codertimo / paper-log

읽어야 하는 논문들을 관리하고, 읽은 논문들의 기록을 남기는 공간
31 stars 5 forks source link

Disentangled Representation Learning for Non-Parallel Text Style Transfer #15

Closed codertimo closed 4 years ago

codertimo commented 4 years ago

어떤 내용의 논문인가요? 👋

Abstract (요약) 🕵🏻‍♂️

This paper tackles the problem of disentangling the latent representations of style and content in language models. We propose a simple yet effective approach, which incorporates auxiliary multi-task and adversarial objectives, for style prediction and bag-of-words prediction, respectively. We show, both qualitatively and quantitatively, that the style and content are indeed disentangled in the latent space. This disentangled latent representation learning can be applied to style transfer on non-parallel corpora. We achieve high performance in terms of transfer accuracy, content preservation, and language fluency, in comparison to various previous approaches.

이 논문을 읽어서 무엇을 배울 수 있는지 알려주세요! 🤔

레퍼런스의 URL을 알려주세요! 🔗

https://www.aclweb.org/anthology/P19-1041/

codertimo commented 4 years ago

Motivation

Method

Flow

각 loss 설명

각 mini-batch 별로 학습 순서를 나누어서 학습합니다.

foreach mini-batch do
2 minimize Jdis(s)(θdis(s)) w.r.t. θdis(s);
3 minimize Jdis(c)(θdis(c)) w.r.t. θdis(c);
4 minimize Jovr w.r.t. θE, θD, θmul(s), θmul(c);
5 end

Experiment

Result

스크린샷 2020-01-11 오후 5 25 39