A self growing learning algorithm for determining the appropriate number of hidden units

The authors propose an algorithm to determine the appropriate number of hidden units in a multilayer feedforward neural network. This algorithm is based on heuristic terminal attractor backpropagation (HTABP), which can finish learning in finite time, reach the global minimum of the error function, and guarantee to converge faster than the backpropagation algorithm. The criteria for adding a hidden unit are the time-varying gain of HTABP and the normalized error function. Several simulation results show that the algorithm is effective and reliable. With this algorithm, the estimation of a number of hidden units by trial and error is no longer necessary