The LSTM manages a cell state vector \({\bf c}^t\) expressing the story of the sequence so far and informing the future decisions (top horizontal pipeline in Fig. \ref{499635}). The LSTM can act on this vector to indicate regime transitions and forget past information (or simply stay the course) and it embodies the Long-Short Term Memory functionality indicated in the name. The input of the decision process is made up of both the current sequence input data \({\bf x}^t\) and the recurring latent state vector \({\bf h}^{t-1}\) (bottom horizontal pipeline in Fig. \ref{499635}). The output of the decision process is a new latent state vector \({\bf h}^t\), from which the actual output can be extracted.