There is a KL divergence function registered for KL(Uniform || Normal)
, but not for KL(Uniform || MultivariateNormal)
. Is there any fundamental reason for this, or it just hasn’t been implemented yet?
Both KL(Normal || Uniform)
and KL(MultivariateNormal || Uniform)
are infinite since Uniform will have bounded support and you’ll divide by zero anywhere outside that region. I’m just trying to understand if there’s anything that should prevent the KL(Uniform || MultivariateNormal)
case. I haven’t tried working through the math yet, I also haven’t found any good references on these derivations. Any pointers are appreciated.