Refereed full papers (journals, book chapters, international conferences)

1991

Though it arouses more and more curiosity, the HJ iterative algorithm has never been derived in mathematical terms to date. We attempt in this paper to describe it from a statistical point of view. For instance the updating term of the synaptic efficacies matrix cannot be the gradient of a single C2 functional contrary to what is sometimes understood. In fact, we show that the HJ algorithm is actually searching common zeros of n functionals by pipelined stochastic iterations. Based on simulation results, advantages and limitations as well as possible improvements are pointed out after a short theoretical analysis.