究了三元系氟化物非晶态形成的判别条件，建立了相应的计算法判别智能软件．识别成功率较高．结果表明，该方法性能良好，可望成为研究化合物非晶态形成条件的有效的辅助手段．     

Atom-centered symmetry functions for constructing high-dimensional neural network potentials

Neural networks offer an unbiased and numerically very accurate approach to represent high-dimensional ab initio potential-energy surfaces. Once constructed, neural network potentials can provide the energies and forces many orders of magnitude faster than electronic structure calculations, and thus enable molecular dynamics simulations of large systems. However, Cartesian coordinates are not a good choice to represent the atomic positions, and a transformation to symmetry functions is required. Using simple benchmark systems, the properties of several types of symmetry functions suitable for the construction of high-dimensional neural network potential-energy surfaces are discussed in detail. The symmetry functions are general and can be applied to all types of systems such as molecules, crystalline and amorphous solids, and liquids.     

Spontaneous and Fast Molecular Motion at Room Temperature in the Solid State

The development of molecular machines requires new building blocks which are easy to characterize and visualize to realize a complexity comparable to their natural counterparts such as biological enzymes. Furthermore, with the desire to build functional nanobots capable of navigating living organisms, it is necessary that the building blocks show mobility even in the solid state. Herein we report a system which is emissive in the amorphous state but is non‐fluorescent in the crystalline state due to the formation of extensive π‐π interactions. This dual nature could be exploited for easy visualization of its solid‐state molecular rearrangement. The emission of the amorphous film was quenched as the molecules spontaneously formed π‐π interactions even in the solid state. Scratching the non‐emissive film destroyed the interactions and restored the emission of the film. The emission quickly disappeared with an average lifetime of 20 s as the compound reformed the π‐network even at room temperature.     

Room Temperature Solid State Reaction Involving Structural Transformation of Covalent Oxide Network

An interesting structural transformation from a two dimensional (2d) covalent oxide network with a layered structure to a three-dimensional (3d) network with a tunnel structure was found at room temperature in the mixture of hydrated alkali-metal molybdenum bronze and amorphous alkali-metal molybdate. From various experimental results it was concluded that the transformation was due to a room temperature solid state reaction.     

POLYMER NETWORK-POLY(ETHYLENE GLYCOL)COMPLEXES WITH SHAPE MEMORY EFFECT

The complexes of poly(methacrylic acid-co-methyl methacrylate) network with poly(ethylene glycol) stabilized byhydrogen bonds were prepared. By introducing the poly(ethylene glycol), a large difference in storage modulus below andabove the glass transition temperature occurred and the complexes exhibited shape memory behaviors. The morphology ofcomplexes was studied by using DSC, WAXD, and DMA. The results indicate that the fixed phase of this kind of novelshape memory materials is the network, and the reversible phase is the amorphous state of PEG:PMAA complex phase. Theshape recoverability almost reaches 100%. This type of complexes can be regarded as a novel shape memory network.     

Cleavage and recovery of molecular water in silica

The network response associated with the incorporation and reactivity of water molecules in bulk phases of amorphous and crystalline silica are investigated using density functional theory. The extent of network relaxation is found to change the relative stabilities of the reactant and product states. A highly reactive site, with a low activation barrier, is associated with a highly strained site in which network relaxation significantly stabilizes the silanol state by effectively annealing the local structure. Diffusion and exchange reaction paths are found to likely be associated with minimum energy paths in which the stability of the product and reactant states are equal. These latter paths are associated with minimal network response, although the ability of the silanol groups to take on several conformations has an overall effect of changing the stability along a given reaction path.     

An investigation of the interconnection between topological characteristics of physical and chemical networks of anhydride adamantane-containing epoxy polymers

The interconnection between topological characteristics of chemical and physical networks of adamantane-containing epoxy polymers of anhydride curing is studied. It is shown that the introduction of adamantane fragments into the network of an epoxy polymer by different methods affects the glass-transition point and stress-strain properties. The changes in polymer properties depending on the concentration of modifying agents are analyzed during comparison of the frequency parameters of the networks of chemical bonds and engagements within a cluster model of the structure of the polymer amorphous state.     

Fluorescence studies on phenylene moieties embedded in a framework of periodic mesoporous organosilica

The excited state characteristics of phenylene (Ph)-bridged periodic mesoporous organosilica (PMO) powders with crystal-like and amorphous wall structures are investigated. Crystal-like Ph-PMO has a molecular ordering of the bridging organic moieties with intervals of 0.76 and 0.44 nm parallel and perpendicular to the mesochannel direction, respectively, whereas amorphous Ph-PMO has no molecular-level periodicity in the wall. Fluorescence from the exciton and excimer of the Ph moieties and the defect center in the silicate network were detected at room temperature, but fluorescence from the excimer and the defect center were not detected at 77 K for crystal-like Ph-PMO dispersed in a methanol/ethanol mixed solvent. The decay curve of the exciton fluorescence of crystal-like Ph-PMO at room temperature was analyzed successfully using a one-dimensional diffusion model quenched by the defect center and the excimer site. The results were discussed in comparison with those for the crystal-like biphenylene-bridged PMO reported in the preceding paper (Yamanaka et al., Phys. Chem. Chem. Phys., 2010, 12, 11688-11696). The existence of excited states with various conformations including ground state dimers or aggregates of the Ph moieties was suggested for amorphous Ph-PMO. It was clearly apparent that the differences in the excited state dynamics reflected the differences in the molecular-level structure in the wall.     

Segmental size of poly(ethylene terephthalate) determined by optical birefringence and ESR spin-probe method

The segmental size in amorphous poly(ethylene terephthalate) (PET) was determined by two methods: a) from deformational behavior of amorphous samples in the elastic state by optical birefringence measurement, and b) from dynamic behavior of a small spin probe by ESR measurement of correlation times. The use of the network model for orientation of segments in amorphous polymer and of the generalized free-volume model for re-orientation of the probe in the amorphous phase gave comparable values ofn=2.3 and 2.1 structural units/segment. We conclude that trans gauche conformational transitions of rather small macromolecular segments are the basic motion mechanism in _a-segmental mobility in PET.