Andy Miller

Research Scientist @apple. ML/Stats & Health.

[Diana Cai @dianarycai]

1/ Come to the Bayesian nonparametrics workshop at #NeurIPS2018 on *Friday* in Room 517D! We have a great set of speakers, panelists, poster sessions, lunch+TensorFlow Probability tutorial!

Schedule: https://t.co/W4Sr3ubvgv
Accepted papers: https://t.co/x0SHKMbqKI

Highlights:

[Ethan Plaut @ethanplaut]

This is not how AI works, or how writing works, or how education works. https://t.co/KBPnzkRTpU

[François Chollet @fchollet]

Bias laundering: when our biases no longer just live in our minds, but get baked into algorithms (e.g. via historically biased data), they become harder to acknowledge, and harder still to identify and fix.

[Andy Miller @_amiller_]

@alexdamour classic mix-up

[Andy Miller @_amiller_]

pretty cool data viz - detecting fake followers on twitter (unrelated, please don't look at my followers) https://t.co/jsJdiVBCzq

[David Duvenaud @DavidDuvenaud]

@StatMLPapers TL;DR: There are two problems: The best Gaussian might still be a bad approximate posterior, and the recognition net might put the Gaussian in a bad place. It turns out the first problem isn't so bad, because the generator can adapt to it. Kudos to @_chriscremer and @lxuechen!

[Alexander D'Amour (@alexdamour.bsky.social) @alexdamour]

Estimate+predict Bayesianly. Use data efficiently; propagate error coherently; generate precise hypotheses.

Validate+falsify frequentist-ly (and frequently!). Run experiments that reject those hypotheses with high probability if they are wrong.

Learn both paradigms. Alternate.

[Kathryn Morrison @kathryn_tm]

I wrote a gentle INLA tutorial, if you've been wanting to learn the gist / get started with INLA, it might be a good place to begin. Happy holidays! #rstats #inla #bayes
https://t.co/4Gwuj8fG7R

[François Chollet @fchollet]

Three challenges in our field

Hype -- Don't oversell. Talk about what you know.
Ethics -- Be aware of the potential for evil in applications of new tech. Do the right thing.
Science -- Empiricism is a valid way to produce new knowledge. So is theory. Hacks & obfuscation aren't.

[Rembert Browne @rembert]

Here’s the thing,
Sane person from Atlanta.

I know you don’t like Kasim.
It makes sense.
Duh.

But don’t let that make you so basic.
That you vote for @marynorwood.

Tomorrow:
Vote for @KeishaBottoms.

[🅐🅜🅘🅡 🅡🅐🅗🅝🅐🅜🅐 @amirrahnama_]

It’s hard to miss #NIPS2017 just because I am an Iranian. I always thought politics should have a limit, but apparently not this time!

[Andy Miller @_amiller_]

Best-ever algorithm? This must be what everyone is talking about at #NIPS2017 #FreeLunch 💸🌳📈 https://t.co/klUZqosYGG

[Quanta Magazine @QuantaMagazine]

A more effective way of tracking trending topics:
https://t.co/YC9ovZqRxD

[Andy Miller @_amiller_]

@axlekb @PatrickRoath a convenience that ought to be enjoyed by those who don't drive!

[Andy Miller @_amiller_]

Time for Mass. to adopt automatic voter registration, @PatrickRoath https://t.co/tyz4FrGwFV

[Alexander D'Amour (@alexdamour.bsky.social) @alexdamour]

For folks who don't like clicking thru, here's the abstract. Overlap in Observational Studies with High-Dimensional Covariates.

https://t.co/dcA2py6roi.

[Vox @voxdotcom]

The GOP tax bill could be a disaster for PhD students https://t.co/cX6t3t07Ek

[Arvind Narayanan @random_walker]

When I tell my computer science colleagues that there are so many fairness definitions, they are often surprised and/or confused. [Thread] https://t.co/K8yY5c5KZS

[Dustin Tran @dustinvtran]

Midnight tomorrow at the latest time zone. :-) (retweeting as a PSA!) https://t.co/WjWw6y9hKu

[Harvard Griffin GSAS @HarvardGSAS]

Analysts like the Dodgers' @dcervone0, PhD '15 in Statistics, have ushered in a new era of baseball https://t.co/Af4PydfLDv #worldseries

[Nate Silver @NateSilver538]

Figuring out *how* to optimize something is mostly a math problem. But figuring out *what* to optimize is mostly not.