一种基于区间优化的神经网络学习算法

神经网络的学习算法通常是采用梯度下降法，此方法容易陷入局部极小而得到次最优解。另外，对于有些应用来说，用于训练网络的样本的输入／输出数据无法精确给出，而只能以一定的范围的形式给出，这就给传统的神经网络带来了困难。该文提出了一种基于区间优化的神经网络学习算法，可以很好地解决上面所提到的传统神经网络学习算法的缺点。     

基于区间优化的神经网络全局优化方法

Hopfield神经网络被广泛应用于优化问题的求解中,而传统的Hopfield网络通常基于梯度下降法,此方法容易陷入局部极小而得到次最优解或收敛到问题的不可行解。另外,当用于训练网络样本的输入/输出数据无法精确给出,而只能以一定的范围的形式给出时,传统的神经网络学习方法就无能为力了。论文提出了一种基于区间优化的神经网络学习算法,可以很好地解决上面所提到的传统神经网络学习算法的缺点。     

基于深度强化学习的存内计算部署优化算法

针对存内计算大规模神经网络部署导致的计算延迟、运行功耗较大等问题，提出了基于深度强化学习的神经网络部署优化算法。首先，建立了马尔可夫决策过程的任务模型，优化神经网络的延迟和功耗，完成片上计算核心的部署。其次，针对优化部署过程中，存在求解空间过大、探索能力不足等问题，提出了一种基于深度强化学习的智能部署优化算法，从而得到近似最优的神经网络部署策略。最后，针对强化学习探索能力不足的问题，提出了一种基于内在激励的奖励策略，鼓励探索未知解空间，提高部署质量，解决陷入局部最优等问题。实验结果表明，该算法与目前强化学习算法相比能进一步优化功耗和延迟。     

基于遗传算法和神经网络预测的再励学习

提出一种基于遗传算法和神经网络预测法相结合的再励学习方法,利用遗传算法对全局进行最优解搜索,将进化过程中产生的数据用来训练神经网络预测器,当再励学习逼近最优解时,利用预测网络估计动作网络的参数、结构与系统响应之间的映射关系,用预测网络逼近最优解的能力引导遗传算法在局部向最优解快速逼近,以解决遗传算法局部振荡问题,从而实现快速学习的能力。将其应用于矢量控制交流电机的速度环控制器自学习中,仿真实验验证了该算法的有效性。     

计算机网络连接增强优化中的均场神经网络算法

计算机网络结构在进行拓扑扩展时，经常要解决这样一类优化问题：在给定预算限制下，选择一组连接，使带来的利润最大。该文利用均场退火技术和神经网络算法，提出了一种新的解决方法——均场神经网络算法。用该算法求解连接增强问题能更快、更有效地得到全局最优解。     

一种优化模糊神经网络的多目标微粒群算法

模糊神经网络优化是一个多目标优化问题．通过对模糊神经网络和微粒群算法的深入分析，提出了一种多目标微粒群算法．在算法中将网络的精确性和复杂性分别作为目标进行优化，再用一种启发性分量加权均值法来选取个体极值和全局极值．算法能够引导粒子较快地向非劣最优解区域移动并最终获得多个非劣最优解，为模糊神经网络的精确性和复杂性的折中寻优问题提供了一种解决方法．茶味觉信号识别的仿真实验验证了该算法的有效性．     

神经网络基于粒子群优化的学习算法研究

研究神经网络基于粒子群优化的学习算法，将粒子群优化算法用于神经网络的学习训练，并与遗传算法进行了比较，结果表明，神经网络基于粒子群优化的学习算法简单容易实现，而且能更快地收敛于最优解。     

基于免疫遗传算法的前向神经网络设计

针对传统BP算法训练速度慢、易陷入局部最优等缺点，该文提出了一种采用免疫遗传算法设计前向神经网络的方法。为解决神经网络权值随机初始化带来的问题，介绍了一种基于免疫的多样性模拟退火法(SAND算法)来进行神经网络权值初始化。仿真结果表明，该算法比混合遗传算法有更高的性能。     

