Panelist for NIPS 2016 Workshop

[Dustin Tran]

Hi all,

Dave Blei, Tamara Broderick, a few others, and I are organizing another NIPS workshop on approximate inference. (See last year’s at http://approximateinference.org) We’re having a panel focused on software tools for both the applied and methodological researcher. Is anyone here going to NIPS and want to represent Stan? 

Michael was on our panel last year so he can give his thoughts. Other panelists we’re planning to invite are from, e.g., bnpy, BayesFlow, Edward, maybe PyMC3 or JuliaStats or from probabilistic programming.

Dustin

[Michael Betancourt]

Last year the entire panel largely reduced to everyone arguing that hard problems 

require trading off bias for reduced variance, then my countering that you can’t do

that robustly in practice without being able to quantify the bias and variance.

Some agreed in concept but perhaps not in practice, some argued that no one has 

ever done that for any algorithm so we why start now?  The conversation hasn’t

evolved much since then so I don’t really have anything new to contribute.

But if the panel this year is focused on software then maybe someone can talk about

setting up model ensembles for benchmarking algorithm accuracy and some of

the early results with Stan and ADVI.  Of course if the focus is on tools for the

applied researcher it would be great for the speaker to emphasize that applied

researchers can do only fragile work when they get nominal results without any

sense of error!

--
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.