Abstract:

A mathematical model of functional networks is proposed. The property of its topology structure and learning process is analyzed. Online incremental learning algorithms based on the block matrix and the property of functions matrix are designed. The learning algorithms make fully use of the training history, and have functions of learning, modification and emergency adaptation. Simulation on a Hénon time series shows effectiveness of the proposed algorithms.

Key words: basis functions, functional parameter, Lagrange multiplier method, online incremental learning, Hénon time series