Richard Lanas Phillips 📈📉

ML researcher at Cornell. Inequality, structured data, and responsible prediction. Amateur cricket player and computational chemist. He/him. PhD Student.

[EAAMO @EAAMO_ORG]

Hello everyone! Discrimination and Equality in Algorithmic Decision-making are taking over Twitter this week. Check us out! https://t.co/RHvgBHCfLD 1/7

[EAAMO @EAAMO_ORG]

@rlanasphillips is a Ph.D. candidate at Cornell. His interests are in the ways we make decisions around ML algorithms, e.g. how should we apply models to groups with different needs, how do model affect community development, or how algorithms augment our existing biases? 4/7

[Richard Lanas Phillips 📈📉 @rlanasphillips]

🙂👨‍💻

[ACM FAccT @FAccTConference]

Announcing the following UPDATE:

*The page limit this year is 12-15 pages, not 8-10!*

Looking forward to reading your submissions 🎉

Further details of the Facct 2022 CFP can be found here: https://t.co/t2g275U4Gi

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@kiran_tomlinson @nytimes Cool! The NYT really does a great job with pop data viz

[Alondra Nelson @AlondraNelson46]

It is pleasure to welcome Dr. Sorelle Friedler to OSTP. Her vast + distinctive experience includes AI applications ranging from materials science to criminal justice policy, and problem-solving tech innovation in industry, government, and the academy. A simply stellar colleague!

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@kdphd @WHOSTP @AlondraNelson46 @geomblog Congrats Sorelle!! A dream team to be sure.

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@PangWeiKoh @shiorisagawa It was an incredible learning experience to get to work with such a talented team! Distribution shift is a big issue and demands solutions as ML is increasingly deployed irl. ET friends, if you're up late I'd definitely recommend dropping in! 🙂

[Richard Lanas Phillips 📈📉 @rlanasphillips]

We're presenting tomorrow at 10:45 AM ET tomorrow and postering from noon to 2 ET. Please come say hi and ask questions if you're interested! https://t.co/NNl0rQ5HFx

[Richard Lanas Phillips 📈📉 @rlanasphillips]

Tomorrow I’ll be presenting my first (!) first-author grad school paper at ICML with my brilliant coauthor @HanShao16. Along with @nhaghtal and Avrim Blum, we study the mathematical concerns behind collaboratively sampling when groups have different needs (eg federated learning).

[Richard Lanas Phillips 📈📉 @rlanasphillips]

Overall, we believe that these considerations will need to be made to ensure machine learning is fair, participatory, and equitable in the long term.

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@roydanroy As in, just because you put it in the description doesn't mean disney isn't going to sue you for uploading the full star wars trilogy on youtube. You'll still get fired if you say something dumb enough on the internet.

[Pang Wei Koh @PangWeiKoh]

We’re presenting WILDS, a benchmark of in-the-wild distribution shifts, as a long talk at ICML!

Check out @shiorisagawa's talk + our poster session from 9-11pm Pacific on Thurs.

Talk: https://t.co/ebYtrHJBlH
Poster: https://t.co/ACK56HAGYn
Web: https://t.co/vA5KxsZf6c

🧵(1/n)

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@roydanroy @_kunal_talwar_ @thesasho @thejonullman I realize I’m accidentally walking into a long, ongoing (and charged?) debate, but I was hoping to understand if the wording is, like you mentioned, intentionally provocative when talking to a broad audience.

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@roydanroy @_kunal_talwar_ @thesasho @thejonullman I definitely get the fundamentals of the theory. The language around statistical inference never being a 'privacy violation' just seems potentially a little glib in a letter partially addressed to courts and the media.

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@_kunal_talwar_ @roydanroy @thesasho @thejonullman Obviously learning the smoking-cancer link or the census may have a lot of scientific legitimacy. I can imagine other instances of imputation that don't have such a good justification, however.

[Richard Lanas Phillips 📈📉 @rlanasphillips]

@_kunal_talwar_ @roydanroy @thesasho @thejonullman Isn't this one of the key breakdowns in the language though? If you can predict things via insuppressible features about someone, I can still say (in the common understanding) that your choice to aggregate the data and do inference violated their privacy.