神经网络中的术语

                                                                                                重要的术语

          adaline       adaptive linear element ，                                                               自适应线性单元

           AIC           Akaike^，s information theoretic criterion ，                                        Akaike的信息论准测

           AND          AND logic function ，                                                                       与逻辑函数

           ANN          artificial neural network，                                                                人工神经网络

           APEX        adaptive pricipal component extraction ，                                          自适应主成分提取

           ARMA       autoregressive moving average，                                                      自动回归平滑平均

           ARMAX     autoregerssive moving average with exogenous inputs，                     外部输入性自回归平滑平均

           BER         bit error rate                                                                                     位误差率

           BFGS                                                                                                              Broyden-FLetcher-Goldfarb—Shanno，修正牛顿法（通过对函数的泰勒展开式进行误差分析,提出了对二次模型进行改进的新模型,在此基础上得到了改进的拟牛顿条件,并得到了与其相应的Broyden-Fletcher-Goldfarb-Shanno(BFGS)算法.证明了在适当条件下该算法全局收敛.从试验函数库中选择标准测试函数,对经典的BFGS算法与改进的BFGS算法进行数值试验,试验结果表明改进的算法优于经典的BFGS算法.）

         

           BIBO          bounded-input bounded-output，                                                         有界输入和有界输出

           BP              backpropagation ，                                                                              反向传播

           CAM           content addressable memory，                                                             按照内容的寻址记忆

           CLS            Classical least-squares，                                                                      经典的最小二乘法

           CPCA          constrained PCA，                                                                               约束PCA（主成分）

           DOA           direction ofarrival，                                                                             到达方向（目标域）

           DPC            discrete Picard condition，                                                                    离散皮卡条件

           DTLFNN      distributed time lagged feedforward neural network，                             分布式的时间滞前网络

           EVD            eigenvalue decomposition，                                                                   特征值分解

           FFPA           fats fixed-point algorithm，                                                                    快速的固定点算法

           FFT             fast Fourier transform，                                                                          快速傅里叶变换

           FIR             finite impulse response ，                                                                        有限冲击反应

           FMMC         fuzzy min-maxclassifier ，                                                                        模糊最小-最大分类器

           GHA           generalized Hebbian algorithm，                                                                广义Hebb算法     

           GSVD         generalized SVD，                                                                                    广义的svd（SVD奇异值分解是LSA的数学基础，本文是我的LSA学习笔记的一部分，之所以单独拿出来，是因为SVD可以说是LSA的基础，要理解LSA必须了解SVD）

           IC              independent component，                                                                        独立成分检验

          ICA             independent component analysis，                                                            独立成分分析

          ILS              inverse least squares，                                                                             逆最小二乘

          ISL              in the sense of Lyapunov，                                                                        李雅普洛夫条件

          KO                                                                                                                     karhunen-Oja(由于PCA实际计算中只涉及到输入数据概率密度分布函数（Pdf）的二阶特性（协方差矩阵），所以解出的各主分量只互相正交（不相关），但并不满足相互独立。而且信号的大部分重要特征往往包含在Pdf的高阶统计特性中，所以只有多变量观测数据是由高斯分布的源信号构成，PCA方法才有效。
非线性PCA（NLPCA）即将高阶累积量引入标准的PCA中，是由芬兰学者Karhunen和Oja首先提出并将其应用于ICA。它的可以完成对输入信号的盲分离。高阶累积量是以隐含的方式引入计算的，采用自适应迭代方法便于工程实现。标准的PCA基于信号的协方差矩阵仅能处理高斯信号，而NLPCA可以处理非高斯信号)

         LAPART           laterally primed adaptive reasonance theory ，                                          偏向初始自适应共振理论

         LDU                lower diagonal upper matrix decomposition ，                                              矩阵三角的LDU分解（注意一般的LDU分解不是对称矩阵分解，网上大多数是对称矩阵分解容易搞混，LU分解完，把U矩阵分解成对角阵和单位上三角矩阵的乘积，实际上就是把U矩阵的对角线元素拿出来就是那个对角阵，单位上三角矩阵是将U矩阵对角线写1，同时在每行除以该行对应的对角线元素（就是被拿出去的那个数）

         LMBP                                                                                                                  levenberg-Maiqardt backpropagtion，（最广泛的非线性最小二乘算法，中文为列文伯格-马夸尔特法。它是利用梯度求最大（小）值的算法，形象的说，属于“爬山”法的一种。它同时具有梯度 法和牛顿法的优点。当λ很小时，步长等于牛顿法步长，当λ很大时，步长约等于梯度下降法的步长。在作者的科研项目中曾经使用过多次。图1显示了算法从起 点，根据函数梯度信息，不断爬升直到最高点（最大值）的迭代过程。共进行了12步）列文博格反馈传播法

