From a new way to show the universality of a two-layer perceptron to a new approach to digital design

摘要: We show in a constructive way the universality of two layer perceptron. how to build any minterm logical function with one perceptron and OR all minterms another This it is possible implement N N+1 perceptrons. Next we addres question minimizing number perceptrons or threshold elements given function. Here have answer open questions that not yet answered literature:1) Is linearly separable? 2) If nonseparable minimize function? know separable can be implemented by trained Rosenblatt's Perceptron Learning Rule [1]-[2] guarantees convergence case existence solution. propose as test non-separable did converge after very great epochs iterations. As method minimization trial error procedure generates manners mnimum elements, if for them learning rule associations then must increment generate again implementation till got Finally present some examples functions minimum elements.