Pages

26 June, 2008

"If By a Man's Works Shall We Know Him...

...then you, sir, are a colourful, nine-dimensional blob."

A quote from a colleague. I take it as a compliment. If you can't make pretty graphs, what's the point of science?

EVD


Apologies for my absence - recent days are mostly spent taking tea in a rough-and-ready way while writing up. With a new position starting for me next week (though happily in the same group), my thesis-writing time will be squeezed into evenings and week-ends alone, so I don't see much chance of improvement in the immediate future. Your patience is appreciated!


Toodlepip,

Hobbes

16 comments:

  1. Thanks for the update! Good luck with your writing.

    Brent

    ReplyDelete
  2. doughnuts with sugar sprinkles..

    what is your theme of research?


    robert

    ReplyDelete
  3. looks kind of Eigen-ish? john

    ReplyDelete
  4. Thanks for the good wishes, I really appreciate them. :)

    The general field is variously known as:

    "Artificial Intelligence"
    "Information Engineering"
    "Signal Processing"
    "Machine Learning"
    "Pattern Analysis"
    "Statistical Pattern Recognition"

    ...depending on which department of which institution you're in. Our name is Legion, for we are many. ;)


    Toodlepip,

    Hobbes

    P.s. John, the graph is an Extreme Value Distribution for a three-dimensional Gaussian Mixture Model of three components, with general covariance matrices. If the model represents "normality", it effectively describes where the extreme bounds of "normality" lie, so that you can identify extreme events (stock crashes, aircraft engine problems, patient organ failures, etc.).

    ReplyDelete
  5. Sooo..."normality" consists of two orthogonal distorted doughnuts centered around the origin, with an undistorted parallel doughnut in positive space and a more eccentrically distorted parallel doughnut in negative space?

    ReplyDelete
  6. Uh-oh, it's CB, better tighten up my loose techspeak. :)

    The image is just a few slices through a four-dimensional blob (contourslice.m in Matlab, if you use it) to give the general impression - of course, a probability distribution defined over 3-D is a 4-D quantity, so it's hard to visualise holistically. I had to slice the lumpy model, so you're looking at a few cross-sections taken at orthogonal axes.

    "Normality" here is a trimodal GMM: three hyperellipsoids that sit within the middle of the doughnuts. Like three American footballs of different stretch and width, fused together into a lumpy bundle. :)


    Toodlepip,

    Hobbes

    ReplyDelete
  7. Allright, I'm off to tape 3 footballs together and see if that will help me predict what's going to happen to our crippled economy.

    ReplyDelete
  8. If the number of dimensions is the least it takes to locate yourself, need a lot less than that. Can't hardly find where at I am no how with just three (time is an illusion so it does not count).

    ReplyDelete
  9. All this talk of doughnuts and footballs...by far the most ridiculously beautiful rant I've seen in a very long time.

    ReplyDelete
  10. If one were to send you some samples where could they send them?

    Peace

    ReplyDelete
  11. Dear Matt,

    That's very kind of you, thank you. I'm always happy to swap tea-samples. Do please e-mail me (hobbesoxon 'at' gmail 'dot' com), or leave your e-mail address here and I'll contact you.


    Toodlepip,

    Hobbes

    ReplyDelete
  12. Aha, you use free photobucket, which has now kicked you out for exceeding your bandwidth. Obviously, you're getting too many hits for your own good :)

    I pony up the money every year for a pro account. It does make photo hosting easier to handle.

    ReplyDelete
  13. I considered it, but apparently the Photobucket bandwidth counter resets on the 4th of each month for my account... why is, by sheer coincidence, today. So let's see if all the images return. Fingers crossed.


    Toodlepip,

    Hobbes

    ReplyDelete
  14. Indeed, they do

    But your bandwidth usage will only go up, not down :)

    Although, one way to avoid it -- if there's a way for you to reduce the number of posts you show on your blog's first page, then you will reduce your bandwidth usage significantly without needing to pay them. After all, I doubt most of your readers are going 40 days back to read old entires there -- if they really want, they can always go dig it up.

    ReplyDelete
  15. That's a clever idea, gracias.

    ReplyDelete

(and thanks)