My guess is that this is related to how LSTM networks operate: they learn to pay attention differently to various parts of data, focusing more on some depending

My guess is that this is related to how LSTM networks operate: they learn to pay attention differently to various parts of data, focusing more on some depending on the input. I believe the broader class is composed of various "attention mechanisms". But I'm mostly giving keywords away, I do not have an intricate understanding of the way they work either.