A sliding mode strategy for adaptive learning in multilayer feedforward neural networks with a scalar output

The features of a novel robust adaptive learning algorithm in analog multilayer feed forward networks are presented. It implements sliding mode control strategy. The zero level set of the learning error is considered as a sliding surface in the space of neural network learning parameters. A sliding mode trajectory can be brought on and reached in finite time on such a sliding manifold. The algorithm is applied to on-line learning of a non-monotonic function and manipulator forward dynamics identification. The learning neural structures come into some of the advantages of variable structure systems, such as high speed of learning and robustness.