Profile

Cover photo
Suresh Venkatasubramanian
Works at U. Utah
Attended Stanford
Lives in Salt Lake City
1,061,041 views
AboutPostsPhotosYouTubeReviews

Stream

 
+Sorelle Friedler on "When big data goes bad"
Corporations are increasingly relying on algorithms to make business decisions and that raises new legal questions.
7
3
Suresh Venkatasubramanian's profile photoBala Krishnamoorthy's profile photoSorelle Friedler's profile photoCarlos Scheidegger's profile photo
2 comments
 
That is definitely a problem. And some cases has even been documented. 
Add a comment...
 
So cool. Reminds me of both the monsters Inc door scene and the toy story II airport scene.
 
Like a crazy amusement ride for your bags! There's a lot more different types of mechanisms than I would have thought.
10 comments on original post
6
Robert Ricci's profile photoSuresh Venkatasubramanian's profile photo
2 comments
 
Yeah. Be quiet, Rob :). 
Add a comment...
 
For people who think algorithmic bias isn't a real problem, are you willing to risk your ability to ever fly again ?

http://www.slate.com/articles/news_and_politics/politics/2015/08/the_u_s_government_is_putting_americans_on_its_no_fly_list_on_a_hunch_and.html
11
Arnab Paul's profile photo
 
I'm looking at your g+ post on Aug-9th to gauge when this perception may change :)   well, sort of .. 
Add a comment...
 
Nice interview. And she hits all the points about algorithmic fairness head on. 
 
A nice feature about Cynthia Dwork talking about interesting work in fairness and machine learning that she has been doing with, among others,  +Moritz Hardt +Toniann Pitassi +Omer Reingold and Rich Zemel in the NYtimes!

http://www.nytimes.com/2015/08/11/upshot/algorithms-and-bias-q-and-a-with-cynthia-dwork.html
Preventing discriminatory algorithms is an issue being taken up by computer scientists as well as policy makers, ethicists and legal experts.
View original post
8
4
Saravanan Thirumuruganathan's profile photoPhil Long's profile photo
Add a comment...
 
Because guilt by association is the best kind of guilt...
Facebook has been granted an updated patent from the U.S. Patent office on a technology that can help lenders discriminate against certain borrowers based on the borrower's social network connections.
7
2
Ian Mallett's profile photoSergey Ten's profile photoAdam Smith's profile photoGershom B's profile photo
5 comments
 
Woo–Hoo! Now people with surplus in account who don't need credits can sell their social connections to people in the red, who want to improve their social-credit balance! Get paid for likes now can have whole new meaning :)))
Add a comment...
 
apparently using safari gives you an extra hour on a macbook vs chrome. Not to mention not burning my legs
5
Suresh Venkatasubramanian's profile photoJoshua Herman's profile photoNeal Patwari's profile photoDaniel Lemire's profile photo
5 comments
 
Sadly, I had to move away from Chrome as it is just too memory and CPU intensive. On one recent Dell laptop, Chrome brought the whole PC to a crawl. When I uninstalled it, it asked whether I uninstalled because it slowed down the computer... I am curious as to what makes Chrome so intense... c.c. +Philippe Beaudoin
Add a comment...

Communities

8 communities
 
 
Data science is just a fancy word for stats. Great answer by our very own David Robinson.
1 comment on original post
8
2
Dragomir Anguelov's profile photoJeffrey Ullman's profile photoHilmar Hoffmann's profile photoCarlos Scheidegger's profile photo
2 comments
 
Data science, "big data," or whatever, is NOT coextensive with statistics.  There are things like locality-sensitive hashing, PageRank, and many other algorithms discussed in MMDS http://www.mmds.org that are neither machine learning nor any form of statistics.  (Yes, I know you could say that a freshman-level knowledge of statistics is necessary to see why some of these algorithms work, but they were neither invented by statisticians nor considered part of the statistics curriculum.)
Add a comment...
 
My student +Amirali Abdullah just defended his Ph.D today ! And now he's off to Michigan to do a postdoc with +Anna Gilbert as a Hildebrand fellow. Congratulations !
19
Add a comment...
1
2
Robert Ricci's profile photoTruyen Tran's profile photoBhaskar Krishnamachari's profile photo
 
When I see these sorts of conversations, and they start with some kind of discussion of "competitive advantage" with other sciences, I tune out immediately. There may in fact be some actual research benefits to be had from changing the publishing system we have in CS, and I'd be interested in that sort of conversation. But the idea of changing publishing systems for what amounts to political reasons...
Add a comment...
 
One of the biggest challenges in scientific outreach is that by the time received scientific wisdom has firmly entered the public consciousness, it has been outdated (or worse, wrong) in the community where it came from.

Things change, but the fundamentals change more slowly. Learn the fundamentals. Be receptive but skeptical of all else. 
6
1
Heinrich C. Kuhn's profile photo
Add a comment...
 
