-
## Abstract
- Train source-to-target NMT (student) without parallel corpora available, guided by the existing pivot-to-target NMT (teacher) on a source-pivot parallel corpus
- X : source, Y : target…
-
## 論文リンク
https://arxiv.org/pdf/1906.02443.pdf
## 概要
NMTにおいて損失を利用したwhite-boxでの敵対的サンプル作成法を提案
作成した敵対的サンプルを用いた学習によるロバストなNMTモデルを提案
-
http://aclweb.org/anthology/P/P17/P17-1065.pdf
-
PaddleNLP/neural_machine_translation/transformer/infer.py报错，请问是什么原因，谢谢？
```
memory_optimize is deprecated. Use CompiledProgram and Executor
W0614 19:29:48.624078 50964 device_context.cc:261] Please…
-
#### Link: [ACL Anthology](https://aclanthology.org/2024.cl-1.2/)
#### Main problem
NMT models struggle when training on low-resource (LR) data.
#### Proposed method
Author proposed two pipelines …
-
#### Link: [ACL Anthology](https://aclanthology.org/2023.acl-long.332/)
#### Main problem
Existing test sets such as IATE, Wiktionary, and TICO employ oversimplified constraint settings and thus lea…
-
## 简介
论文分析了机器翻译在**推理**时的校准`Calibration`问题：即模型输出分布与结果不吻合，存在差距。 文章使用的评价指标是`ECE(期望校准误差)`，发现训练的`ECE`远远小于推理`ECE`,说明缩小训练和推理之间的差距需要做很多工作。
论文分析了一些`NMT`在语言学方面的现象：
`Frequency, Position, Fertility, Syntact…
-
https://arxiv.org/abs/1409.0473
https://www.youtube.com/watch?v=upskBSbA9cA&index=56&list=PLlMkM4tgfjnJhhd4wn5aj8fVTYJwIpWkS 참고
* 사실 이해가 잘 가지 않아 밑에 보면 여러군데 찾아봄.
-
## 简介
把Capsule Network套到了建模past-feture上。感觉写的比较套路了，了解了一下dynamic route咋工作的，再看论文感觉基本就是用的标准的做法套了一下呢。。写的有点不是很清楚。用pre-trained baseline继续训的。
## 论文信息
* Author: NJU
* [Paper](https://arxiv.org/pdf/1904.0…
-
# [Paper Review] Bahdanau Attention (2014) | woodong's log
Neural machine translation by jointly learning to align and translate (2014) 논문 리뷰 Bahdanau, D., Cho, K., & Bengio, Y. (2014). Neural machin…