deepbias @deepbias

a deep bias exists in the datasets used to train neural networks. deepbias.co San Francisco, CA Joined June 2020

Tweets

23
Followers

64
Following

531
Likes

1

Dan Roy @roydanroy

5 years ago

Amazing... but at the same time I'm really curious how dataset bias influences how these generated images might look. Do they look like Hollywood actors because that's what GANs are often trained on? Are they whitened for that same reason? Studying role of bias here fascinating.

hardmaru @hardmaru

5 years ago

26 532 2K 0 128

Download Image

3 4 58 0 3

Diletta Goglia 🇸🇪🇮🇹 @dilettagoglia.bsky.social @dilettagoglia

5 years ago

What's in my planner for today: this amazing session by Women in Analytics (WIA)! A seminar on impact of data representation. #WomenInAnalytics #Analytics #Bias #Data #MachineLearning bit.ly/3gvvs5t

0 2 3 0 0

deepbias @deepbias

5 years ago

theverge.com/2020/7/29/2134…

0 0 0 0 0

deepbias @deepbias

5 years ago

Equality

Andrew Ng @AndrewYNg

5 years ago

Equality

2K 516 3K 0 360

0 0 0 0 0

Rachel Thomas @math_rachel

5 years ago

ICYMI here's a talk I gave on algorithmic bias & a thread of the slides:

Rachel Thomas @math_rachel

6 years ago

ICYMI here's a talk I gave on algorithmic bias & a thread of the slides:

6 202 642 0 229

Download Image

0 35 103 0 33

deepbias @deepbias

5 years ago

Key word “subjectively”

0 0 0 0 0

deepbias @deepbias

5 years ago

“...there is arguably a connection between the biased outputs and the unreliable ones that point to a larger problem. Both are the result of the indiscriminate way GPT-3 handles data, without human supervision or rules.”

The Verge @verge

5 years ago

0 10 45 0 7

Download Image

0 0 1 0 0

deepbias @deepbias

5 years ago

This.

0 0 1 0 0

Andrew Beam @AndrewLBeam

5 years ago

Language models are known to exhibit bias and it seems that GPT-3 is no different in this regard, but it is always shocking to see. I gave it the first prompt and had it generate the rest.

27 98 420 0 17

Download Image

Mark Riedl @mark_riedl

5 years ago

A call for anticolonialism in AI from DeepMind. I’m glad this paper was written, and by a group that gets a lot of attention. I hope they walk the walk and that others take this seriously as well arxiv: arxiv.org/abs/2007.04068 venturebeat.com/2020/07/11/dee…

1 3 39 0 9

Chris Anderson @chr1sa

5 years ago

Have you noticed that people (mostly tech ppl) are starting to use the word "priors" instead of "biases"? Priors comes from Bayesean stats, so it's a bit of smarty signaling, but also reflects that "biases", which originally came from textiles, now has a pejorative connotation

16 20 143 0 10

hardmaru @hardmaru

5 years ago

Towards Learning Convolutions from Scratch “As ML moves towards reducing the expert bias and learning it from data, a natural next step seems to be learning convolution-like structures from scratch.” Would be great to find the "ConvNet" for new domains. arxiv.org/abs/2007.13657

Behnam Neyshabur @bneyshabur

5 years ago

11 160 773 0 151

Download Image

3 39 195 0 44

Download Image

Paul Graham @paulg

5 years ago

People get mad when AIs do or say politically incorrect things. What if it's hard to prevent them from drawing such conclusions, and the easiest way to fix this is to teach them to hide what they think? That seems a scary skill to start teaching AIs.

211 342 3K 0 137

Elon Musk @elonmusk

5 years ago

Will those who write the algorithms ever realize their negativity bias?

3K 8K 115K 0 424

Kyle Wiggers @Kyle_L_Wiggers

5 years ago

‘Fundamentally flawed’ study describes facial recognition system designed to identify non-binary people ift.tt/30afUgq

1 1 3 0 0

deepbias @deepbias

5 years ago

Incredible read. nytimes.com/2020/06/24/tec…

0 0 0 0 0

deepbias @deepbias

5 years ago

“Yet, bias internalized in a widely deployed AI system can be insidious — precisely because a single set of biases become institutionalized with little variance.”

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) @rao2z

5 years ago

“Yet, bias internalized in a widely deployed AI system can be insidious — precisely because a single set of biases become institutionalized with little variance.”

7 21 51 0 9

0 0 0 0 0

(((ل()(ل() 'yoav))))👾 @yoavgo

5 years ago

my main takeaway from conversations at ACL: our datasets are crap. I am not talking about MRC, which goes without saying but also improving. I'm talking about real-world stuff: NER, IE, document classification. What we have is toyish and unrealistic. How do we get good ones?

38 60 306 0 30

deepbias @deepbias

5 years ago

It’s so incredibly easy to make rapid ‘small’ decisions at the beginning of your research... ...that could potentially inflict serious societal harm. ML engineers need to consider this ‘amplification of bias’ from day one.

Rachel Thomas @math_rachel

6 years ago

1 31 89 0 7

Download Image

0 0 0 0 0

deepbias @deepbias

5 years ago

- Systematic under-representation women in search results for occupations - Object detection that detects pedestrians with higher error rates for demographic groups with dark skin tones - Lighter-skin males are classified with the highest accuracy” analyticsindiamag.com/image-datasets…