A general principle adopted by most of these techniques is the decomposition of the problem into smaller problems that are easier to handle.
For each smaller problem, a single node is employed. Its parameters are determined either iteratively using appropriate learning algorithms.
From the way these algorithms build the network, they are sometimes referred to as constructive techniques. (查看原文)
The other direction we will follow to design a multilayer perceptron is to fix the architecture and compute its synaptic parameters so as to minimize an appropriate cost function of its output.
This method not only overcomes the drawback of the resulting large networks of the previous section but also makes these networks powerful tools for a number of other applications, beyond pattern recognition.
However, this method is confronted with a serious difficulty. This is discontinuity of the step(activation) function, prohibiting differentiation with respect to the unknown parameters(synaptic weights). (查看原文)
0 有用 流萤天下 2018-11-11 15:49:56
原版内容丰富细致值五星，但中文版翻译糟糕像机翻，而且中文版公式有些错误，建议对照着英文原版看
0 有用 为神马还没暴富 2014-04-10 10:46:46
看了一点。
1 有用 慎思园六栋 2013-10-04 20:43:00
书里还是公式的列举为多，推导相对还是少一些，需要根据书中的reference去查找文献以得到严谨的数学证明。因为做人脸识别，有训练集，所以略去看后面无监督学习（聚类）的章节。
0 有用 murfey 2013-04-05 20:58:22
不错了，里面有好多的方法都是人工智能时代的经典算法
0 有用 daphoon 2014-01-13 21:34:47
太多了。。。
0 有用 第四紀 2023-04-03 14:57:34 福建
一些本科痛苦编程的回忆
0 有用 Salooloo 2020-09-30 17:38:51
可以。
0 有用 why 2020-04-26 13:12:07
两个王一个李，他们简直是翻译界的败类，计算机科学的耻辱，拿着机翻来出书，语言不通逻辑混乱，符号与原文相反或者错误，出这样的译书你们良心不痛吗？ 原版英文是本好书，这些人尽会通过机翻来糟蹋好书。
0 有用 violet 2019-05-22 21:14:04
可怕的理论
0 有用 orange 2018-11-30 09:34:32
书的框架不错，但翻译得太糟糕