Dear friends,
I would like to design a network, that I will call net_R similar to this one, but with the following charactersitics:
Design a basic module with four neurons:
w1x1 + w2x2 + k1<= 0
w1x1 + w2x2 + k2 > =0
-(1/w1)x1 + w2x2 + k3 <= 0
-(1/w1)x1 + w2x2 + k4 >=0
that is, net_R(x1,x2) = 1 <=> (x1,x2) is insided the rectangle nearned from a sample. Not that this 4 perceptrons network depends just of four parameters: (w1,w2,k1,k2,k3,k4). This should imply in less data due to the sharp
modeling that imply in smaller VC-dimension.
ii-design the netowork Net-4 that integrates 4 net_R(x1,x2) by the OR connective, that is,
Net-4(x1,x2) = 1 <=> net_R(y)(x1,x2)=1 for y=1,2,3 or 4,
y being the indides of the rectangle networks.
Thank you very much
Junior Barrera (Univrsity of SP - Brazil)
