Re: comp.ai.neural-nets FAQ: Neurofuzzy

Aki Vehtari (Aki.Vehtari@hut.fi)
Tue, 18 Jun 1996 17:06:09 +0200


"WS" == Warren Sarle <saswss@hotellng.unx.sas.com> writes:
WS> I am updating the entry on fuzzy logic in the comp.ai.neural-nets FAQ
WS> in light of some recent arguments. Any comments?
<...>
WS> discussion) about combining fuzzy logic with neural network approaches
WS> in comp.ai.neural-nets.
<...>

There is a researh project "Using external knowledge in neural network
training and structure" at Lappeenranta University of Technology
(Finland) <URL:http://www.lut.fi/menes/extknow.html>

They have presented algorithms, where constraints can be stated with
fuzzy-like terms for different order derivatives of the mapping,
allowing a considerable amount of background knowledge to be trained
to the MLP network. Fuzzy-rules have now been incorporated into a
commercial neural network process control tool.

Algorithms are presented in the following two articles:

Jouko Lampinen and Arto Selonen, "Multilayer Perceptron Training
with Inaccurate Derivative Information", Proc. 1995 IEEE
International Conference on Neural Networks ICNN'95, Perth, WA, Vol.
5, pp. 2811-2815, 1995
(ftp://lut.fi/reports/neural-nets/Lampinen_ICNN-95.ps.gz)

Jouko Lampinen and Arto Selonen, "Using Background Knowledge for
Regularization of Multilayer Perceptron Learning", Submitted to
International Conference on Artificial Neural Networks, ICANN'96,
Bochum, Germany, 16-19 July, 1996

There is another researh project At Tampere University of Technology
(Finland) "Neuro-Fuzzy Systems"
<URL:http://dmiwww.cs.tut.fi/nfs/Welcome_uk.html>

List of Publications at <URL:http://www.cs.tut.fi/~tpo/group.html>
Most of them are about fuzzy self-organizing maps and neuro-fuzzy
controllers.

Aki Vehtari

-- 
Quid lucrum istic mihi est?