arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2104.00664cs.CLcs.LG
收藏

神经机器翻译蒸馏数据的采样与滤波

Sampling and Filtering of Neural Machine Translation Distillation Data

Institute of Formal and Applied Linguistics, Charles University(查尔斯大学形式与应用语言学研究所)

Vilém Zouhar

摘要:

在大多数神经机器翻译提取或窃取场景中,目标是保留目标模型(教师)的性能。教师模型的最高分假设通常用于训练新模型(学生)。如果参考翻译也可用,则可以对更好的假设(相对于参考)进行上采样,并且去除或欠采样差的假设。本文探讨了重要性采样方法景观(修剪,假设上采样和欠采样,重复数据删除及其组合)与英语到捷克语和英语到德国的MT模型使用标准的MT评估指标。我们表明,与仅在原始数据或合成数据或其直接组合上进行训练相比,仔细的上采样和与原始数据的组合会带来更好的性能。

英文摘要:

In most of neural machine translation distillation or stealing scenarios, the goal is to preserve the performance of the target model (teacher). The highest-scoring hypothesis of the teacher model is commonly used to train a new model (student). If reference translations are also available, then better hypotheses (with respect to the references) can be upsampled and poor hypotheses either removed or undersampled. This paper explores the importance sampling method landscape (pruning, hypothesis upsampling and undersampling, deduplication and their combination) with English to Czech and English to German MT models using standard MT evaluation metrics. We show that careful upsampling and combination with the original data leads to better performance when compared to training only on the original or synthesized data or their direct combination.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