作为一名测量人员，xx认真履行岗位职责，紧密配合施工，制定切实可行的的测量放线方案。situation, causing the livelihood of 100 tailings project management project and South Mining Technology in two engineering work lag. Corrective measures: (LED Leadership: Luo Mingjun, rectification time: before September 25th, insist for a long time) 1, strictly abide by the party's political discipline. Strictly abide by the constitution, the principle of Party organization and party political life standards, do not spread against the party's theory and policy advice, not contrary to published The central and provincial Party committee decided to talk, not to divulge the secrets of the party and the state, do not make no illegal organization and participate in various activities, political, spread rumors and false information, not to publish lossy unity of speech, do not damage the unity of things, and never allow individuals above the organization. 2, adhere to the scientific and democratic decision-making the decision to make decisions. Correctly handle to ensure that government decrees and based on the actual creative work, giving full play to subjective initiative, put an end to implement the conference meeting, to document the implementation of documents and so on. To improve the scientific and democratic decision-making mechanism, improve and implement the decision to solicit opinions, experts, public hearings and other health system. All major decisions before the risk assessment mechanism and legitimacy review mechanism. Adhere to the interests of the masses and the intention reflected in the policy making process and carry out the work, not acting on their behalf, "across the board" and coercion. (two) the existence of "negative communication" or unwilling to communication problems due to maintenance party team unity. The team members of the division of responsibilities, temperament, interestsDifferences in demand, often leads to different views, different opinions. Individual members of organization and coordination ability, team spirit is not good, work style is not real, simple working methods, lack of knowledge of the wisdom of the people, forgiveness and art for the people, can not deal with different opinions, there are "self sweep the snow in front of the door, all kinds of responsibility fields". Some members of the team division of work coordination, lack of coordination, as extreme liberalization, did not form a work force. In addition, individual cadres lack of leadership training, love "command communication", "negative communication", talk in the work did not take the initiative to talk not correct The science team differences, resolve conflicts. Corrective measures: (LED Leadership: Luo Mingjun, rectification time: before September 25th, insist for a long time) 1, to further strengthen the education and guidance on the implementation of the central eight provisions, the provincial nine provisions of the county committee of the "ten provisions" extensive publicity, eliminate the ideological
第 5 章 CMAC 小脑神经网络
前面几章介绍的 BP 神经网络、Hopfield 神经网络和 BAM 神经网络分别属于前馈和反馈
神经网络，这主要是从网络的结构来划分的。如果从神经网络的函数逼近功能这个角度来分，
神经网络可以分为全局逼近网络和局部逼近网络。若神经网络的一个或多个可调参数（权值
和阈值）在输入空间的每一点对任何一个输出都有影响，则称该神经网络为全局逼近网络，
前面介绍的多层前馈 BP 网络是全局逼近网络的典型例子。对于每个输入输出数据对，网络
的每一个连接权均需进行调整，从而导致全局逼近网络学习速度变慢，对于有实时性要求的
应用来说常常是不可容忍的。如果对网络输入空间的某个局部区域只有少数几个连接权影响
网络输出，则称网络为局部逼近网络。对于每个输入输出数据对，只有少量的连接权需要进
行调整，从而使局部逼近网络具有学习速度快的优点，这一点对于有实时性要求的应用来说
至关重要。目前常用的局部逼近神经网络有 CMAC 网络、径向基函数 RBF 网络和 B 样条网络
等，其结构原理相似，本书主要介绍 CMAC 神经网络和 RBF 神经网络。
1975 年 J.S.Albus 提出一 种模 拟小 脑功 能的 神经 网络 模型 ，称 为 Cerebellar Model
Articulation Controller
[1-2]
，简称 CMAC。CMAC 网络是仿照小脑控制肢体运动的原理而建立
的神经网络模型。小脑指挥运动时具有不假思索地作出条件反射迅速响应的特点，这种条件
反射式响应是一种迅速联想。CMAC 网络有 3 个特点：
· 作为一种具有联想功能的神经网络，它的联想具有局部推广（或称泛化）能力，因
此相似的输入将产生相似的输出，反之则产生独立的输出。
· 对于网络的每一个输出，只有很少的神经元所对应的权值对其有影响，哪些神经元
对输出有影响则由输入决定。
· CMAC 的每个神经元的输入输出是一种线性关系，但其总体上可看做一种表达非线
性映射的表格系统。由于 CMAC 网络的学习只在线性映射部分，因此可采用简单
的
δ
算法，其收敛速度比 BP 算法快得多，且不存在局部极小问题。CMAC 最初主
要用来求解机械手的关节运动，后来被进一步应用于机械人控制、模式识别、信号
处理以及自适应控制等领域
[3-4]
。
5.1 CMAC 结构及工作原理
[5]
5.1.1 CMAC 结构
简单的 CMAC 结构如图 5-1 所示，图中
X
表示
p
维输入状态空间，
A
为具有
n
个单元的
存储区（又称为相联空间或概念记忆空间）。设 CMAC 网络的输入向量用
p
维输入状态空间
X
中的点
Ti
p
iii
xxxx ),,,(
21
��
表示，对应的输出向量用
),,,(
21
i
p
ii
i
xxxFy ��
表示，图中
3,2,1�i
，
输入空间的一个点
i
X
将同时激活
A
中的
L
N
个元素（图 5-1 中
L
N
＝4），使其同时为 1，而
其他大多数元素为 0，网络的输出
i
y
即为
A
中 4 个被激活单元的对应权值累加和。
L
N
称为
泛化参数，反应网络泛化能力大小，也可将其看做信号检测单元的感受野大小。对 CMAC 来
说，其工作过程一般包括两个方面：① 结果输出计算及误差生成阶段；② 权值调整阶段。
