digg

Facebook Connect Join Digg About

MIT model helps computers sort data more like humans

web.mit.edu — Humans have a natural tendency to find order in sets of information, a skill that has proven difficult to replicate in computers. Faced with a large set of data, computers don't know where to begin -- unless they're programmed to look for a specific structure, such as a hierarchy, linear order, or a set of clusters.

Who dugg this? Made popular Sep 1, 2008

Digg DudeWhat is Digg?

Digg is a place for people to discover and share content from anywhere on the web. Digg surfaces the best stuff as submitted and voted on by the community. Take the Tour to learn more.

36 Comments

show profanity settings
expand all
  • nicktx, on 08/31/2008, -3/+21The average human can't properly sort their laundry, shouldn't MIT be instead teaching the students to sort data more like computers?
  • inactive, on 09/01/2008, -0/+12I put whites and darks together when I do laundry.
  • jc7012, on 09/01/2008, -0/+11And who says models are all looks and no brains?
  • arjie, on 09/01/2008, -0/+9Ah, civil rights for clothes. Nelson Mandela is proud of you, son.
  • katieedwards, on 09/01/2008, -0/+8I think a couple people are missing the point here. Sorting is one of the fundamental bases of Computer Science and many algorithms in all sorts of applications do some kind of sorting. There are many well known algorithms that can sort a set of data and we are always looking for the fastest/most efficient way of doing this. Thing is, some algorithms are really fast on certain sets of data and really slow on others. Humans subconsciously use variants of these algorithms and we're pretty good at recognizing what the data we have looks like and what technique we're going to use to make sense of it. This research is aiming for a way to get a computer to pre-analyze data in the same way we do.
  • ASSASSYN360, on 09/01/2008, -0/+5Great now my future PC will forget things.
  • thedez, on 09/01/2008, -1/+6Column A: porn
    Column B: everything else
  • Relikh, on 09/01/2008, -0/+4WHAT? NOT IN MY COUNTRY!
  • suinmind, on 09/01/2008, -0/+3So the algorithm brute-force searches through a list of structures.

    Yeah, right, big break-through. If the data has a single/separable structure; if the lists is complete; if the structure is static over time......

    Being busy publishing "break-through" papers and fighting for funds everyday destroy any hope for a potential scientific break-through.
  • missingnoh4x, on 09/01/2008, -0/+3Somehow I imagine that simulating the way humans look at these problems would only take more CPU time...
  • Cappez, on 09/01/2008, -0/+2Here is the article: http://www.psy.cmu.edu/%7Eckemp/papers/kempt08.pdf
    (btw the author is not at MIT, he moved to Carnegie Mellon University)
  • drpaidout, on 09/01/2008, -1/+3This can help with predicting what humans are more likely to do eg in the context of marketing ...
  • maninalift, on 09/01/2008, -0/+2Bloody press release bollocks. Next time point me to something written by the researcher (web page, article...).
  • inactive, on 09/01/2008, -0/+2How is that annoying?
  • ratboy4001, on 09/01/2008, -0/+1Dijkstra would be disappointed.
  • Culyt, on 08/31/2008, -2/+3I would prefer a way to help me sort more like a computer...

    There are a few annoyance with computer sorting, namely dates on in filenames, you need yyyy-mm-dd for proper ordering.
  • kalos, on 09/02/2008, -0/+1The link ended in .pdf, what part of that wasn't clear in indicating that it was a PDF link?
  • KingGorilla, on 09/01/2008, -0/+1R2-D2????
  • tomega, on 09/01/2008, -0/+1Digg.com needs that. Just try to search for an entry which was a minute ago on the fornt page.
  • Taiyoryu, on 09/01/2008, -0/+1Did anyone else hope the MIT model was a female grad student who happens to model on the side to earn tuition? Brainy women are so sexy.
  • catxors, on 09/01/2008, -0/+1Thank you. The press release was remarkably content-free.
  • catxors, on 09/01/2008, -0/+1Your one-sentence description is essentially correct, but actually getting it to work is really hard. You have come up with the right probability distributions for scoring a whole bunch of different kinds of structures, making sure they are internally consistent *and* yield good results when applied. Also, the search space is huge, so you can't search it all. You have to come up with an efficient algorithm and/or structure the probability model so a simple algorithm works well.

    Yes, this is not the crazy breakthrough the press release touts it to be. It is an application of standard statistical learning theory. But I myself have done research of this type and to me this data set learning is very good work.

