degregat

Research Engineer: Mechanism Design, Privacy Preserving Machine Learning, Distributed Systems. "The purpose of a system is what it does." pronouns: they/any

[degregat @degregat]

We wrote up a draft on our work at @anoma on how to formulate intents as constraint satisfaction problems: https://t.co/3dvVTU8Bp4

[degregat @degregat]

@alicemazzy If you find an open minded project to contribute to as a SWE and then apprentice your way into an adjacent area (e.g. by simulating whatever systems are built) it will be a much better use of everyone’s time. If that fails, you can still study.

[degregat @degregat]

@badcryptobitch @Tim_Roughgarden Awesome!
I started one on mechanism design a while ago, should we add references, or even merge them?

https://t.co/RysQ7Jkb5v

[degregat @degregat]

Does anyone know who coined the phrase "there are DAGs and then there are bad data structures"? Or is it a folk aphorism with unknown origin?

[degregat @degregat]

@bobpoekert Is this feature always „capturing information + context in a universal format“, or do other examples come to mind as well?

[degregat @degregat]

@zornsllama @mrclbschff Then I’m 𝕂onfused as well

[degregat @degregat]

@mrclbschff @zornsllama Yeah, I think I’ve seen that once too in a German math course

[degregat @degregat]

@moyix The roundest trip.

[degregat @degregat]

@dystopiabreaker Feels way more prime than 59 too.

[degregat @degregat]

Got myself a mastodon account @[email protected]

[degregat @degregat]

@lambdatotoro I sadly don’t see how this will be a norm quickly and robustly enough without a more explicit incentive structure.

[degregat @degregat]

So what’s the heuristic to pick a mastodon instance? Mutuals: Which ones are you on?

[degregat @degregat]

@lambdatotoro I see it now XD

[degregat @degregat]

@lambdatotoro Why the smallest?

[degregat @degregat]

Evolutionarily stable strategies rule everything around me.

[Pierre Richemond 🇪🇺 @TheOneKloud]

New on arXiv: a mathematical note on how to do diffusion directly over categorical distributions ! Introducing Simplex Diffusion - if you've ever wondered how to replace masking noise with diffusion noise, that's how. https://t.co/Zb8pRtsg6Q

[degregat @degregat]

@cronokirby I feel this is true more generally. Also one of the reasons why many researchers recommend thinking about a problem on ones own for a while _and then_ looking at the literature, to not constrain your imagination too much from the start.

[Harvard Data Science Initiative @harvard_data]

TalkToModel – an interactive dialogue system for explaining #MachineLearning models through conversations – is more effective for model explainability than existing systems, introducing a new category of explainability tools for practitioners: https://t.co/XwGcOf8X0v @arxiv #ML

[Tim Dockhorn @timudk]

📢 Excited to announce Differentially Private Diffusion Models! 🔒

https://t.co/iJRvZxnbIg

We train diffusion models with strict differential privacy guarantees and outperform previous methods by large margins.

w/ @tianshi_cao, @ArashVahdat, @karsten_kreis

(1/n)

[degregat @degregat]

@bobpoekert IIRC Firefox is using (parts of) it