最近读了Michael Colins 大神关于感知器收敛的证明,居然能看懂。瞬间有了信心写下自己关于其的理解。原文如下链接:
http://www.cs.columbia.edu/~mcollins/courses/6998-2012/notes/perc.converge.pdf
其 Novikoff 定理如下:
Assume that there exists some parameter vector such that and some, such that for all t = 1…n,
Assume in addition that for all t = 1. . .n,
Then the perceptron algorithm makes at most
errors.
注意,这里的范数是2-范数(欧几里得),有兴趣的同学可以看看别的范数。
http://blog.csdn.net/zouxy09/article/details/24971995
先看看感知机的算法:
Definition: