The question has been asked to me about using a GRU (gated recurrent unit) instead of an LSTM. From what I can tell from my googling is that GRU's are becoming more popular and are rivaling LSTM's for performance. They have a simpler structure with 1 fewer gate than LSTM's, so they should train faster. I think sometime, it'd be worth a try (maybe on the single segment model to begin with).
The question has been asked to me about using a GRU (gated recurrent unit) instead of an LSTM. From what I can tell from my googling is that GRU's are becoming more popular and are rivaling LSTM's for performance. They have a simpler structure with 1 fewer gate than LSTM's, so they should train faster. I think sometime, it'd be worth a try (maybe on the single segment model to begin with).