Neural Network Based Inverse System Identification from Small Data Sets

Many applications in control, signal processing and manufacturing require the inverse model of a system. Identification of the inverse of a system (inverse modelling) is often an ill-posed problem and therefore a challenging task. The learning capability of the artificial neural network (ANN) has been exploited in the past to identify the inverse of a system. However, in certain applications, such as manufacturing, where the available data samples are less, the complexity of fitting an inverse model increases significantly. This results in small data learning problem. The present study solves this small data learning problem from the perspective of the inverse system identification using neural networks. Initially, the effectiveness of different combinations of various virtual sample generation (VSG) methods and machine learning tools are investigated to determine the optimum combination which gives the highest learning accuracy. Simulation results are included to demonstrate the effectiveness of ANN in identifying the inverse of various systems from small data.