I am confused in your tutorial, do you consider the activation function?

Question

I am confused in your tutorial, do you consider the activation function?

Closed this issue 8 years ago · 4 comments

xiaohu2015 commented 9 years ago

Answer 1 · 2016-07-02T10:54:28.000Z

The tutorial is from other's homepage, not mine. And would you like to describe your question in detail?

Answer 2 · 2016-07-03T03:21:29.000Z

I mean that a activation function such as sigmoid function is adopted in the the weighted sum $W*X$. However, in the PDF tutorial, I find there is no activation function. Hence, it is reasonable if we don't use a activation function? Thanks, now I am a beginner in CNN.

Answer 3 · 2016-07-03T09:07:28.000Z

In CNN model we usually use ReLU (part 5 in the PDF tutorial) instead of sigmoid as activation function. The sigmoid function has fallen out of favor in CNN, please refer to http://cs231n.github.io/neural-networks-1/#actfun for more detial about "sigmoid" and "ReLU"

Answer 4 · 2016-07-03T12:27:50.000Z

Thanks, I got it. In the tutorial, a Conv layer is followded by a ReLu layer. Hence, the ReLu layer is served as a activation function. I am sorry that I have not read the turorial in a comprehensive way.