Lars Roemheld

63 posts

Lars Roemheld

Lars Roemheld

@LarsRoemheld

Making machines learn, preferably about causality. Leading data science in Microsoft’s Office of the Chief Economist.

San Francisco, CA Katılım Ekim 2011
421 Takip Edilen413 Takipçiler
Lars Roemheld retweetledi
Tobias Werner
Tobias Werner@_ToFeWe·
How do humans interact with pricing algorithms within firms❓ We explore this question using two large-scale field experiments in collaboration with @ZalandoTech in a new WP. This work is a collaboration with Tobias Huelden, @VJascisens and @LarsRoemheld. 🧵Thread below.
Tobias Werner tweet media
English
3
20
68
13.5K
Lars Roemheld retweetledi
Towards Data Science
Towards Data Science@TDataScience·
"I find that 'applied science' has different expectations than research science. So I wrote up what, in my experience, helps produce great applied science work." by Lars Roemheld buff.ly/3KU3YYd
English
1
5
19
6.9K
Lars Roemheld
Lars Roemheld@LarsRoemheld·
A big part of leadership is making expectations explicit. Over the last year, I've formalized what I believe to be great applied (data) science work. Whether as a "definition of done", inspiration for personal development, or ideas to shape team culture. @lars.roemheld/5739daf13dd3?source=friends_link&sk=a446647ab7adc5c3fc7711fba37ff25c" target="_blank" rel="nofollow noopener">medium.com/@lars.roemheld…
English
9
2
9
4.1K
Lars Roemheld retweetledi
Tobias Schmidt
Tobias Schmidt@RoyalTS·
I don't know if AI will kill us all but this already seems like a crime against humanity youtu.be/HyfQVZHmArA
YouTube video
YouTube
English
1
1
1
596
Lars Roemheld
Lars Roemheld@LarsRoemheld·
@law42_DE Aussagekräftig über longitudinales Versorgungsgeschehen aber dennoch - insb. bei schlauer Berücksichtigung der Biases. Zeitlicher Verzug der Daten ist ein Problem. Die Daten sind überwiegend schon digital, werden aber durch viele Zwischenschritte verlangsamt.
Deutsch
1
0
2
0
Lars Roemheld
Lars Roemheld@LarsRoemheld·
@law42_DE Biases ja. Auf ganz konzeptioneller Ebene einerseits durch Anreize durch die Abrechnung selbst (Mehr-Umsatz durch „upcoding“), andererseits historisch wohl auch durch Risikostrukturausgleich der Kassen (Morbi-RSA). 1/2
Deutsch
0
0
1
0
Lars Roemheld
Lars Roemheld@LarsRoemheld·
"As I ate the still-living digital body of my friend in the company of a stranger in the shape of a wild dog, I thought, 'Oh. This is **special**.'" Deeply beautiful essay about finding home and meaning on the internet by @everestpipkin pioneerworks.org/broadcast/i-kn…
English
0
0
0
0
Lars Roemheld retweetledi
Corona-Warn-App
Corona-Warn-App@coronawarnapp·
Meldewochen 2020/2021 im Vergleich: Die 7-Tage-Inzidenz zeigt starken Anstieg bei den 15-34-Jährigen.😮 Angesicht der #DeltaVariante ist es wichtig, die #CoronaWarnApp (empfohlen >16) jetzt zu nutzen. Ihr könnt im Infektionsfall andere direkt warnen. ➡️rki.de/DE/Content/Inf…
Corona-Warn-App tweet media
Deutsch
15
73
203
0
Lars Roemheld retweetledi
AK
AK@_akhaliq·
VQGAN + CLIP "matte painting of someone reading papers and burning the midnight oil | trending on artstation"
AK tweet media
English
2
17
163
0
Lars Roemheld
Lars Roemheld@LarsRoemheld·
... it neatly summarizes what's happening: using 2 RandomForests, confounding noise is reduced. This comes at a cost, as we explain away and therefore lose variation in our cause of interest.
English
0
0
0
0
Lars Roemheld
Lars Roemheld@LarsRoemheld·
Auch für #DeltaVariante: Es gibt inzwischen gute Hinweise, dass die #CoronaWarnApp funktioniert! Mehr freiwillige Warnungen = langsamere Verbreitung. Mehr-Nutzer bringen wegen Netzwerkeffekten überproportional Mehr-Nutzen. Installieren hilft.
Nationale Akademie der Wissenschaften Leopoldina@Leopoldina

Digitale Werkzeuge können helfen, die #Pandemie einzudämmen. Die Leopoldina hat heute ein Diskussionspapier zum Einsatz automatisierter #Kontaktnachverfolgung, #CoronaApps und #Wearables veröffentlicht: leopoldina.org/presse-1/nachr… #Corona #Coronavirus

Deutsch
2
2
1
0
Lars Roemheld
Lars Roemheld@LarsRoemheld·
@XXVmile Yes, imaging might be helpful for diagnosis of severe cases. But for those, we have relatively little training data and even less ground-truth understanding of this very new disease. Great case for human experts, not clinical AI. (1/2)
English
1
0
0
0
Lars Roemheld
Lars Roemheld@LarsRoemheld·
Many important points. #1 issue is missing though: reliable antigen/antibody tests are cheap. Why put somebody in expensive scanners for the same diagnostic quality in the first place?
Vlad Haltakov@haltakov

Can you detect COVID-19 using Machine Learning? 🤔 You have an X-ray or CT scan and the task is to detect if the patient has COVID-19 or not. Sounds doable, right? None of the 415 ML papers published on the subject in 2020 was usable. Not a single one! Let's see why 👇

English
2
0
3
0
Lars Roemheld
Lars Roemheld@LarsRoemheld·
Don't touch the collider, and be clever about observational, interventional, and randomized data: very much enjoyed teaching a practitioner's perspective on causal inference. Thanks for having me!
Lars Roemheld tweet media
Peter Steinbach@psteinb_

What an awesome opportunity our @helmholtz_ai provided me today: a full day workshop on causal inference by 2 superb teachers & formidable experts @k__niki @LarsRoemheld! I am super grateful 🙏 and overwhelmed at the same time. ☕️ Mind the colliders! 💥⚠️

English
1
0
1
0
Lars Roemheld retweetledi
Markus Beckedahl
Markus Beckedahl@netzpolitik·
24.300 bestätigte Neuinfektionen. Viel zu viele! Und 4.449 warnende Personen über die Corona-Warn-App, die offensichtlich besser funktioniert als ihr Talkshow-Image verspricht.
Markus Beckedahl tweet mediaMarkus Beckedahl tweet media
Deutsch
6
56
261
0

Keşfet

@ZalandoTech @VJascisens @law42_DE @everestpipkin @VC31415 @XXVmile @elonmusk @BarackObama