D人工神经网络4A.pptVIP

4.2 自组织特征映射(SOM)神经网络 4.2.3.3 功能分析 　　SOM网的功能特点之三是特征抽取。 例4.3 SOM网用于字符排序。 　　图4.12给出通过自组织学习后的输出结果。 　　SOM网完成训练后，对于每一个输入字符，输出平面中都有一个特定的神经元对其最敏感，这种输入-输出的映射关系在输出特征平面中表现得非常清楚。 　　SOM网经自组织学习后在输出层形成了有规则的拓扑结构，在神经元阵列中各字符之间的相互位置关系与它们在树状结构中的相互位置关系相当类似，两者结构特征上的一致性是非常明显的。输出平面上的●号表示处于自由状态的神经元，它们对任何输入样本都不会发生兴奋。 4.2.3 自组织特征映射网(络)的运行原理与学习算法 　　To demonstrate this algorithm, Kohonen used the set of 32 vectors reproduced in the table below. The vectors have dimension 5, and are labeled A-Z,1-6. A B C D E F G H I J K L M N O P Q R S T U V W X Y Z 1 2 3 4 5 6 1 2 3 4 5 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 0 0 0 0 0 1 2 3 4 5 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 0 0 0 0 0 0 0 0 0 0 1 2 3 4 5 6 7 8 3 3 3 3 6 6 6 6 6 6 6 6 6 6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 3 4 1 2 3 4 2 2 2 2 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 3 4 5 6 　　There is order in this set of vectors that can be displayed in a minimal spanning tree, obtained by linking all the vectors together, using the smallest possible square difference between linked vectors: 　　Using Kohonens algorithm on this set of vectors generates the map below: Extensions 　　A traditional Kohonen map groups similar input vectors together. It does not show any discontinuities. Distance implies difference, but nearness does not imply resemblance. Compare the pair of units marked G and H with the pair marked A and 5, on the map above. However, this information is available. All you have to do is make it visible. This can be done by calculating the square difference between neighboring units on the trained map, and use this value to color the edged separating the units. This is done in the map below, where dark lines indicate strong difference, and light lines indicate strong resemblance. 　　Additional information can be added. You can create a minimal spanning tree, and overlay this on the map, as is done on the map below. A B C D E F G H I J K L M N O P Q R S T U V W X Y Z 1 2 3 4 5 6 1 2 3 4 5 3 3 3