|
• Rule Extraction from Neural Networks
The literature records some number of efforts directed at rule extraction from trained artificial neural networks, such as:
http://citeseer.ist.psu.edu/262926.html
http://citeseer.ist.psu.edu/261232.html
http://www.di.uniba.it/~castella/papers/NNW2000.pdf
http://www.cis.ysu.edu/~alazar/pdf/anniec.pdf
http://www.ci.louisville.edu/zurada/papers/SetionoLeowZuradaTNN02.pdf
http://osiris.sunderland.ac.uk/~cs0kmc/MyPapers/icann99.pdf
Obviously, one could simply train a neural network and fire it over a large set of (designed) query points to produce a training data set for a conventional tree- or rule-induction algorithm (CART, ID3, RIPPER, ITRULE, etc.). It would seem more straightforward to me to just run such an algorithm over the original data, though.
-Will Dwinnell
http://will.dwinnell.com
|