 
Preserving validity in adaptive data analysis

With all data analysis, there is a danger that findings observed in a particular sample do not generalize to the underlying population from which the data were drawn. Adaptive analysis of a data set - where the analyst is informed by data exploration, as well as the results of previous analyses of the data set - can lead to an increased risk of spurious discoveries that are neither prevented nor detected by standard approaches. 

In order to increase the reliability of data driven insights, researchers from Google, Microsoft, IBM, the University of Toronto, Samsung and the University of Pennsylvania have introduced the reusable holdout mechanism, a new methodology for navigating the challenges of adaptivity which allows the analyst to safely validate the results of many adaptively chosen analyses without the need to collect costly fresh data each time. 

Learn more on the Google Research blog, linked below.
3 comments on original post
3
3
Ken Clarkson's profile photoHeinrich C. Kuhn's profile photoRasmus Pagh's profile photoHilmar Hoffmann's profile photo
Add a comment...
 
Very well put. I could not agree more. 
 
This is one of those ideas which is superficially good and becomes terrible if you look at it more closely. The concept is that in sentencing, we should do a risk assessment, based on objective criteria rather than on the prejudices of the individuals in the court, and use that to influence the decision. Those more likely to re-offend, in this system, would receive harsher and longer sentences.

The first problem with this is that if you build a 100% honest model of re-offending probability, what you're building is a model of your own system, not of the person. For example: If you lock someone in prison for a few years, offering them no training or rehabilitation, and then upon release have various penalties on them which basically prevent them from getting a job -- ranging from the simple "nobody wants to hire someone with a conviction record" to "explicit legal bars to their getting certain kinds of job, living in certain areas, etc" -- then it will probably not surprise you that this person is significantly more likely to turn to a professional life of crime. A model which correctly recognized and predicted this would therefore conclude that the only solution is to lock this person up for life, since at any point after they're released, they're simply likely to become criminals again.

This highlights the deeper problem in such a model, of course, which is that its basic design parameters, where the only variable it controls is "imprison more" or "imprison less," create a false dichotomy: rather than asking "which course of action is most likely to lead to the person no longer engaging in crime," it only considers one possible course of action, and that action (again, by the design of the system) most often increases the probability of future crime. 

The criticism of this system that it will end up encoding implicit racial biases is only sort-of correct. This model will definitely end up having a strong racial component; even if you eliminate race as an input, your race is so strongly correlated to other things like where you live that the system will end up modeling your race, and basing its decisions upon that, one way or the other. And that will, indeed, end up increasing sentences for Black and Latino offenders, for all the reasons specified above.

But in this case, the racial biases which the system would acquire are simply one manifestation of the even deeper and more profound problem that this model is simply designed to optimize for throwing people into prison.

If you want a variation on this which actually works, give the model access to a wide range of possible consequences, and ask it which of those will minimize the odds of re-offense, presumably balanced against various costs. You'll almost certainly find that rehabilitation, training, and treatment overwhelmingly work best to minimize that. (And in the cases where they don't, your best bet is likely to simply take them out and shoot them)

I would actually quite strongly favor such a project, because it would require its creators to make very explicit the thing for which they are trying to optimize. You can't lie to a computer about what you want it to do; if you want to minimize the chance of re-offense, you have to tell it to do so. If you instead want to optimize the system for retribution, or to cow a broad population into submission, or to maximize revenue, the model will absolutely be able to do that as well -- but you would have to tell it explicitly to do so, and it's very hard to lie to yourself about that.

h/t +Amy Quispe over on Twitter for prompting me to actually write about this one.
Interactive graphics by Matthew Conlen, Reuben Fischer-Baum and Andy Rossback This story was produced in collaboration with The Marshall Project. Criminal sentencing has long been based on the pres…
68 comments on original post
2
2
Neal Patwari's profile photoSuresh Venkatasubramanian's profile photoVijay Chidambaram's profile photoCristóbal Guzmán's profile photo
5 comments
 
I like that line: but I'm not sure how to interpret it. 
Add a comment...
Communities
8 communities
Work
Occupation
Researcher
Employment
  • U. Utah
    Associate professor, present
Places
Map of the places this user has livedMap of the places this user has livedMap of the places this user has lived
Currently
Salt Lake City
Previously
Aarhus, DK - Stanford, CA - Philadelphia, PA - Morristown, NJ - New Delhi, India - Berkeley, CA
Story
Introduction
CS prof, interested in algorithms, geometry, data mining, clustering
Education
  • Stanford
  • IIT Kanpur
Basic Information
Gender
Male
Other names
geomblog
That this restaurant has an overall rating less than 4 is a travesty. Skip Finca's overrated food and preserve your ear drums: Cafe Madrid is a much more intimate (read: quiet and charming) Spanish fine dining experience, with possibly the best service I've ever had in Salt Lake City. Call ahead if you want the paella: it's worth it.
Public - 6 months ago
reviewed 6 months ago
3 reviews
Map
Map
Map