基于QPSO算法的RBF神经网络参数优化仿真研究

针对粒子群优化(PSO)算法搜索空间有限，容易陷入局部最优点的缺陷，提出一种以量子粒子群优化(QPSO)算法为基础的RBF神经网络训练算法，将RBF神经网络的参数组成一个多维向量，作为算法中的粒子进行进化，由此在可行解空间范围内搜索最优解。实例仿真表明，该学习算法相比于传统的学习算法计算简单，收敛速度快，并由于其算法模型的自身特性比基于PSO的学习算法具有更好的全局收敛性能。     

非线性动态系统优化与参数估计集成的神经网络算法

基于动态系统优化与参数估计集成的迭代算法，提出一个求解系统优化的神经网络方法，得到一种动态系统优化与参数估计集成的神经网络算法，该算法通过重复求解参数估计问题和悠神经网络，获得原问题的精确最优解，由于系统优化问题用神经网络求解，因此该算法具有求解速度快，易于硬件实现等优点，特别适用于在线优化与控制。     

Detection and reconstruction of freeform sweeps

We study the difficult problem of deciding if parts of a freeform surface can be generated, or approximately generated, by the motion of a planar profile through space. While this task is basic for understanding the geometry of shapes as well as highly relevant for manufacturing and building construction, previous approaches were confined to special cases like kinematic surfaces or “moulding” surfaces. The general case remained unsolved so far. We approach this problem by a combination of local and global methods: curve analysis with regard to “movability”, curve comparison by common substring search in curvature plots, an exhaustive search through all planar cuts enhanced by quick rejection procedures, the ordering of candidate profiles and finally, global optimization. The main applications of our method are digital reconstruction of CAD models exhibiting sweep patches, and aiding in manufacturing freeform surfaces by pointing out those parts which can be approximated by sweeps.     

自由曲面之间最短距离的一种新的改进遗传算法

遗传算法具有独有的特性,它采用选择、交叉和变异等策略,获取的解为全局最优解,而且无需计算函数的导数,是一种只考虑输入与输出关系的黑箱方法,因而适用于处理各种复杂问题。由于自由曲面的不规则性,自由曲面最短距离是CAD/CAM领域一个最重要的研究课题之一,也是一个难题。文章基于自由曲面的特性,在遗传算法中引入新的特殊个体,通过大量的计算与分析,提出了求自由曲面之间最短距离的一种新的改进遗传算法,并给出了计算实例,效果显著。     

面向层析测量反求工程的自由曲面轮廓集分割技术

轮廓集的分割分为粗略分割和精化两个阶段．粗略分割是指从其他层上抽取和种子轮廓最相似的轮廓，提出了基于轮廓串匹配的优化方法；在精化阶段，提出了基于转角映射的特征连接点辨识技术，用来调整粗略分割轮廓段的两端点．实例表明，文中算法解决了自由曲面轮廓集精确分割的问题，为进一步实现三维特征参数识别和CAD模型重构打下了基础．     

Efficient Hausdorff Distance computation for freeform geometric models in close proximity

We present an interactive-speed algorithm for computing the Hausdorff Distance (HD) between two freeform geometric models represented with NURBS surfaces. The algorithm is based on an effective technique for matching a surface patch from one model to the corresponding nearby surface patch on the other model. To facilitate the matching procedure, we employ a bounding volume hierarchy (BVH) for freeform NURBS surfaces, which provides a hierarchy of Coons patches and bilinear surfaces approximating the NURBS surfaces (Kim et al., 2011 [1]). Comparing the local HD upper bound against a global HD lower bound, we can eliminate the majority of redundant surface patches from further consideration. The resulting algorithm and the associated data structures are considerably simpler than the previous BVH-based HD algorithms. As a result, we can compute the HD of two freeform geometric models efficiently and robustly even when the two models are in close proximity. We demonstrate the effectiveness of our approach using several experimental results.     

