Talk:Smooth maximum

this page needs good references to generalized mean, theoretical & implementation details, uses and possible links back to the softmax function. So far only John Cook's blog posts seem to be useful, http://www.johndcook.com/blog/2010/01/13/soft-maximum/. He does however use a different function that converges in the limit and call the topic soft maximum.

definition according to who?

The article gives a definition of a smooth maximum in general - a category of functions that must fulfil some criteria - but does not say who gave this definition, or who uses it, or why this definition might exist. Is it real? Is there supporting literature?--mcld (talk) 15:33, 8 March 2017 (UTC)[reply]

LogSumExp

The article claims that LogSumExp is a smooth maximum, but this is a category error, since we have defined a smooth maximum to be a *family* of functions, rather than a single function. There is no discussion on how we can parameterize LogSumExp to get a better approximation to the maximum function. J.Gowers 12:47, 2 January 2020 (UTC)[reply]

Another possible approximation

I think it could be approximated by different functions for $x>0,\ y>0,\ x\neq y$ like:

{\text{max}}\{x,\ y\}\approx \ln(e^{x}+e^{y})\approx {\dfrac {x\,e^{\frac {x}{\ln(2)}}-y\,e^{\frac {y}{\ln(2)}}}{e^{\frac {x}{\ln(2)}}-e^{\frac {y}{\ln(2)}}}}

Notice it has different signs than the Boltzmann approximation.

I believe is interesting since I have never seen before the approximation for splitting the logarith of a sum before finding it accidentally here {https://math.stackexchange.com/q/4838311/909869}. Maybe a superuser could incorporate it after checking it do works as intended. 45.181.122.234 (talk) 12:54, 29 September 2024 (UTC)[reply]