Editing Deep Learning Revolution (section)

=== Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) ===
RNNs are a category of neural networks designed for sequential data.<ref>Arisoy, E., Sethy, A., Ramabhadran, B., & Chen, S. (2015, April). Bidirectional recurrent neural network language models for automatic speech recognition. In ''2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)'' (pp. 5421-5425). IEEE.</ref><ref>Graves, A. (2013). Generating sequences with recurrent neural networks. ''arXiv preprint arXiv:1308.0850''.</ref> RNNs are particularly useful in ASR due to their ability model longer-distance context than word n-gram models. LSTMs, a specific type of RNN, aim to deal with the vanishing gradient issues present in traditional RNNs.<ref>Graves, A., & Graves, A. (2012). Long short-term memory. ''Supervised sequence labelling with recurrent neural networks'', 37-45.</ref> LSTMs particularly excel in the handling and modeling of sequential data, and speech signals are highly sequential. They can model the probabilistic relationships between words in a sentence, which aids in improving the recognition accuracy.