一种新型神经网络在自由曲面重构中的应用

以层次划分和模块化为思想基础,提出了一种新型神经网络模型对自由曲面进行重构,即基于径向基函数（RBF）神经网络的混合网络模型。先后运用减聚类方法、正交最小二乘法、最大似然法对网络进行有无监督的混合训练,旨在解决大样本集的简化建模和快速训练问题,提高混合网络输出精度。实验结果表明该网络模型使得曲面的拟合精度有了明显提高。     

Constructive shell representations for freeform surfaces and solids

We usually model freeform surfaces (mathematically, 2D r-sets embedded in 3D Euclidean space E³) as a finite union of patches represented in the traditional parametric or the recently developed algebraic forms. The article introduces a new representation scheme for freeform surfaces called constructive shell representation (CSR), that draws on recent research on algebraic patches. CSRs of surfaces that constitute boundaries of solids are very useful for solid modeling. They represent thick shells derived from freeform surfaces and provide a means to compute exact CSG representations of freeform solids     

Sweep曲面自相交的快速检测和消除

借鉴offset曲线去除自相交的方法提出了一种去除sweep曲面自相交的方法.首先根据sweep曲面上的点到脊线的距离来判断是否产生自相交;然后确定自相交候选区间集,并去除自相交点集.该方法可以推广到一般曲面与sweep曲面的求交,也可以推广到sweep曲面的offset操作中.在北京大学自主开发的几何造型系统PUM2.0中,通过大量的实例验证了该方法的稳定性和有效性.     

Surface Fairing towards Regular Principal Curvature Line Networks

Freeform surfaces whose principal curvature line network is regularly distributed, are essential to many real applications like CAD modeling, architecture design, and industrial fabrication. However, most designed surfaces do not hold this nice property because it is hard to enforce such constraints in the design process. In this paper, we present a novel method for surface fairing which takes a regular distribution of the principal curvature line network on a surface as an objective. Our method first removes the high‐frequency signals from the curvature tensor field of an input freeform surface by a novel rolling guidance tensor filter, which results in a more regular and smooth curvature tensor field, then deforms the input surface to match the smoothed field as much as possible. As an application, we solve the problem of approximating freeform surfaces with regular principal curvature line networks, discretized by quadrilateral meshes. By introducing the circular or conical conditions on the quadrilateral mesh to guarantee the existence of discrete principal curvature line networks, and minimizing the approximate error to the original surface and improving the fairness of the quad mesh, we obtain a regular discrete principal curvature line network that approximates the original surface. We evaluate the efficacy of our method on various freeform surfaces and demonstrate the superiority of the rolling guidance tensor filter over other tensor smoothing techniques. We also utilize our method to generate high‐quality circular/conical meshes for architecture design and cyclide spline surfaces for CAD modeling.     

Smooth surfaces from rational bilinear patches

Smooth freeform skins from simple panels constitute a challenging topic arising in contemporary architecture. We contribute to this problem area by showing how to approximate a negatively curved surface by smoothly joined rational bilinear patches. The approximation problem is solved with help of a new computational approach to the hyperbolic nets of Huhnen-Venedey and Rörig and optimization algorithms based on it. We also discuss its limits which lie in the topology of the input surface. Finally, freeform deformations based on Darboux transformations are used to generate smooth surfaces from smoothly joined Darboux cyclide patches; in this way we eliminate the restriction to surfaces with negative Gaussian curvature.     

点云模型法矢调整优化算法

点云中存在奇异情况时,采用最小生成树法进行法矢调整会出现错误,而采用曲面重建方法运算效率又较低,为此提出一种点云模型法矢调整的优化算法.算法分别处理薄壁特征、垂直法向和相邻曲面3种奇异情况.对薄壁特征,算法提取特征点并在该处强制进行法矢取反;对垂直法向,算法通过扩大邻域搜索范围来获得法矢变化趋势;对相邻曲面,算法在K邻域中剔除歧义邻域点,避免在最小生成树中生成错误边.实验结果表明,该算法在点云中存在奇异情况时能够进行正确的法矢调整,并且相较于曲面重建方法具有较高的效率.