| Re: Neural Networks [message #61700 is a reply to message #18986] |
Sun, 27 July 2008 05:13   |
![Chris[6] is currently offline](theme/default/images/xoffline.png.pagespeed.ic.XRkd1fkXye.png "Chris[6] is currently offline")
Chris[6]
Messages: 84
Registered: July 2008
|
Member |
|
|
On Jul 26, 11:13 pm, "|Rob|" <r08...@gmail.com> wrote:
> Hi all,
>
> I posted a query on here about a year ago relating to neural networks
> in IDL but I didn't get any response so I thought it might be worth
> trying again.
>
> The problem I'm currently working on for my PhD breaks down to
> matching a known pattern/signature against a huge amount of data to
> find cases where the pattern is present.
>
> I don't know that much about neural networks (just a few undergraduate
> lectures on the basic principles) or exactly how I'd go about
> implementing something like this but it seems that IDL is fairly well
> suited to this kind of task due to it's ability to sort and match
> large arrays of data.
>
> Have many people worked on neural networks in IDL before? I haven't
> had much luck searching the group for answers apart from a few old
> (and very old!) posts on the subject asking pretty much the same as I
> am.
>
> I'd much prefer to use IDL for this if possible but I'm starting to
> wonder whether something like Matlab might be better suited as there
> seems to be quite a bit more documentation and example code for using
> neural networks in IDL.
>
> Cheers
>
> Rob
I haven't used IDL for NN programming, though I've been interested in
the idea.
My suspicion is that, while you could probably implement a pre-learned
neural network in IDL, it would not be efficient at actually learning
the proper weights. Typical algorithms for learning NN weights (eg
backpropagation) are dominated by looping. IDL usually chokes on such
tasks unless there is a way to eliminate some of the loops by using
its optimized array operations. I may be wrong, but I don't think a
nonlinear, iterative process like backpropagation could be vectorized
very much.
For general info on implementing NNs, I might look at Tom Mitchell's
"Machine Learning," and maybe some open source implementations (like
JOONE for Java)
chris
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
