A note on the simulation of exponential threshold weights

Abstract

It is known (see [GHR],[GK]) that for F(x) = w0 + w1x1 + …+ wnxn, a threshold gate G(x) = sgn(F(x)) which may have arbitrarily large integer weights wican be computed (“simulated”) in threshold circuits of depth 2 with polynomial size. In this paper, we modify the method from [GK] to obtain an improvement in two respects: The approach described here is simpler and the size of the simulating circuit is smaller.