(more on 1. Of course, if one has unary neurons with "log" and "exp" activation functions, then multiplication is doable at a relatively moderate cost. Multiplication only becomes difficult, if one has the usual (rather impoverished) set of activation functions.
But even with "log" and "exp", there is some difficulty, because the most important case of multiplication is multiplication by zero. So, a bit more thought would be needed on handling this near zero. And, of course, sign reversal would be a separate unary operation. Still the idea of presenting multiplication in this fashion is neat, might be useful for some things, so let it be recorded here.)
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
timeasmymeasure
no subject
Date: 2016-12-30 05:57 am (UTC)But even with "log" and "exp", there is some difficulty, because the most important case of multiplication is multiplication by zero. So, a bit more thought would be needed on handling this near zero. And, of course, sign reversal would be a separate unary operation. Still the idea of presenting multiplication in this fashion is neat, might be useful for some things, so let it be recorded here.)