Ecological Neural Networks for Object Recognition and Generalization

Generalization is a critical capacity for organisms. Modeling the behavior of organisms with neural networks, some type of generalizations appear to be accessible to neural networks but other types do not. In this paper we present two simulations. In the first simulation we show that while neural networks can recognize where an object is located in the retina even if they have never experienced that object in that position ('where' generalization subtask), they have difficulty in recognizing the identity of a familiar object in a new position ('what' generalization subtask). In the second simulation we explore the hypothesis that organisms find another solution to the problem of recognizing objects in different positions on their retina: they move their eyes so that objects are always seen in the same position in the retina. This strategy emerges spontaneously in ecological neural networks that are allowed to move their 'eye' in order to bring different portions of the visible world in the central portion of their retina.     

前馈神经网络模式分类器的推广能力

研究了前馈神经网络模式分类器的推广能力．从几何和概率两个方面分析了前馈网络用于模式识别的分类机理，在前人证明的基于误差最小的有导师（１和０）前馈网络输出端输出为模式样本后验概率估计结论的基础上，给出了径向基函数网络推广的核函数定理；对于两层感知器网络，提出使用加性噪声的样本来增加网络的推广能力．最后使用实测的五种飞机目标一维距离（纵向）象实验数据进行了模拟实验．     

基于GPCA的高技术产业技术创新能力演化分析

将具有时序性特征的全局主成分分析(GPCA)方法引入到我国高技术产业技术创新能力的分析和评价中．根据评价方法及产业的特点，确定了评价指标体系；对由国有企业构成的5个具体产业的技术创新综合能力、技术创新基础能力、技术创新转换能力和技术创新盈利能力的动态发展状态和轨迹进行了综合分析．结果表明，各产业的技术创新能力在4个方面的发展状态和；演化对程存在较大的差异，创新能力评价指标的逐年变化程度不存在规律性。     

How do tufted capuchin monkeys (Cebus apella) understand causality involved in tool use?

Four tufted capuchin monkeys (Cebus apella) were trained to choose from 2 hook-like tools, 1 of which successfully led to collecting food, whereas the other did not because of inappropriate spatial arrangement of the tool and the food. In Experiment 1, all of the monkeys successfully learned the basic task. The monkeys performed successfully with tools of novel colors and shapes in Experiments 2-5. These results demonstrate that the monkeys used the spatial arrangement of the tool and the food as a cue. However, they failed when there were obstacles (Experiment 6) or traps (Experiment 7) on the path along which she monkeys dragged tools. These results may suggest that capuchin monkeys understand the spatial relationship between 2 items, namely, food and. the tool, but do not understand the spatial relationship among 3 items, namely, food, tool, and the environmental condition. The possible role of stimulus generalization is also considered. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Shape deformation in continuous map generalization

Given a collection of regions on a map, we seek a method of continuously altering the regions as the scale is varied. This is formalized and brought to rigor as well-defined problems in homotopic deformation. We ask the regions to preserve topology, area-ratios, and relative position as they change over time. A solution is presented using differential methods and computational geometric techniques. Most notably, an application of this method is used to provide an algorithm to obtain cartograms.

Nanobody Paratope Ensembles in Solution Characterized by MD Simulations and NMR

Variable domains of camelid antibodies (so-called nanobodies or V_HH) are the smallest antibody fragments that retain complete functionality and therapeutic potential. Understanding of the nanobody-binding interface has become a pre-requisite for rational antibody design and engineering. The nanobody-binding interface consists of up to three hypervariable loops, known as the CDR loops. Here, we structurally and dynamically characterize the conformational diversity of an anti-GFP-binding nanobody by using molecular dynamics simulations in combination with experimentally derived data from nuclear magnetic resonance (NMR) spectroscopy. The NMR data contain both structural and dynamic information resolved at various timescales, which allows an assessment of the quality of protein MD simulations. Thus, in this study, we compared the ensembles for the anti-GFP-binding nanobody obtained from MD simulations with results from NMR. We find excellent agreement of the NOE-derived distance maps obtained from NMR and MD simulations and observe similar conformational spaces for the simulations with and without NOE time-averaged restraints. We also compare the measured and calculated order parameters and find generally good agreement for the motions observed in the ps–ns timescale, in particular for the CDR3 loop. Understanding of the CDR3 loop dynamics is especially critical for nanobodies, as this loop is typically critical for antigen recognition.     

数控铣削加工程序通用模块化的实现

在合理安排工艺过程的基础上,如何编制结构简洁、刀路科学的数控程序很大程度上影响着工件加工效率和质量,探讨如何合理运用子程序和刀具半径补偿功能,实现数控铣削加工程序的模块化、通用化。     

基于PAC-Bayes边界理论的SVM模型选择方法

PAC-Bayes边界理论融合了贝叶斯定理和随机分类器的结构风险最小化原理,它作为一个理论框架,能有效评价机器学习算法的泛化性能。针对支持向量机（SVM）模型选择问题,通过分析PAC-Bayes边界理论框架及其在SVM上的应用,将PAC-Bayes边界理论与基于交叉验证的网格搜索法相结合,提出一种基于PAC-Bayes边界的SVM模型选择方法（PBB-GS）,实现快速优选SVM的惩罚系数和核函数参数。UCI数据集的实验结果表明该方法优选出的参数能使SVM具有较高的泛化性能,并具有简便快速、参数选择准确的优点,能有效改善SVM模型选择问题。     

选择策略的集成学习方法研究

差异性是提高分类器集成泛化性能的重要因素。采用熵差异性度量及数据子集法训练基分类器,研究了爬山选择、集成前序选择、集成后序选择以及聚类选择策略选取个体模型的集成学习。实验结果表明,由选择策略选取差异性较大的个体模型,其集成性能表现出较好的优势;从总体角度考虑,爬山选择策略的集成性能优于集成前序选择和集成后序选择的集成性能;另外,由聚类技术选取的集成模型,当集成正确率较稳定时,则模型间的差异性变化较小;簇数也对集成性能与集成模型间的差异性产生一定的影响。     

广义超球面SVM研究

超球面支撑向量机是不均衡样本分类的一种重要方法.然而,目前引入间隔的超球面支撑向量机中,当一类样本集中不存在支撑向量时,两类样本之间的间隔解是不确定的;在两类样本均存在正常支撑向量的情况下,两类样本之间的间隔为零.间隔不确定或为零在很大程度上影响分类器的推广性能.为此提出了一种广义的超球面支撑向量机算法,通过引入参数n和b,理论推导得出n>b,这样可以保证获得不为零的间隔解.理论分析和实验结果表明,所提供算法在具有较小经验风险的同时,可获得较好的推广性能.