If you can read this, either the style sheet didn't load or you have an older browser that doesn't support style sheets. Try clearing your browser cache and refreshing the page.

(New Scientist) Cool "Matrix Multiplication", a key theoretical mathematical tool for solving problems in physics, economics and science, makes first major discovery in 24 years. I was told there would be no math   (newscientist.com) divider line 61
More: Cool  
•       •       •

6827 clicks; posted to Geek » on 09 Dec 2011 at 11:07 AM   |  Favorite    |   share:  Share on Twitter share via Email Share on Facebook   more»   |    Get this fabulous T-Shirt and impress the methane out of your friends! shirt it!



61 Comments   (+0 »)
   

First | « | 1 | 2 | » | Last | Show all
 
2011-12-09 11:06:08 AM
I took the Blue Pill so I'm getting a kick out of these replies.
 
2011-12-09 11:11:39 AM
Whoa. I know Kung Fu.
 
2011-12-09 11:12:18 AM
I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.
 
2011-12-09 11:16:00 AM
cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

I remember doing them in college, with variables, in matrices 4x4 and higher. THAT sucked.
 
2011-12-09 11:19:58 AM
I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.
 
2011-12-09 11:21:02 AM
Oh god, not another shiatty sequel.
 
2011-12-09 11:21:43 AM
Hollie Maea: I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.


Matlab is the most amazing program ever. Days of drudge work reduced to hours, sometimes minutes.
 
2011-12-09 11:21:54 AM
"Matrix Multiplication", a key theoretical mathematical tool for solving problems in physics, economics and science, makes first major discovery in 24 years. I was told there would be no math was no spoon"
 
2011-12-09 11:26:11 AM
mainstreet62: Hollie Maea: I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.

Matlab is the most amazing program ever. Days of drudge work reduced to hours, sometimes minutes.


I use R for all my heavy math work personally.
 
2011-12-09 11:27:01 AM
mainstreet62: Hollie Maea: I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.

Matlab is the most amazing program ever. Days of drudge work reduced to hours, sometimes minutes.


This. MATLAB has been consuming my life lately. The most fun I've had recently with it is reducing 105,000 data points (solar output for a year at 5 minute intervals) into an equation with 21 numbers as inputs.
 
2011-12-09 11:27:20 AM
cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Number theory is just a theory, just like evolution is just a theory.
 
2011-12-09 11:27:29 AM
I recall a course titled Differential Equations & Matrix Methods. I now realise the wisdom of choosing a career that does not require regular usage of such maths skills.
 
2011-12-09 11:29:18 AM
entropic_existence: mainstreet62: Hollie Maea: I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.

Matlab is the most amazing program ever. Days of drudge work reduced to hours, sometimes minutes.

I use R for all my heavy math work personally.


I used to use Mathematica, but it varied so wildly from computer to computer, and I had a TA that sucked so badly at teaching it, that I gave up on it.
 
2011-12-09 11:30:06 AM
They found the one?
 
Bf+
2011-12-09 11:30:40 AM
planet-quacksilver-forums.1049324.n5.nabble.com
 
2011-12-09 11:35:08 AM
mainstreet62: I used to use Mathematica, but it varied so wildly from computer to computer, and I had a TA that sucked so badly at teaching it, that I gave up on it.

I've just picked up R in grad school. It's been pretty useful. I only ever used MATLAB in one course in my undergrad.
 
2011-12-09 11:37:53 AM
Lord Dimwit: cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Number theory is just a theory, just like evolution is just a theory.


You're probably trolling, but I'll just say for the record that mathematical theory and scientific theory are two whole different things.
 
2011-12-09 11:38:21 AM
Me. Me, me, me, me, me.

/me too.
 
2011-12-09 11:40:17 AM
Hollie Maea: The most fun I've had recently with it is reducing 105,000 data points (solar output for a year at 5 minute intervals) into an equation with 21 numbers as inputs

Whoa! You do know that 21 is half of 42?

Another five million years, and you might just have the answer!
 
2011-12-09 11:40:36 AM
Awesome! Anything that can help speed up my Artificial Neural Networks (and SVMs, etc) is cool by me!
 
2011-12-09 11:45:39 AM
torusXL: Lord Dimwit: cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Number theory is just a theory, just like evolution is just a theory.

You're probably trolling, but I'll just say for the record that mathematical theory and scientific theory are two whole different things.


I wasn't attempting to troll, I was attempting to be funny.

/ "attempting" apparently being the operative word
 
2011-12-09 11:45:55 AM
torusXL: Lord Dimwit: cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Number theory is just a theory, just like evolution is just a theory.

You're probably trolling, but I'll just say for the record that mathematical theory and scientific theory are two whole different things.


But that's just, like, your opinion, man.
 
2011-12-09 12:06:26 PM
evilmrsock: torusXL: Lord Dimwit: cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Number theory is just a theory, just like evolution is just a theory.

You're probably trolling, but I'll just say for the record that mathematical theory and scientific theory are two whole different things.

