|
PureBytes Links
Trading Reference Links
|
pierre.orphelin:
>> Unlike what you think, fuzzylogic, whent properly trained, is able to generalize ...And this is more true with neurofuzzy logic...... This froggyfied piece of software is at least 4-5 years ahead the current US technology.<<
I can attest that, from my own R&D, fuzzy neurons can be incredibly powerful. My investigation led me to discover a large class of functions ( eg. simple, geometric and harmonic averaging, minima and maxima, L1 and L2 distance metrics, etc. ) as mere points in a fuzzy-neural function space. All other points in the space represent undiscovered functions that can be easily accessed by setting a few parameters in the fuzzy neuron. Imagine, if you will, a function Fn(x) where subscript n defines the function: n=-1 yields harmonic averaging, n=0 yields geometric averaging, n=1 yields arithmetic averaging, and so on. Parameter n can actually be any floating point value from -infinity to + infinity, thereby unifying an broad class of functions parametrically. Training might be set n=0.37, producing a function that is nearly impossible to intuit. Now imagine 7 adjustable parameters, producing new functions in 7-D function space. This goes well beyond the capabilities of percep!
tron based and template based NN models.
Pierre's neuro-fuzzy software may very well be years ahead of mainstream USA.
Mark Jurik
Jurik Research
http://www.jurikres.com
|