In this paper we prove a finite convergence of online BP algorithms for nonlinear feedforward neural networks when the training patterns are linearly separable.

• 当训练样本线性可分时；本文证明前馈神经网络的在线BP算法是有限次收敛的.