Dry Goods--lstm detailed, about Lstm's previous incarnation

Recently in doing lip reading field of research, design to C3D and RNN combination, so carefully observe the next LSTM series of papers, summed up as follows:

The ppt total length is 98 pages, the content includes:
1.conventional lstm (Detailed explanation of BPTT algorithm)
The proposed 2.forget gate
3.Peephole mechanism
4.encoder-decoder
5.GRU
6. Gated feedback for processing long-term and short-term information RNN
7. The attention mechanism, which is now widely used, includes the original version and the modified version
And the corresponding references, after reading, you will certainly benefit ^_^.

2017-12-18 update: Added understanding of "attention is" all your need (2017, NIPS)

Link: https://pan.baidu.com/s/1bXH2lK Password: Kpik
Only for learning and exchange use, I hope to help you.