【摘要】 提出一种基于公平神经网络的学习算法.设置一个合理的信源数初始值,通过构造的一个稳定性判决器,能够自适应调整神经网络的维数,并估计出信源数真实值,从而使信源得以成功分离.理论分析表明,在其数学统计意义上缩减了训练时间;而计算机仿真结果表明,在其不同信源数条件下均能快速收敛.更多还原

【Abstract】 This paper proposes a fair neural-network-based algorithm.It initiates the estimatied source number to be a proper value,and constructs a stability discriminator,which can adjust dimensions of the nerual network and estimate the actual source number.Hence the algortihm is capable of separating sources sucessfully.Theoretical analysis indicates that it reduces the training time in mathematical statistical sense,and simulation results proves that it can converge quickly under different source number cases.更多还原