        LMS                least mean-square，                                                                                     最小均方差

        LP                   linear programming，                                                                                    线性规划

        LVQ                learning vector quantization，                                                                        线性移位寄存器

       LU                   lower upper matrix decompostion，                                                               矩阵三角的lu分解法

       Madaline          multiple Adaline，                                                                                  多重自适应线性单元

       MAJ                 majority logic function  ，                                                                                逻辑表决函数

       MAW               mountain associatd wave，                                                                              波峰

       MDL                minimum description length，                                                                          最小描述长度

       MIMO              multiple-input multiple-output，                                                                      多元 输入和输出

       MLP                multilayer perceptron  ,                                                                                   多层感知器

       MLPNN            multilayer perceptron neural network，                                                         多层感知器神经网络

       MRAC              model reference adaptive control，                                                                 模型参考自适应控制

       MRI                Madaline rule I,                                                                                               网络规划 I 类

       MSE                mean square error  ，                                                                                      均方误差

       MUSIC            multiple signal classification ，                                                                      多信号分类

       NARMA           nonlinear autoregressive moving average，                                                    非线性自回归滑动平均法

       NARMAX         nonlinear autoregressive moving average with exogenous inputs，                 具有外部输入的 非线性自回归滑动平均法

       NARX             nonlinear autoregressive with exogenous inputs，                                           具有外部输入的的非线性自回归

       NGE               nested generalized exmplar，                                                                         嵌套广义标本

       NIPALS          nonlinear iterative partial least squares ，                                                       非线性迭代的部分最小二乘法

       NN_c               neural network controller，                                                                            神经网络控制器

       NN_I               neural network to perfrom system identification，                                            执行系统辨别的神经网络

       NOR             NOT—OR logic function，                                                                                非或逻辑函数

       NP                nonlinear programming or nondeterministic polynomial（time complete），         非线性规划或非确定多项式（时间完全）

       OLS              orthogonal least-squares,                                                                                  正交最小二乘法

       OR               OR logic function，                                                                                           或逻辑函数

       PCA             principal component analysis，                                                                          主成分分析

       PCR             principal component regression，                                                                       主成分回归

       PLSNET        partial least-squares regression neural network，                                              部分最小二乘法回归神经网络

       PLSNET—C   PLSNET-calibration，                                                              PLSNET-校准

       PLSNET—P   PLSNET-prediction，                                                               PLSNET-预测

       PLSR           partial least-squares regression，                                            部分最小二乘法回归

       PN               pseudo noise，                                                                       伪造音

       PRESS         predicted residual error sum of squares ，                            预测残亮误差平方和

       QP              quadratic programming，                                                        二次规划

       RBF             radial basis function，                                                            径向基函数

       RBF NN        radial basis function neural network，                                  径向基函数神经网络

       RLS             recursive least-squares                                                          递归最小二乘法

       RMS            rool-mean-squares，                                                             均方根

       ROB            robust，                                                                               鲁棒性

       RWLS          recursive weighted least-sqares，                                      递归加权最小二乘法

       SEC            standard error of calibration，                                             校正的标准误差

       SEE            standard error of estimation ，                                             估计的标准误差

       SEP            standard  error of prediction  ，                                            预测的标准误差

       SISO          single-input single-output,                                                     但输入单输出

       SGA           stochastic gradient ascent，                                                随机梯度上升

       SNR           signal-to-noise ratio，                                                         信噪比

       SOM           self-organizing map，                                                        自组织映射

       SOR           successive overrelaxation，                                              逐次超松弛

       SRN           simple recurrent network，                                               简单递归网络

       SSE            sum-squared error ，                                                       误差平方和

       SVD           singular value decomposition，                                       奇异值分解

      TDNN          time delay neural network，                                           时间延迟神经网络

      TLS             total least-squares，                                                        总体最小二乘法

      TLU            threshold logic until ，                                                       阈值逻辑单元

      TSVD          truncated SVD，                                                               截断SVD

       VOL            volcano,                                                                           火山（峰值或极限值）

       WSS            wide-sense staionary，                                                  宽平稳显示

       XNOR          exclusive NOR logic function，                                      异或非逻辑函数

       XOR            exclusive  OR logic function，                                        异或逻辑函数

（说明：上面的这些说明是我参考和抄录神经计算原理的书上的，这是为了便于接下来从理论到代码的解释做的第一步，第一次写博客错误很多，希望大家谅解A）

Modules References.Machine Learning.Initialize Model.Regression