I kind of agree with this downvoted comment. We need more gender diversity in anime-ML projects. The anime dataset bias issue should be a challenge, rather than an excuse (cc: @gwern @kvfrans) https://t.co/MprP1NLLkM https://t.co/Ly2QCpHUpl
— hardmaru (@hardmaru) July 24, 2019
Thread
— Guillaume Chaslot (@gchaslot) July 14, 2019
My first op-ed in @WIRED: how the AI feedback loops I helped build at YouTube can amplify our worst inclinations, and what to do about it.https://t.co/0rpIkDT264
1/
Introduction to Algorithmic Bias: covers types of basis, debunks common myths about bias, and shares steps towards addressing biashttps://t.co/jv53akGMfI pic.twitter.com/8CLVmTG3Co
— Rachel Thomas (@math_rachel) July 8, 2019
Fairwashing: the risk of rationalization. It it is possible to forge a fairer explanation from a truly unfair black box. Paper presented at #ICML2019 by @umaivodj https://t.co/c4PGpAXEAT pic.twitter.com/ePsU5lDHeB
— Rachel Thomas (@math_rachel) June 16, 2019
Nice new work from FAIR on assessing bias in datasets.
— Yann LeCun (@ylecun) June 7, 2019
Turns out image recognition systems will recognize everyday objects more reliably if the picture was shot in developed countries than... https://t.co/Nq2ErHND15
This doesn't surprise me. Algorithms match jobseekers to jobs in ways that propagate past practices. STEM jobs will go to men, daycare jobs to women, and we'll see tons of automated ageism. https://t.co/2yMG0BOEqL
— Cathy O'Neil (@mathbabedotorg) June 7, 2019
Object Detection systems (commercial and academic) are trained on biased data
— Soumith Chintala (@soumithchintala) June 7, 2019
This disproportionally affects accuracy in lower-income households and continents like Africa & Asia.
Work by my colleagues at FAIR using the Dollar Street dataset from GapMinderhttps://t.co/gdm7Ar9f4l pic.twitter.com/vmtkh3i4eo
Data Statements for NLP: a proposal to help address issues of exclusion, overgeneralization, underexposure, & reproducibility in NLP @emilymbender https://t.co/kWaDqa7Z64 pic.twitter.com/k6c56xHRVI
— Rachel Thomas (@math_rachel) June 6, 2019
Algorithms aren’t some pure math form that falls ready-formed from above in etched stone tablets. Mathematical applications are subject to bias!
— Angela Bassa (@AngeBassa) June 5, 2019
For instance, 3/5 is just math. Counting someone as three fifths of a person IS NOT. https://t.co/iLxasZyXZc
The paper argues that rather than attempting to debias embeddings, bias should be addressed at the point of action/decision in a system https://t.co/OPBNV501MI
— Rachel Thomas (@math_rachel) May 31, 2019
Was just rereading @aylin_cim @j2bryson @random_walker paper on bias in word embeddings. They use "small baskets" of words (from heavily cited psychology papers) to represent a concept, and compare the distance/similarity between different concepts.
— Rachel Thomas (@math_rachel) May 31, 2019
https://t.co/vkhxUuiJ8N https://t.co/OPBNV501MI
Cognitive Bias Codex-- chart of 188 cognitive biases, organized by the problem they're trying to solvehttps://t.co/31SoY54ppS@buster @jm3 h/t @purpultat pic.twitter.com/SfNwO9bufO
— Rachel Thomas (@math_rachel) May 26, 2019
