David Widmann

BTW I just checked and already without this PR, as expected, we have ```julia julia> eltype(truncated(Normal(0f0,1f0), 0f0, Inf32)) Float32 ``` so the `eltype` definition above is not needed. In fact,...

It's not fixed by the PR alone, so the test would fail.

FYI I applied similar changes also to the reverse-mode rule, and moved computations of constants outside of the pullback. That leads to a significant performance improvement of the pullback function...

I tried to address your comments but I was not completely sure what you had in mind. Can you check if I changed it correctly?

I'd like to get this in, so I'll merge since I think I addressed the two comments. I'm happy to open a follow-up PR if something could/should be improved.

It seems in Distributions there is no performance improvement if `quantile` is not broadcasted but would support arrays (correct me if I am wrong). Thus it seems the PR would...

> One alternative would be to override the definition of broacast for quantile in StatsBase to call the optimized method under the hood (i.e. to sort data only once). Another...

As discussed on Slack, `+` is not defined for sum of Distributions but one has to use `convolve`. I'll open a PR that would allow to use `\oplus` which was...

This PR is really not the right place to discuss Unicode operators but I disagree. `\otimes` is a standard notation for tensor products (and used in this way e.g. in...

I think it would be great if it would be possible to use different matrices and types as covariance matrices. I'm not happy about the added complexity in Distributions though...