Need help with mathlab and parallel computing

phobos

Right I am in the middle of a parallel computing lab session, and once again I am quite confused. The start of the course went fine, and I understood the concepts behind parallel computing, but then we started using Mathlab, and it all went pear shaped after that.

What I don't understand is why we are using linear algebra, and mathlab. I don't understand how this can enable a process to use parallel processors. So far we have used two matrices elimination methods.

Gaussian Elimination (Direct method)
Jacobi (Iterative method)

I haven't got a clue what these things are doing. Also what the hell is the diag function for, and.........[I think I'm going to go mad]

Any help would be appreciated. If anyone else has done a parallel computing course, will you tell me where you got some decent tutorial information for people that have never touched the subject before.

I seem to understand things if I am taught it from a high level first, which gives me a general overview of what's going on. Then after that you can break it down, and as long as I know where everything fits in to the high level picture, I'm OK. But unfortunately we are not being taught that way.

Damn

;-phobos-)

phobos

Actually it's been a while since I first posted on this thread, and things have gotten a little bit better. TBH it doesn't take a lot of brains to figure out what parallel computing is about, but I am still quite confused with the above problem.

At least we have moved on to the more practical aspects of the topic, using the PVM (Parallel Virtual Machine). That, I find more interesting, and seem to grasp a lot easier.

But my question at this point is...
AFAIK the reason we were using linner algebra was just as an example of a computational process, who's common problem could be distributed amongst (n)processors. As far as I'm concerned the only reason we had more than one method for doing this is so we could see the various efficiency levels, and use our findings to determine which would be the best method to use. But why MATHLAB??

Why couldn't we have just written programs with a PVM from day one. Why did we have to do all that maths?. The fact that I can't answer these questions scares me, because the exam is just after Christmas. Fortunately for me, this module doesn't carry much towards my final year average, but I don't want to fail it either

Any light on the subject, please!!

;-phobos-)

marauder

I used matlab back in the day to do some neural network simulation which is simiar in some respects to parallel computing in that you use and array of computational elements that operate in parallel on an input to generate an output (listen to me ffs )
Anyway, the reason you would use matlab is that it is extremely good at matrices and a parallel system can be represented by a matrix


so lets say you have 2 input a, b and you want to peform some operation on them to end up with an output c

c = f(a,b)

you can use a matrix to represent the process

| w x |
| y z |

then

| a b|  | w x |   =  |wa+yb  xa+zb| 
        | y z |   

so then you can do some operation on the output vector like
modulus to get a single output.

c = mod(|wa+yb xa+zb|)


graphically this looks like this (edit: should look OK now)

         --->[w]-----------\
a--------|                 |
         --->[x]-----\     |-->[+]--------\
                  /--+-----/               \---------->[mod]-----> c
                  |  |                     /
         --->[y]--/  \--|----->[+]--------/
b--------|              |
         --->[z]--------/

this is a simple linear example. you can get freaky with w x y z and make them functions of some external variables that define the process. In the Neural Network case I used a nonlinear function

I belive you are doing something similar - using the matrix algebra to represent parallel processes. I agree though that in your case its a strange way of doing it....

I this has confused you more then apologies

DadaKopf

Does this mean you're studying Parallel Distributive Processing Groups? AI, cognitive science? neural nets, functionally isomorphic systems etc?

Is anyone here doing the UCD Cognitive Science course at all? I'm a philosophy student and Philosophy of Mind is the hot topic these days so I'd be interested in finding out how people find the course.

phobos

Does this mean you're studying Parallel Distributive Processing Groups? AI, cognitive science? neural nets, functionally isomorphic systems etc?

I'm just studying Parallel Distributive Processing Groups for now. I have done some AI with CBR and Vision based technologies in the past. We're dropping parallel after Christmas, and picking up AI again AFAIK.

Thanks for the help lads. Tis always appreciated

;-phobos-)

DadaKopf

Let me know how the rest of the AI course goes will you? Anything about programs which are able to evolve and write themselves like that Life-World program?