But that's just, like, your opinion, man.


Must....not....bite......

Phew. I was able to resist.
 
2011-12-09 12:07:02 PM
Hollie Maea: I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.


I'm working with MATLAB right now. It is indeed awesome.

Doing my calculations on a Core 2 Quad Q6600 (in my desktop that I built a few years ago) is practically instantaneous. Doing them on an Intel Atom N550 (my netbook I bought last year, and occasionally use for MATLAB when away from my desktop) is bloody painful.
 
2011-12-09 12:16:30 PM
cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Me too. And that was in 1981.

Just sayin'.
 
2011-12-09 12:52:27 PM
The constant factor is probably so high that this algorithm won't be used in practice.

That's the case with the current Coppersmith-Winograd algorithm.

What?
 
2011-12-09 12:57:37 PM
Hollie Maea: The most fun I've had recently with it is reducing 105,000 data points (solar output for a year at 5 minute intervals) into an equation with 21 numbers as inputs.

nlinfit?

What do you mean the data doesn't follow first principles? Just add more @#(* terms!
 
2011-12-09 01:09:39 PM
darkscout: What do you mean the data doesn't follow first principles? Just add more @#(* terms!

This.

Actually, my equation is based on a repeating gaussian curve, and the 21 numbers are various polynomials to quantify the variation so I can add it back -normally-distributed-randomly. There's nothing first principles about it, for sure. And it's messy as hell and pointless (I was complaining to my professor about the slowness of file accesses and he jokingly said "well you can always just fit an equation to the data" and I took it as challenge).
 
2011-12-09 01:34:41 PM
Hey they can actually make that living earth simulator now, right?
 
2011-12-09 01:48:59 PM
ko_kyi: Me. Me, me, me, me, me.

/me too.


www.starstore.com
 
2011-12-09 02:22:01 PM
They discovered the truth, that there is no spoon.
 
2011-12-09 02:32:57 PM
torusXL: evilmrsock: torusXL: Lord Dimwit: cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Number theory is just a theory, just like evolution is just a theory.

You're probably trolling, but I'll just say for the record that mathematical theory and scientific theory are two whole different things.

But that's just, like, your opinion, man.

Must....not....bite......

Phew. I was able to resist.


You have a high capacitance for that.
 
2011-12-09 02:45:09 PM
I did matrix multiplication in C, using tiling. Does that count?
 
2011-12-09 03:25:44 PM
Donnchadha: torusXL: evilmrsock: torusXL: Lord Dimwit: cgraves67: I remember multiplying matrices in high school. I'm pretty sure it wasn't just a theory.

Number theory is just a theory, just like evolution is just a theory.

You're probably trolling, but I'll just say for the record that mathematical theory and scientific theory are two whole different things.

But that's just, like, your opinion, man.

Must....not....bite......

Phew. I was able to resist.

You have a high capacitance for that.


It was developed by Don Coppersmith and Shmuel Winograd and put omega at 2.376.

Math has the OMEGA.
 
2011-12-09 03:41:58 PM
mainstreet62: Hollie Maea: I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.

Matlab is the most amazing program ever. Days of drudge work reduced to hours, sometimes minutes.


I'm jealous...and so old my undergrad matrix algebra book doesn't even mention Strassen.

jake3988: I did matrix multiplication in C, using tiling. Does that count?

Jimmy-come-Lately...i did matrix algebra using APL after graduating and getting my first job.
 
2011-12-09 04:25:39 PM
StoneColdAtheist: using APL

*takes a step back off your lawn*
 
2011-12-09 04:33:43 PM
StoneColdAtheist: mainstreet62: Hollie Maea: I remember doing it 10 minutes ago. Well, making MATLAB do it for me.

/Working on an optics model.
//n^2.373 won't make my slow-ass laptop suck any less.

Matlab is the most amazing program ever. Days of drudge work reduced to hours, sometimes minutes.

I'm jealous...and so old my undergrad matrix algebra book doesn't even mention Strassen.

jake3988: I did matrix multiplication in C, using tiling. Does that count?

Jimmy-come-Lately...i did matrix algebra using APL after graduating and getting my first job.


I think I'm the only person I know under 35 who has any real experience with APL. It's...beautiful in a way. I just wish Inconsolata had APL symbols.

I've made a decent side-career of knowing obscure and/or old technologies. I'm 31, but I've written assembly for DG Novas, fixed up old Fortran code, reverse-engineered stuff whose source has been long-lost, and can help you out if your system is written in REBOL or Eiffel. It's fun. :)
 
2011-12-09 04:39:21 PM
WhyteRaven74: StoneColdAtheist: using APL

*takes a step back off your lawn*


LOL...yeah, that'sI'm pretty farking archaic.
 
2011-12-09 04:46:17 PM
In 1987, this culminated in what remained the fastest known algorithm for 24 years. It was developed by Don Coppersmith and Shmuel Winograd and put omega at 2.376.

