Lightnews — Scholar-powered news

Ehud Karavani

@ehudk.bsky.social

250 followers 230 following 430 posts

Research Staff Member at IBM Research. Causal Inference 🔴→🟠←🟡. Machine Learning 🤖🎓. Data Communication 📈. Healthcare ⚕️. Creator of 𝙲𝚊𝚞𝚜𝚊𝚕𝚕𝚒𝚋: https://github.com/IBM/causallib Website: https://ehud.co

github.com

Posts Media Videos Starter Packs

Ehud Karavani @ehudk.bsky.social · 4h

not fake news, just alternative facts

Ehud Karavani @ehudk.bsky.social · 1d

🙃

Ehud Karavani @ehudk.bsky.social · 1d

I think it correlates. I'm not a Rothman's Causal Pies person, but the change in designer was probably a necessary but not a sufficient cause, following "iphonism" was already well-established, the chief designer might not have had the mandate for such a drastic change.
But that's just my conspiracy

Ehud Karavani @ehudk.bsky.social · 1d

This has already happened once. the first iphones where very 3D. people were not used to touch screens so buttons aimed to look like the physical ones as much as possible, and UIs looked like their physical counterparts more generally. And then when people got used to it everything became flat.

Ehud Karavani @ehudk.bsky.social · 1d

The best explanation I heard is Apple prepping us for their virtually augmented future, where windows will need to float between our eyes and the natural environment, so they better be transparent/translucent so they could blend smoothly.

Ehud Karavani @ehudk.bsky.social · 2d

I mean the line between community rabbis and grifters can be very blurry, but mostly these people strongly believe that everything is godly planned, which leads to a very passive approach to life more generally

Ehud Karavani @ehudk.bsky.social · 11d

i imagine you before class like a boxer prepping for a match

Ehud Karavani @ehudk.bsky.social · 11d

when scicomm is at the end of its ROPE

Ehud Karavani @ehudk.bsky.social · 14d

oh, and also congrats 😄🐣

Ehud Karavani @ehudk.bsky.social · 14d

on the upside: making this type of intelligence agents is often much more fun than coding

1 2

Ehud Karavani @ehudk.bsky.social · 15d

Nowadays there are excellent git interfaces (JetBrain's/PyCharm's is no less than amazing), and I really hope people catch on to that because, imo, reasonable git workflows do magic in organizing the very process of research

1 1

Ehud Karavani @ehudk.bsky.social · 15d

the recurring theme in the replies here seems to be the UX. Which is unfortunate, because people seem to be deeply aware of the benefits of version control.
people use version control on Google docs all the time, git is just a way to make each version meaningful.

1 1

Ehud Karavani @ehudk.bsky.social · 26d

@alxndrmlk.bsky.social ☝️
(if you won't go for it, I might try)

Ehud Karavani @ehudk.bsky.social · 27d

I don't think it's a replacement either, and imo these formal components can be part of it (estimands in the casual question/contrast, identification in the observational emulation part...). I just find the TT mental framework useful for converting design to analysis code, not just as a comm tool.

Ehud Karavani @ehudk.bsky.social · 28d

I think the advantage of treating TTE as a method is that it makes "time zero" more concrete than other designs, and this in turn reduces bias that arises when organizing data for an analysis (which results in treatment groups bring more comparable).
and it also helps making the estimand clearer,imo

Ehud Karavani @ehudk.bsky.social · Sep 7

fair enough

Ehud Karavani @ehudk.bsky.social · Sep 7

I see it more as a stand up to publishers signing multi-million dollar contacts to give access to papers the authors paid them for being published, and still seeing no attribution or compensation.

Ehud Karavani @ehudk.bsky.social · Sep 1

Be careful Aki "it's just polishing work" are very famous last words. Good luck! 🤞

Ehud Karavani @ehudk.bsky.social · Aug 12

the number of times i was in rooms with a ~100 years of cumulative experience, watching a bunch of color-coded RNAseq UMAPs, examining them like art critics interpreting a Jackson Pollock

Ehud Karavani @ehudk.bsky.social · Aug 12

>10 years on and i still haven't seen a single question answered with clustering

2 2 24

Ehud Karavani @ehudk.bsky.social · Aug 10

IVF is such a brutal process and success is so rare, that I can't not think about the physical and psychological toll women might go through by buying into the hype and optimizing for barely noticeable behavioral traits rather than whichever embryo has the highest chances of successful birth. 2/2

Ehud Karavani @ehudk.bsky.social · Aug 10

when we were working on the polygenic scores-based embryo selection paper, i had a classmate (unrelatedly) working on computer vision methods for morphological-based prediction of embryo implantation chances (in cows). 1/2

Ehud Karavani @ehudk.bsky.social · Aug 10

nice! thank you for explaining and for contributing to better open source statistical vis 💐

Ehud Karavani @ehudk.bsky.social · Aug 9

Norm, Frank, I love your substantiated opinions and stories from your long and fascinating careers, but just note this deborah thingy is a bot ❤️

Ehud Karavani @ehudk.bsky.social · Aug 8

I remember the same poasters taking a dump on sklearn (and python by extension) for defaulting to ridge also fiercely advocating for weakly informative bayesian priors.
the internet can be a funny place i guess.