    Oh yeah, and they're not exactly simulating human reasoning. Rather, they've (a) produced an algorithm for something that has been harder for AI than many other problems that (b) plausibly has some resemblance to human intuition.
  • LibertyForever, on 09/02/2008, -0/+1From a total dreamer: why use algorithms to solve those problems - I mean, structured, predefined, predictable programming? Shouldn't we be solving those problems using self-organizing, self-programming artificial intelligence systems that work 24 hrs a day and get better and better at solving those problems? Maybe our brain doesn't use rigid algorithms - I doubt my brain does ;) I'm no scientist, but why don't we emulate human brains using some sort of brain-like chaos chain of events - and give our artificial systems a chance to sort data BETTER than humans?

    I don't want a machine whose sole ambition is to emulate me - and even less the people I disagree with! Hoping there is a scientist out there deigning to elaborate on this.
  • flarn2006, on 09/01/2008, -0/+1You know, it's good netiquette to warn people when you are giving a PDF link. Not everybody uses Foxit Reader! (Although I do.)
  • AdamsGuitar, on 09/01/2008, -1/+1In order to properly compare "human" sorting to algorithmic sorting, you have to remove a human's ability to visualize. If you had a giant pile of laundry that you couldn't see and were handed each garment one at a time, you wouldn't have any problem sorting them just like a computer and you'd be slower. The ability to visualize and see, say, a grouping of colored clothes that you can grab at once can't be easily duplicated in the realm of computers without a lot of parallel processing. Obviously the analogy breaks down under too much scrutiny, since in reality a computer could sort your laundry a lot faster than you could, but the point is still true.
  • p3ngwin, on 09/01/2008, -1/+1computers are nototiously bollocks at certain pattern recognition, unlike humans that can recognize faces, pictures and many other types with ease since birth.

    try conversing with someone in a noisy crowded room, and realise how you "naturally" do it with ease, compared to a computer that would have great difficulty.

    try getting a computer to understand what an "optical illusion" is, and then understand it's having difficulty because it can't "see" what the picture is supposed to be, hence the illusion fails.
  • blacktriangle, on 09/01/2008, -1/+1HAL-9000?
  • AdamsGuitar, on 09/01/2008, -0/+0If you're sorting lexicographically, then yes, you do need the date represented in that format. Fundamentally, though, sorting in a computer comes down to the ability to compare two pieces of data to determine equality or relative inequality (ie, which is larger). Using the date example, actual dates are stored in numerical units, normally "ticks" since a particular starting date. The particular date format, such as the yyyy-mm-dd format or the US mm/dd/yyyy format are all produced by applying those "ticks" to a specific calendar and producing the correct year, month, day (along with hour, minute, second, millisecond, etc.). Those particular numbers can then be formatted into the display method that you like to see. When a computer compares what it knows to be an actual date, then it simply compares the number of ticks in each date and returns the date with more ticks as being greater than the other. The display format is irrelevant.
  • AXNJAXN, on 09/01/2008, -1/+1Does the new human sort have order of n * "whatever" complexity?
  • MadHarvey, on 09/01/2008, -0/+0It would be a breakthrough, if they could take arbitrary sets of unstructured data, find patterns within the data, and apply a data structure that fits best with those patterns.

    If they are just applying data structures one by one, and calculating which one worked best, yeah, not so much a break through.
  • filldeviant, on 09/01/2008, -1/+1T-800?
  • homersaysdoh, on 09/01/2008, -1/+1Wouldn't that be taking a step backwards?
  • inactive, on 09/01/2008, -2/+1I was going to digg you up, till you started suggesting EVIL purposes
  • fuzzy889, on 09/01/2008, -1/+0humans can't recognize ***** at birth
  • GelfTheElf, on 09/01/2008, -1/+0Umm.. didn't we already show computers "Selection Sort" which is how humans sort data...
  • mambanamba, on 09/01/2008, -3/+1one step closer to i,robot

  • Add a Comment

    Login or register to comment!
    Or, sign-in super fast with your Facebook account ...

© Digg Inc. 2009 — Content created and posted by Digg users is dedicated to the public domain | Terms of Use Updated | Privacy Policy
DIGG, DIGG IT, DUGG, DIGG THIS, Digg graphics, logos, designs, page headers, button icons, scripts, and other service names are the trademarks of Digg Inc.