Now, Virginia Vassilevska-Williams, who splits her time between the University of California, Berkeley, and Stanford University, has shown this assumption was wrong - by tweaking the algorithm to produce an omega of 2.373.


Her parents must be so proud.
 
2011-12-09 04:47:14 PM
Lord Dimwit: I've made a decent side-career of knowing obscure and/or old technologies. I'm 31, but I've written assembly for DG Novas, fixed up old Fortran code, reverse-engineered stuff whose source has been long-lost, and can help you out if your system is written in REBOL or Eiffel. It's fun. :)

Cool...I wrote a shiat-load of Fortran code back in the day. Elegant language as I recall.

/you're welcome on my lawn any time... :)

media.giantbomb.com
 
2011-12-09 04:49:20 PM
"Matrix Multiplication", a key theoretical mathematical tool for solving problems in physics, economics and science 3d video games...

FIFY
 
2011-12-09 04:55:15 PM
Bacontastesgood: Don Coppersmith and Shmuel Winograd ... put omega at 2.376.

Now, Virginia Vassilevska-Williams, who ... tweakinged the algorithm to produce an omega of 2.373.

Her parents must be so proud.


In a matrix of sufficient size to be useful that might be the difference to computing an answer in less than the projected lifespan of the observable universe. ;)
 
2011-12-09 05:05:18 PM
Interesting, but really not of great interest. Fast algorithms have been around since Strassen and they aren't much use in practice because they have serious stability/conditioning problems when you try to implement them with finite precision computers.
 
2011-12-09 05:13:45 PM
StoneColdAtheist: Lord Dimwit: I've made a decent side-career of knowing obscure and/or old technologies. I'm 31, but I've written assembly for DG Novas, fixed up old Fortran code, reverse-engineered stuff whose source has been long-lost, and can help you out if your system is written in REBOL or Eiffel. It's fun. :)

Cool...I wrote a shiat-load of Fortran code back in the day. Elegant language as I recall.

/you're welcome on my lawn any time... :)

[media.giantbomb.com image 400x282]


I think you may be suffering from that disease that causes you to think things were better when you were younger.
 
2011-12-09 05:18:22 PM
StoneColdAtheist: Bacontastesgood: Don Coppersmith and Shmuel Winograd ... put omega at 2.376.

Now, Virginia Vassilevska-Williams, who ... tweakinged the algorithm to produce an omega of 2.373.

Her parents must be so proud.

In a matrix of sufficient size to be useful that might be the difference to computing an answer in less than the projected lifespan of the observable universe. ;)


I worked on a very mission critical piece of embedded software back when I worked at *mumble*. One of the algorithms ended up having a superlinear time complexity (not like, Ackerman bad, but, bad enough). Basically, it came down to if you had more than about 30 *mumble* specified in the *mumble*, the system would freeze, as the calculation would end up completing some time after the heat death of the Universe (I will point out that I did not write the offending code). This code was burned into ROM and already shipped, so it wasn't going to change. I had to go through and write patches to the compiler to prevent code from getting deployed on top of the hardware that would trigger the issue. It's more fun if you know where this software is deployed and what would happen if it went wrong. :)
 
2011-12-09 05:33:20 PM
Lord Dimwit: StoneColdAtheist: Elegant language as I recall.

I think you may be suffering from that disease that causes you to think things were better when you were younger.


I should have added by the standard of the day. ;)

Lord Dimwit: StoneColdAtheist: ...computing an answer in less than the projected lifespan of the observable universe. ;)

...the calculation would end up completing some time after the heat death of the Universe


It seems we worked under similar constraints. :)
 
2011-12-09 05:47:23 PM
Whoa...Lord Dimwit has committed a random act of kindness in sponsoring me for a month of TotalFark. However shall I earn this singular honor?

Thank you, kind sir!

/Mrs SCA harrumphs in mild derision, but returns to her knitting...
 
2011-12-09 05:49:18 PM
Lord Dimwit: what would happen if it went wrong.

I reckon that the consequences would involve some serious *mumble*, amirite?
 
2011-12-09 05:50:31 PM
Hollie Maea: Lord Dimwit: what would happen if it went wrong.

I reckon that the consequences would involve some serious *mumble*, amirite?


They might even graduate to full on throat-clearing and meaningful glances.
 
2011-12-09 06:26:14 PM
StoneColdAtheist: In a matrix of sufficient size to be useful that might be the difference to computing an answer in less than the projected lifespan of the observable universe. ;)

Not really. Even for matrices of a trillion tera-elements (10^12 x 10^12), you only get 15% time savings. If you have googol element matrices, it comes out to half the time. If you need a matrix that big, you farked up your model or data reduction HARD.

Also: Although today's computers can't take advantage of this specific speed advance

www.enjoytherandom.com
 
Displayed 50 of 61 comments

First | « | 1 | 2 | » | Last | Show all


This thread is closed to new comments.

Continue Farking
Submit a Link »