seq2seq attention相关

seq2seq模型实例：用Keras实现机器翻译
玩转Keras之seq2seq自动生成标题

https://baijiahao.baidu.com/s?id=1627587324043258333&wfr=spider&for=pc
https://www.jianshu.com/p/923c8b489604
https://www.cnblogs.com/DLlearning/p/7834018.html
https://mp.weixin.qq.com/s/QwVImqc66GP3_KBwSaU2CQ

https://mp.weixin.qq.com/s?__biz=MzU1NTU3Njg4Mw==&mid=2247484354&idx=1&sn=2028329ca69a800628b6129c3936dd10&chksm=fbd3789ecca4f1881dee66c054bc57c68138bd429a668876d92c6e8a56e98fd3d6a06674f5de&scene=21#wechat_redirect

https://yq.aliyun.com/articles/669616
https://zhuanlan.zhihu.com/p/40920384
https://www.zhihu.com/people/cheshengyuan/posts
https://zhuanlan.zhihu.com/p/36361833
https://zhuanlan.zhihu.com/p/39034683
https://spaces.ac.cn/archives/5861/comment-page-2#comments

https://github.com/bojone/seq2seq
https://cloud.tencent.com/developer/news/46171

https://github.com/kmsravindra/ML-AI-experiments/blob/master/AI/Neural Machine Translation/Neural machine translation - Encoder-Decoder seq2seq model.ipynb

https://github.com/chenjiayu0808/ML-AI-experiments/tree/master/AI/Neural Machine Translation

https://github.com/keras-team/keras/blob/master/examples/lstm_seq2seq.py

---

Attention目前基本上已经是seq2seq模型的“标配”模块了，它的思想就是：每一步解码时，不仅仅要结合encoder编码出来的固定大小的向量（通读全文），还要往回查阅原来的每一个字词（精读局部），两者配合来决定当前步的输出。

干货 | Attention注意力机制超全综述
阿里云社区-Attention模型
完全图解RNN、RNN变体、Seq2Seq、Attention机制
苏剑林-《Attention is All You Need》浅读（简介+GitHub keras 代码）
真正的完全图解Seq2Seq Attention模型