April 2018
Intermediate to advanced
334 pages
10h 18m
English
This section explains the neural intra-attention model on the encoder-decoder network. Here,
represents the sequence of input (article) tokens, and
represents the sequence of output (summary) tokens. The encoder part of the network consists of bi-directional LSTM (see Appendix A, Further topics in Reinforcement Learning). Thus, the input sequence x is read using a bi-directional LSTM which computes the hidden states from the embedding vectors of , where || represents concatenation of the vectors.
In the decoder ...