弹性波理论研究──关于球面波首项(Ⅰ)

本文推导了球面波在弹性介质中传播的首项公式，包括直达波，反射波和透射波的首项。通过对球面波分解。用柱面波合成球面波；在界面上由压力和速度连续条件，给出了人射波，反射波和透射波的表达式；对汉格尔函数用大自变量展开，用最陡下降法分别求出这三种波的首项解析表达式。     

Approximate Bayesian methods

Summary This paper develops asymptotic expansions for the ratios of integrals that occur in Bayesian analysis: for example, the posterior mean. The first term omitted isO(n ⁻²) and it is shown how the termO(n ⁻¹) can be of importance.     

Computing depth maps from descent images

In the exploration of the planets of our solar system, images taken during a landers descent to the surface of a planet provide a critical link between orbital images and surface images. The descent images not only allow us to locate the landing site in a global coordinate frame, but they also provide progressively higher-resolution maps for mission planning. This paper addresses the generation of depth maps from the descent images. Our approach has two steps, motion refinement and depth recovery. During motion refinement, we use an initial motion estimate in order to avoid the intrinsic motion ambiguity. The objective of the motion-refinement step is to adjust the motion parameters such that the reprojection error is minimized. The depth-recovery step correlates adjacent frames to match pixels for triangulation. Due to the descending motion, the conventional rectification process is replaced by a set of anti-aliasing image warpings corresponding to a set of virtual parallel planes. We demonstrate experimental results on synthetic and real descent images.Received: 7 November 2002, Accepted: 13 September 2004, Published online: 25 February 2005 Correspondence to: Clark F. Olson     

Parameter determination for a generalized fuzzy model

This paper underlines a way to evolve a generalized fuzzy model (GFM), using the interpolation of CRI and TS models in their consequent parts of fuzzy rules. The GFM possesses the index of fuzziness of CRI model and the local model of the TS model. The parameters of the GFM are estimated by a two-step process. The consequent part of fuzzy rules is reformulated to suit the LSE framework for estimating the associated parameters. By assuming Generalized Gaussian membership function for the premise parts, Gradient descent technique is used to update its parameters. The performance of two classes of GFM has been tested on two systems and it is shown that class II GFM is the best out of all the fuzzy models tested.     

Adaptive Preconditioned Variational Methods For Solving Self-Adjoint Partial Differential Equations

The implementation of the Preconditioned Conjugate Gradient method for the solution of large linear systems arising from the discretization of differential operators, requires the pre-determination of only one iteration parameter. The numerical determination of the optimal value of this constant parameter, involves the spectral bounds of some matrices and can be obtained in O ( N 2 ) sine function evaluations, where 1/ N is the discretization mesh size. It is shown that this parameter can be chosen in a stable manner in O (1) operations per iteration, if it is allowed to vary with the iteration index from information derived from the gradient parameters.     

双模态语音识别中乘积HMM权重系数与瞬时SNR的关系研究

在有噪声污染等复杂情况下,为了能够得到更高的语音识别率,提出了一种新的乘积隐马尔可夫模型(HMM)用于双模态语音识别,研究并确定了模型中权重系数与瞬时信噪比(SNR)之间的关系.该模型在独立训练音频和视频HMM的基础上,建立二雏训练模型,并使用重估策略保证更高的准确性.同时引入广义几率递减(GPD)算法,调整音视频特征的权重系数.实验结果表明,提出的方法在噪声环境下体现出了良好稳定的识别性能.     

基于监督下降方法的左心室超声图像分割

针对基于表观模型的图像分割算法在特征点迭代定位过程中计算量过大、对非线性局部特征的优化较为困难等问题,采用一种基于监督学习的梯度下降算法,建立4层多分辨率金字塔模型,并使用一种基于巴氏系数的特征提取函数（B-SIFT）替代原方法中的尺度不变特征变换（SIFT）特征,对左心室心内膜及心外膜进行特征点定位。首先对训练集进行归一化处理,统一经食道超声心动图像（TEE）的尺度;然后建立基于多分辨率金字塔和B-SIFT特征的监督下降模型,得到特征点趋近于真实值的梯度下降方向序列;最后将得到的方向序列作用于测试集中,得到最终的左心室分割结果。将该方法与传统监督下降方法进行对比,其得到的分割平均误差相比传统监督下降方法降低了47%,迭代得到的最终值相对单一尺度的梯度下降算法更加逼近真实值。     

深度学习应用技术研究

本文针对深度学习应用技术进行了研究性综述。详细阐述了RBM（Restricted Boltzmann Machine）逐层预训练后再用BP（back-propagation）微调的深度学习贪婪层训练方法,对比分析了BP算法中三种梯度下降的方式,建议在线学习系统,采用随机梯度下降,静态离线学习系统采用随机小批量梯度下降;归纳总结了深度学习深层结构特征,并推荐了目前最受欢迎的5层深度网络结构设计方法。分析了前馈神经网络非线性激活函数的必要性及常用的激活函数优点,并推荐ReLU （rectified linear units）激活函数。最后简要概括了深度CNNs(Convolutional Neural Networks), 深度RNNs(recurrent neural networks), LSTM(long short-termmemory networks)等新型深度网络的特点及应用场景,并归纳总结了当前深度学习可能的发展方向。     

快速的演化神经网络模型的研究与应用*

针对演化学习的速度较慢、模型的稳定性较差等,利用梯度下降计算改进演化算法,极大地提高了模型的学习速度,对那些需要及时优化工艺参数的生产过程,该种模型有很好的应用前景。     

改进的最速下降法─最好点最速下降法

提出了用最速下降法解无约束多变量最优问题时，存在一些特殊点，本文称为最好点，这些最好点的负梯度方向总是直指目标函数的中心，即最优点．因此通过最好点一次迭代就能得到最优点．阐述了如何搜索得到最好的计算过程和计算程序．这个改进的最速下降法－最好点最速下降法既保持了最速下降法的优点，又消除了其锯齿现象，提高了计算速度．