New paper on "Accelerating Asymptotically Exact MCMC"

[Avraham Adler]

While the referenced is beyond my skills, I was recently informed of a paper accepted to JASA title "Accelerating Asymptotically Exact MCMC for Computationally Intensive Models via Local Approximations" which describes a way to greatly accelerate the performance of MCMC, and so my first thought was of Stan. The Science Daily article is here: <http://www.sciencedaily.com/releases/2015/11/151117112646.htm> and the paper itself can be found at: <http://www.tandfonline.com/doi/abs/10.1080/01621459.2015.1096787>. I'm sure you're all ASA members and can get the article directly; if not, I'd be happy to e-mail it too you.

Is this something that would prove of use for Stan?

Thanks,

Avi

[Michael Betancourt]

No. For all that is good in the world, no.
If there’s something generically useful enough to go into Stan then we’re very likely already working on it.

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Bob Carpenter]

I'm allergic to paywalled journals, but there's also an
old-fashioned double-spaced arXiv paper (probably required
for the submission to the old-fashioned paywalled journal):

http://arxiv.org/abs/1402.1694

But listen to Michael. Approximating a function and its gradients in
high dimensions is hard! Note the qualification on p. 28 of this paper,
which uses GPs to approximate densities:

Finding nearest neighbors is a hard problem asymptotically
with respect to the parameter dimension and size of the sample set,
but our sample sets are neither high dimensional nor large.

The nice thing about HMC is that it scales well with dimensionality.

- Bob

[Avraham Adler]

On Thursday, November 19, 2015 at 11:20:22 AM UTC-5, Michael Betancourt wrote:
> No. For all that is good in the world, no.
> If there’s something generically useful enough to go into Stan then we’re very likely already working on it.
>

Understood; just wanted to drop y'all a note just in case.

Avi

[Bob Carpenter]

> On Nov 19, 2015, at 12:24 PM, Avraham Adler <avraha...@gmail.com> wrote:

>
> On Thursday, November 19, 2015 at 11:44:30 AM UTC-5, Bob Carpenter wrote:
>> I'm allergic to paywalled journals
>

> See attached

Please don't post non open-access material on our mailing lists.
I may hate the paywalled journals, but that doesn't mean I think
we should be pirating their content on our mailing list.

I know you were just trying to help, but I'm going to go and delete
the post from our list.

- Bob

[Daniel Lee]

Thanks, Bob.

- Bob

[Daniel Lee]

(I was going to suggest the same. Thanks for taking care of it.)