I've been close reading a paper and was mystified when the author mentioned two kernels: The logistic kernel $\frac{1}{e^x + 2 + e^{-x}}$ and the sigmoid kernel $\frac{2}{\pi}\frac{1}{e^x + e^{-x}}$. I'm mystified because the logistic function, also called the sigmoid function, is $\frac{1}{1+e^{-x}}$, which is a different function that lacks the symmetry of either of those two.

When I asked the author, he pointed to the Wikipedia page called Kernel (statistics). That page does list these two kernels, but I can't find a reference that introduces these two, or why they have these names when they're so different from the logistic/sigmoid function.

Could I have a clearer reference please?


1 Answer 1


Kernel is basically a PDF of a respective distribution without parameters. Logistic distribution uses logistic function as CDF, and its PDF would be exactly as you stated in your question (you can check that out by taking a CDF derivative).

I'm pretty curious about the sigmoid example as well, while its CDF is still a sigmoid in a broad sense (DL frameworks mostly stick to a narrow sense where sigmoid=logistic), it looks like a hyperbolic secant distribution, which does not seem that widely used to me to be generalized as a sigmoid kernel.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.