JavaScript is not available.

Nov 30, 2022

we test 3 academic and 3 commercial face detection models on 4 different datasets which we corrupt using the imagenet-c paradigm. (1) we see that all models on all datasets are very susceptible to these edge cases of noisy or corrupted images.

Show this thread

1

...or tell me who is your favorite workboat!!! ⛴️⛴️⛴️⛴️⛴️⛴️ mine is miss katie (for obvious reasons) but i'm open to other suggestions 2/2(workboats)

Show this thread

Bias persists in face detection systems from Amazon, Microsoft, and Google

Arthur

@itsArthurAI

Nov 21, 2022

📝: Robustness Disparities in Face Detection 👥:

,

,

, and

📆: NeurIPS, Datasets & Benchmarks Track

1

2

4

this work was done with the wonderful

@gzhihongwei

@tomgoldsteincs

@johnpdickerson

this work has been in the making for a while and i'm so grateful for all the collaboration!

GIF

read image description

ALT

we also provide some recommendations to industry and the public sector about how our findings and methodology can be used and adapted. you can also read more in this article by

@Kyle_L_Wiggers

from a while back about an early release of this study

venturebeat.com

A new study from the University of Maryland finds evidence of bias in face detection algorithms from Google, Microsoft, and Amazon.

1

and (3) both academic and commercial face detection systems exhibit a similar degree of their sociodemographic bias. which might be (not) surprising and disappointing given the resources and commitments made by companies to address performance disparities in face rec

GIF

read image description

ALT

1

(2) in almost all cases, we find that images of individuals who are masculine-presenting, older, darker skinned, or dimly lit fare worse than their counterparts in face detection models

GIF

read image description

ALT

1

we test 3 academic and 3 commercial face detection models on 4 different datasets which we corrupt using the imagenet-c paradigm. (1) we see that all models on all datasets are very susceptible to these edge cases of noisy or corrupted images.

2

since facial detection is everywhere, it get used in a variety of edge cases — like settings where the image is noisy or corrupted we ask (1) how robust is face det to noise, (2) are there demographic disparities, (3) are commercial providers better than academics?

GIF

read image description

ALT

1

🚨neurips poster thursday: 4-6, hall j #1023🚨 how can we audit algorithms for their performance in edge cases in real world use cases? stop by our poster tomorrow afternoon to learn more! arxiv.org/pdf/2211.15937

2

1

9

maybe you’re feeling like this heading into the neurips deadline but if ms patti can still stand and deliver an iconic performance, so can you 📈 youtu.be/z84QdJlPpHE?t=

GIF

read image description

ALT

6

Veronica A. Rivera

@veronrivera

May 4, 2022

Really interesting work by

at #CHI2022 on the adoption of Covid-19 apps. It was eye opening to learn that collective good appeals are significantly more effective at encouraging adoption than individual appeals.

1

5

23

May 4, 2022

i'll be giving my #chi2022 talk at 3:00 (central) in room 296 stop by to see me lip sync for my life and ponder how we make privacy decisions when they can serve a collective goal programs.sigchi.org/chi/2022/progr

GIF

read image description

ALT

6

22

a fantastic talk by

at #chi2022 insightful and important findings that women gig workers are not getting necessary support by gig platforms which generally make one size fits all design decisions. gender agnostic choices cause specific harms for these users

1

5

13

Great to see

on Twitter! If you are in New Orleans for #chi2022 later this month, pop by his talk looking at privacy assurances and COVID app adoption, & say hello! Teaser: youtube.com/watch?v=6CmdBg His first-author work is joint with

and

Quote Tweet

do you think it is better to include reassurances about privacy when talking about apps which use sensitive data? in two weeks, i'll be at #chi2022 to talk about what we learned about human behavior from using different messages to advertise covid contact tracing apps twitter.com/eredmil1/statu…

Show this thread

3

7

Elissa M. Redmiles, Ph.D.

@eredmil1

Apr 20, 2022

Woo

is on twitter & headed to #chi2022 in person for his first-ever in-person conference. Check out the talk at CHI & meet Samuel :)

Quote Tweet

Apr 20, 2022

do you think it is better to include reassurances about privacy when talking about apps which use sensitive data? in two weeks, i'll be at #chi2022 to talk about what we learned about human behavior from using different messages to advertise covid contact tracing apps twitter.com/eredmil1/statu…

Show this thread

1

4