Neural network learning using time-varying two-phase optimization

A time-varying two-phase (TVTP) optimization algorithm is proposed based on the two-phase neural network (NN) and the time-varying programming NN. The proposed algorithm is most useful when the problem is a time-varying optimization programming which may have some constraints. Thus it can be applied to the training of the NN where it has some constraints on weights. Computer simulations show that the proposed TVTP algorithm has good adaptability in online learning and is less sensitive to the learning step size than the conventional error back propagation (EBP) method.<>