Sliding mode-based adaptive learning in dynamical Adalines

A sliding mode control strategy is proposed for the synthesis of adaptive learning algorithms in perceptron-based feedforward neural networks whose weights are constituted by first order, time-varying, dynamical systems with adjustable parameters. The approach is shown to exhibit strong robustness and fast convergence properties. A simulation example, dealing with an analog signal tracking task, is provided, which illustrates the feasibility of the approach.