new

87f3e405 · zengxin · 98151118 · 98151118 · 87f3e405 · 87f3e405
Commit 87f3e405 authored Dec 25, 2020 by zengxin
--- a/Continuous Output for NMT/ReWE RegressingWord Embeddings for Regularization of Neural Machine Translation Systems.NAACL2019.pdf
+++ b/Continuous Output for NMT/ReWE RegressingWord Embeddings for Regularization of Neural Machine Translation Systems.NAACL2019.pdf
--- a/Continuous Output for NMT/continuous-output-for-nmt.md
+++ b/Continuous Output for NMT/continuous-output-for-nmt.md
@@ -111,7 +111,7 @@

    其中$s_j$是解码器端隐藏向量，$W_1、W_2、b_1、b_2$是模型参数

-  * $ReWE_{loss}=l(e_j,e(y_j))$
+  * $ReWE_{loss}=\sum_{j=1}^m (1-cos(e_j),e(y_j))$

                   其中$e_j$是模型输出，$e(y_j)$是真实目标语言词嵌入向量，


--- a/unclassified/2019-acl-Best-Bridging the Gap between Training and Inference for Neural Machine Translation.pdf
+++ b/unclassified/2019-acl-Best-Bridging the Gap between Training and Inference for Neural Machine Translation.pdf