Paper: https://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf
Translation: Implementing learning sequence to the sequence of neural network
Summary
Depth neural network model in many difficult learning tasks have achieved superior performance. Although in the case of a large number of annotation data, the DNN can work well, but they are not the sequence is mapped into a sequence . In this paper, we propose a generic method for end to end, to sequence learning, and do minimal assumptions on the sequence structure. Our method of using a multilayer LSTM (Long Short-Term Memory) , an input sequence is mapped into a fixed dimension vector, and then another depth LSTM decode the target sequence from the vector .