Augmented Seq2seq

• Vanilla Seq2seq
• Conditioning decoder on external context
  • Dynamic Decoder (teacher forcing removed)
• Language Modeling with encoder
• Bidirectional encoder
• RNNSearch : Soft Alignment
• Multi-turn Conversation Modeling with Hierarchical Recurrent Encoder-Decoder (HRED)
• Memory augmentation

Reference

1. Ed Grefenstette, Beyond Sequence to Sequence with Augmented RNNs video, slides
2. Neural Machine Translation by Jointly Learning to Align and Translate
3. End-to-End Memory Networks
4. Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models
5. Advanced Seq2seq in tensorflow
6. Beam Search
7. Multi-task Learning in tensorflow
8. Enhanced Attention-based Encoder-Decoder model for NMT

Credits

• thank you @ematvey, for helping me understand raw_rnn