I've been thinking about it, but without too much ideas. I never
heard about SE applications done with Neural Networks also... But
I would like to discuss it AND will give some of the ideas I am
developing.
First you have to imagine what are the uses of Neural Networks.
Mainly I see two areas: classification and associative memory.
Others areas of interest would be more connected with AI.
In the classification area I believe one can build a system that would
learn characteristics of various different software. I did some work
on classification of programming languages (not using neural networks)
where the idea was to classify the languages accordingly to some characteristics, and them ask the user which characteristics he expects from a language, finding the most appropriate for him latter. This would be a kind
of System to Support Decision.
In the associative memory, I am working with the idea of passing different
(thousands?) program specification through a NN and them using the NN to
find the specification that is more similar. Some difficulties in the
representation mechanisms make this work very difficult for me.
Other area that I discovered just now is in the papers:
"The Connectionist Scientist Game:Rule Extraction and Refinement in a
Neural Network", by McMIllan, Mozer and Smolensky (to appear in
Proc. of the 13th Annual Conference of the Cognitive Science Society)
and
"rule Induction through Integrated Symbolic and SubSymbolic Processing"
(same authors) Advances in Neural INformation Processing Systems IV.
In this paper they use RuleNet to learn explicit symbolic condition action
rules, i.e., learning grammars from examples of the language!
SO, what are your ideias?
Jerry
--
Geraldo Xexeo | xexeo@dxcern.cern.ch | In Virtual Reality: MindWeb
CERN-PPE Div | xexeo@rio.cos.ufrj.br (*) | (*) after Dec. 22
1211 Geneva 23| bt268@cleveland.freenet.edu | I am going back to Brazil !!
Switzerland | gxexeo@cernvm.bitnet | "...Rio eu gosto de voce..."
