An entropy estimator with least square error [biological signals]

Presents a new entropy estimator which minimizes mean squared error between its estimate and true entropy. It uses a new entropy function approximated by a polynomial when an information source outputs two kinds of source symbols independently. In conventional entropy estimation, entropy has been estimated by replacing true occurrence probabilities of each source symbol in the entropy function with their estimates. In this case, the entropy estimator is not optimum in the meaning of least square error. This article demonstrates that the proposed entropy estimator achieves excellent mean squared error compared to a conventional entropy estimator.