Apply signum-activated WASD neuronet to learning XOR logic via noisy input and output data

In this paper, a novel signum-activated weights-and-structure-determination neuronet (SAWASDN) is proposed, investigated and tested. Being different from the past WASD neuronet, the proposed SAWASDN employs discontinuous functions as its activation functions. In addition, we can determine the optimal weights directly and the optimal neuronet structure automatically by the WASD method. Finally, numerical experiments of learning and testing XOR logic via noisy input and output data are conducted, with Gaussian noise and with uniform noise added. Numerical results substantiate the feasibility, efficacy and robustness of the SAWASDN.