g_harrison@vger.nsu.edu (01/07/91)
Neural-Netters: Most of my research work recently has been in program language theory, but I have been working a lot in neural net learning algorithms. I have been wondering if there is a "language of choice" for those of you who do some experimental twiddling with neural net algorithms! Most of the the stuff I've seen has been in C or Turbo Pascal, but I've actually found that Ada gives me some interesting possibilities: (a) it allows me to do a backpropogation network with various momentum and learning constants, sizes of input, middle, and output nodes, and some other variants without ever recompiling (use of generics). (b) it also opens some extremely interesting simulations by assuming random imputs (with noise, errors, etc) from individual tasks representing input nodes. I have had some interesting results but have been told that Ada and Neural Nets are "mutually exclusive," "flawed in concept," etc. - by some "experts" in the field. If you wish to flame me for using Ada, please do so - I have a rather thick skin and am willing to learn. Must I use LISP and/or Prolog to satisfy the AI types or C to satisfy those who are of that faith? George..... -- George C. Harrison -------------- || -- Overworked, Underpaid, -------- ---|| Professor of Computer Science || -- Unappreciated, but enjoying --- ---|| Norfolk State University, ---- || -- it all. ----------------------- ---|| 2401 Corprew Avenue, Norfolk, Virginia 23504 ----------------------- ---|| INTERNET: g_harrison@vger.nsu.edu --------------------------------- -- || These are not necessarily the views of my employer, my family, or -- || even myself.
greenba@gambia.crd.ge.com (ben a green) (01/09/91)
In article <436.27876349@vger.nsu.edu> g_harrison@vger.nsu.edu writes:
I have had some interesting results but have been told that Ada and Neural Nets
are "mutually exclusive," "flawed in concept," etc. - by some "experts" in the
field.
If you wish to flame me for using Ada, please do so - I have a rather thick
skin and am willing to learn. Must I use LISP and/or Prolog to satisfy the AI
types or C to satisfy those who are of that faith?
I found Ada useful for constructing nets of different architectures by
use of generic layers with parameters.
In general, I think of Ada as follows:
Ada is to C what C is to Fortran.
Unfortunately, there are still many people stuck in Fortran, and it will be
a long, long time before the C folks learn what they're missing. I have
to work in C nowadays for lack of an accessible Ada compiler, but the
time I have wasted chasing C bugs that Ada would not have let me make
in the first place is a strong argument for buying the compiler.
Ben
--
Ben A. Green, Jr.
greenba@crd.ge.com
Speaking only for myself, of course.ins_atge@jhunix.HCF.JHU.EDU (Thomas G Edwards) (01/09/91)
In article <GREENBA.91Jan8111008@gambia.crd.ge.com> greenba@gambia.crd.ge.com (ben a green) writes: >In article <436.27876349@vger.nsu.edu> g_harrison@vger.nsu.edu writes: > I have had some interesting results but have been told that Ada and Neural Nets > are "mutually exclusive," "flawed in concept," etc. - by some "experts" in the > field. >Unfortunately, there are still many people stuck in Fortran, and it will be >a long, long time before the C folks learn what they're missing. Actually, I have recently been pseudo-forced to learn APL for a certain job involving neural networks. It actually is a really neat hacky language for mathematics. Matrix algebra can easily be done in just a few operators on a single line, and its workspace environment (and interpretive nature) make throwing together a network architecture easy. Sure, it's slow, uses wierd characters which you need to stick on your keyboard, and is not the most easily readable language. But it is a good method of prototyping neural architectures (before you actually code them in C or what-have-you). -Tom