>
> Has anyone looked at finding optimal discretizations by having a
> discrete variable as a child of a continuous variable and then
> optimizing over the distribution? [I was teaching my class about mixing
> continuous and discrete variables and sketched out how this could be
> done, but I couldn't find a reference.]
>
Part of my Ph.D. thesis work was exactly about this (also, optimizing
discretization over the distribution given evidence). It is still
available at:
http://robotics.stanford.edu/~alexvk/Public/thesis.ps
In short, I was trying to minimize the KL distance of the answer to a
query given a BN on discrete and continuous variables. It can be done
by propagating information back and forth throughout a network (in a
fashion very similar to LS algorithm).
The initial work was published in UAI-97.
-- Alexander V. Kozlov | alexvk@engr.sgi.com | (650) 933-8493
This archive was generated by hypermail 2b29 : Fri Apr 07 2000 - 12:20:25 PDT