Pierwsze dwie strony wygladaja jak wprowadzenie dla studentow - to jest konferencja ekspertow!
Recenzent moze sie zdenerwowac ... 

and wages vector => and weight vector

straight forward:

network tranforms the data into space of a higher dimension => network tranforms the data into space of a lower dimension  !

backprop algorithm,

Linearly separable data and the separating line ? 

One of them was to introduce a  ???

A set is an .... 

częste "we" - bezosobowo!

There is no measure of distance that could be minimized.  ??? Distance from the hyperplane?

but a fuzzy value, => but a graded response,

The similarity to distribution functions => The similarity of these requirements to the properties of distribution functions

increasing the dimensionality of the problem eventually makes it separable. => projecting the problem into a space of increasing dimensionality, or increasing the dimensionality of the feature space by adding combinations of inputs,  eventually makes it separable \cite{Cover or Haykin}.  

numeric age. ?  Cała część tłumacząca rozmytą koncepcje "young" jest tu mało potrzebna.

products of matrices => products of weight matrices

...propagation, not only with sigmoid functions, though in general this is not valid.  ???

sudden chaos. ??

216 vector dataset,  ??? brak opisu

color - nie drukuja w kolorze!

synchronize the function ??

Fig. 5-8 - po ilu iteracjach? jakie wyniki koncowe?

unable to provide a satisfactory separation, => dziwne - może błąd w znaku? Albo aktywacje wskakuja na część stałą tej funkcji? Jeśli działa semilinowa czemu tu są takie trudności? 

Fig. 10 - convex hulls?

Testing on the usual dataset shows great separation capabilities.  =>  Testing the network with these transfer functions on the usual  dataset shows great separation capabilities.

The smoothness of the function causes the fuzzy hyperplanes to move in their place with ease. 
The plot simply reads itself as the geometry of the transformation is clearly visible.  - niezrozumiale. 

Potrzebne jest jakies porówanie uśrednionej liczby iteracji do zbiezności + końcowego błedu.