Karl Krauth

Postdoc @Stanford genetics. Previously PhD student @berkeley_ai. Trying to understand proteins with microfluidics and machine learning.

[Polly Fordyce @fordycelab]

Congratulations to postdoc @Krauth for being awarded a Branco Weiss fellowship! Thrilled that this computer scientist dared to learn to do experiments and excited to see what we can learn about protein sequence/function relationships! https://t.co/ObfZY7aCAk

[Karl Krauth @Krauth]

@fordycelab Thanks Polly! 😁

[Karl Krauth @Krauth]

@ml_angelopoulos A good first bet would be to contact an equipment center with the right tools e.g. https://t.co/vguJ62HYDq

They will usually have full-time staff there who can train you on how to use the equipment or point you to potential collaborators.

[Valeriy M., PhD, MBA, CQF @predict_addict]

Great to see @ml_angelopoulos, @stats_stephen, Michael I Jordan, @Krauth and @yixinwang_ paper on applying conformal prediction to recommendation systems receiving best paper award 🥇 at conformal prediction conference COPA this year.

Using conformal prediction for improving recommendation systems has large potential and companies like Amazon have started to publish their own research about it with a paper at NeurIPS2023.

https://t.co/nf5fe6crIo

https://t.co/MBLvA3CBI1

#conformalprediction

[Karl Krauth @Krauth]

Link to the paper:
https://t.co/3Jlw6AuiCc
and ICLR talk:
https://t.co/7WOFIqYmuz

[Karl Krauth @Krauth]

I'll be presenting our paper: Modelling Content Creator Incentives on Algorithm-Curated Platforms at #ICLR2023 in-person today. Come chat with me!

Talk: 15:40 CAT (Oral 2 Track 5)
Poster session: 16:30-18:30 CAT.
Co-authors: @jirimhron Michael Jordan @k__niki and Sarah Dean
🧵

[Karl Krauth @Krauth]

Finally, we also show how to use exposure games to perform pre-deployment audits of recommendation algorithms. For example, we show how to check if an algorithm will incentivize the creation of content that a given subgroup of users might like more.

[Karl Krauth @Krauth]

Just found out about this new software package to compute numerical integrals called Photoshop.

[Karl Krauth @Krauth]

GPT-2 was much better at answering questions from yahoo answers than ChatGPT :(

[Karl Krauth @Krauth]

Say what you will about GPT-2 but it's really good at generating nonsensical Yahoo! answers text when given nonsensical questions from the site.

[Clara Fannjiang @clara_fannjiang]

pumped to share our latest work! if *any* trained model decides what data to test—for example, NNs designing novel proteins—we can quantify its uncertainty w/ finite-sample guarantees.
https://t.co/q6eaEfhj7x
w/ ​​@stats_stephen, @ml_angelopoulos, @jlistgarten, M.I. Jordan 1/5

[BART @SFBART]

If you haven't mixed us, a public transit system, up with a nonparametric Bayesian regression approach which uses dimensionally adaptive random basis elements, are you even from the Bay?

[Becca Roelofs @BeccaRoelofs]

Ever hear that “networks learn low frequency functions first”?
Turns out that in realistic settings the story is more nuanced!

Read on to hear about our new work “Spectral Bias in Practice”
https://t.co/pxMaoVreMe

Work led by Sara Fridovich-Keil, joint with @iraphas13.
🧵1/7

[Karl Krauth @Krauth]

@roydanroy @stephenltu at Google works in this area.

[Karl Krauth @Krauth]

@ccanonne_ I think the shark lassoing incident might have desensitized me to these headlines now.
https://t.co/kxPRCjcQ8G

[Karl Krauth @Krauth]

@ahsin_s @UCBerkeley @lschmidt3 @berkeley_ai @beenwrekt Looking forward to seeing you there!!!

[Karl Krauth @Krauth]

@PavloManovi 🤔

[Alex Klotz @AlexanderRKlotz]

Google Scholar has parsed this cafeteria lunch menu as an author list, and it's delightful