Open tm4roon opened 5 years ago
ニューラル機械翻訳(NMT)におけるdata augmentationの手法を提案。言語モデルの出力分布をNMTの入力として利用する。NMTへの入力は基本的にone-hot表現の入力であるが、ここでは、確率的に言語モデルの出力分布を入力する。データが少ない場合、多い場合どちらにおいても、提案手法により性能を改善することが可能。
Soft Contextual Data Augmentation for Neural Machine Translation
ニューラル機械翻訳(NMT)におけるdata augmentationの手法を提案。言語モデルの出力分布をNMTの入力として利用する。NMTへの入力は基本的にone-hot表現の入力であるが、ここでは、確率的に言語モデルの出力分布を入力する。データが少ない場合、多い場合どちらにおいても、提案手法により性能を改善することが可能。
文献情報