Profile cover photo
Profile photo
Suresh Venkatasubramanian
I like algorithms. And I hope they're fair.
I like algorithms. And I hope they're fair.
About
Suresh's posts

Post has attachment

Post has attachment

Post has attachment

Post has shared content
The SOCG 2017 call for papers is out. 300-word abstracts are due November 28; papers are due December 5.




Post has shared content

Post has attachment
As part of the research we're doing in algorithmic fairness we're looking to hire a post-doctoral researcher who can help us bridge the gap between the more technical aspects of algorithmic fairness and the ways in which this discussion informs and is informed by the larger context in the social science.

+Sorelle Friedler

https://algorithmicfairness.wordpress.com/2016/10/03/post-doc-in-fairness-at-data-and-society/

Post has attachment

Post has attachment
21st century living, in a nutshell.

Post has shared content
Moritz tackles a very important issue in this post: defining goals for fairness in Machine Learning. It's trickier than it seems, but a very important topic.
Many have given up on the very concept, often arguing that machine learning isn't biased, but merely mirrors our own biases by fitting models to 'naturally occurring' biased data. That's simply not true: machine learning applied without care is inherently biased toward where there is a lot of data in the first place: think groups of people who consume more, or ethnicities that show up more often on medical records. It is something we can control and characterize however, as long as we have the proper definitions in place of what it means to a probabilistic model to be 'fair'.

Post has shared content
What's also nice about his post is the distinction between information obtained from you directly and information obtained about you.

DP doesn't protect the latter. And his complaint is in how people expect it should.

While this is true,I think it's a distinction that has taken some time to emerge (because privacy is such a slippery notion) and in some ways DP has helped emphasize this distinction. So a possibly better line of "attack" as it were might be to spell this out more clearly (if it already hasn't been done so) instead of playing a valiant but losing game of whack-a-paper :)
Another great post by Frank McSherry resolving some misconceptions about the guarantees of differential privacy under correlations between data entries. My favorite passage:

"By way of analogy, we use cryptography to protect our financial information, even though there is probably lots of correlation among our bank accounts. If I see you moving into your new house and call to ask about your mortgage, because often people with new houses have mortgages, have we violated the security guarantees of the cryptosystem your bank uses? Is it now time to write a paper: "cryptography vulnerable to correlated data", announcing that to provide its guarantees, modern cryptography assumes all messages are independent? Please don't."

https://github.com/frankmcsherry/blog/blob/master/posts/2016-08-29.md
Wait while more posts are being loaded