Deep Learning: Grid LSTM

GRID Long Short Term Memory.

Recurrent Neural Network has been developed to handle issues of time series, time dependant event predictions or generations.
General problem in order to minimize the cross entropy of the loss function, we need to propagate the variations through the network cells. With a large of recurrent step, signal becomes smaller and noise overcome the signal level.

So, Long ShortTerm memory architecture have been developed to handle recurrent process using memory cells and reducing the effect of noise overcoming the variational signal.

Training of Grid LSTM Neural Network