Long Short-Term Memory Units (LSTMs) are a subset of deep learning used in artificial intelligence, they were proposed by German researchers Sepp-Hochreiter and Juergen Schmidhuber, they are a solution used in recurrent networks to create a more constant backpropagation through time and layers.
Simple recurring networks would use the current input and what they have perceived previously in time (previous moment).
LSTM creates a mechanims to store information over cycles (not just the previous), they is an important building block to create networks that can keep learning from new input.
