Skip to main content
7 events
when toggle format what by license comment
Feb 8, 2022 at 10:19 history edited Sextus Empiricus CC BY-SA 4.0
added 6579 characters in body
Feb 8, 2022 at 10:13 history edited Sextus Empiricus CC BY-SA 4.0
added 6579 characters in body
Feb 8, 2022 at 4:41 comment added Rylan Schaeffer I might also suggest that $\kappa \mu $ has a physically intuitive meaning, perhaps one that's even more clear. Imagine you wrap a sphere in cloth (the density) and then pull the cloth away from the sphere at a particular point. How hard you pull dictates how sharp the peak will be.
Feb 8, 2022 at 1:03 comment added Rylan Schaeffer To explain why one might want to use 1 parameter, I have three answers: (1) notational succinctness, (2) avoiding confusion (e.g. my own) and (3) reducing computation when performing variational inference. (3) was what motivated this question. In my particular problem (mixture of von Mises-Fisher distributions), my update equations relate $\kappa_{nk} \mu_{nk} = \kappa_{n-1,k} \mu_{n-1,k} + \pi_{nk} o_n$, and so thinking in terms of 2 parameters requires at every step I decompose $\kappa_{nk} \mu_{nk}$ into 2 components, only to compose them at the next step.
Feb 7, 2022 at 16:31 history edited Sextus Empiricus CC BY-SA 4.0
added 427 characters in body
Feb 7, 2022 at 16:08 history edited Sextus Empiricus CC BY-SA 4.0
added 5 characters in body
Feb 7, 2022 at 16:02 history answered Sextus Empiricus CC BY-SA 4.0