The only activation function I have seen for neural networks so far is the logistic function. Are there other functions commonly used? If so, how do you choose the activation function?
可以将文章内容翻译成中文,广告屏蔽插件可能会导致该功能失效(如失效,请关闭广告屏蔽插件后再试):
问题:
回答1:
Yeap there are: http://en.wikipedia.org/wiki/Activation_function
Selection depends on the problem.
回答2:
Hyperbolic Tangent function is another widely used activation function for neural network. The hyperbolic tangent function will produce positive numbers between -1 and 1. Because the hyperbolic tangent activation function has a derivative, it can be used with gradient descent based training methods.
You can also use your own custom activation function but you need to design your own cost function. For novice, it is better you follow the literature.
标签:
machine-learning