A new LSTM model by introducing biological cell state

Abstract

Long Short Term Memory (LSTM) has been a very successful augmented recurrent neural network model employed to learn sequential information with long term dependencies where LSTM can store and compute information for a long period of time. In this study, a biologically inspired variation has been incorporated in LSTM by introducing additive cell state into the functionally computational system. The novel biological variant of LSTM model has been employed to conduct sentiment analysis of textual data. As the learning dataset, fifty thousand movie reviews have been used from IMDB where equal number of review data has been used for training and testing purposes. The comparative performance of the new variant is found to be promisingly better and show more stability than that of the traditional LSTM.