A Fine-Tuning Method Using Pruning of Recurrent Neural Network for Prediction of the Anesthetic Effects

This paper proposes a fine-tuning method combined with the pruning of a recurrent neural network (RNN). The purpose of this study is to realize a prediction model used for predictive control of anesthesia by RNN. RNNs also useful for modeling non-linear and time-varying systems such as dose-response to the human body. However, to improve the accuracy of prediction by the RNN, it is necessary to consider the number of neurons in the RNN. If the number of neurons is too small, the accuracy of the model is reduced. While, too many neurons can lead to overfitting and increase computational complexity. Considering that problem, we propose a fine-tuning method using the pruning of redundant neurons.