Repository logo
  • English
  • 中文
Log In
Have you forgotten your password?
  1. Home
  2. College of Electrical Engineering and Computer Science / 電機資訊學院
  3. Electrical Engineering / 電機工程學系
  4. MetaEx-GAN: Meta Exploration to Improve Natural Language Generation via Generative Adversarial Networks
 
  • Details

MetaEx-GAN: Meta Exploration to Improve Natural Language Generation via Generative Adversarial Networks

Journal
IEEE/ACM Transactions on Audio Speech and Language Processing
Journal Volume
31
Date Issued
2023-01-01
Author(s)
Chuang, Yun Yen
Hsu, Hung Min
Lin, Kevin
RAY-I CHANG
HUNG-YI LEE  
DOI
10.1109/TASLP.2023.3317571
URI
https://scholars.lib.ntu.edu.tw/handle/123456789/636986
URL
https://api.elsevier.com/content/abstract/scopus_id/85173337578
Abstract
Generative Adversarial Networks (GANs) have been popularly researched in natural language generation, so-called Language GANs. Existing works adopt reinforcement learning (RL) based methods such as policy gradients for training Language GANs. The previous research of Language GANs usually focuses on stabilizing policy gradients or applying robust architectures (such as the large-scale pre-trained GPT-2) to achieve better performance. However, the quality and diversity of sampling are not guaranteed simultaneously. In this article, we propose a novel meta-learning-based generative adversarial network, Meta Exploration GAN (MetaEx-GAN), for ensuring the quality and diversity of sampling (sampling efficiency). In the proposed MetaEx-GAN, we develop an explorer trained by Meta Exploration to sample from the generated data to achieve better sampling efficiency. MetaEx-GAN employs MetaEx first applied to Language GANs to achieve better performance. We also propose a critical training method for MetaEx-GAN on the NLG task. According to our experimental results, MetaEx-GAN achieves state-of-the-art performance compared with existing Language GANs methods. Our experiments also demonstrate the generality of MetaEx-GAN with different architectures (involving GPT-2) and how MetaEx-GAN operates to improve Language GANs.
Subjects
Electronic mail | Generative adversarial networks | Generative adversarial networks | Generators | large-scaled pre-trained model | meta reinforcement learning | Monte Carlo methods | natural language generation | Speech processing | Task analysis | Training
Type
journal article

臺大位居世界頂尖大學之列,為永久珍藏及向國際展現本校豐碩的研究成果及學術能量,圖書館整合機構典藏(NTUR)與學術庫(AH)不同功能平台,成為臺大學術典藏NTU scholars。期能整合研究能量、促進交流合作、保存學術產出、推廣研究成果。

To permanently archive and promote researcher profiles and scholarly works, Library integrates the services of “NTU Repository” with “Academic Hub” to form NTU Scholars.

總館學科館員 (Main Library)
醫學圖書館學科館員 (Medical Library)
社會科學院辜振甫紀念圖書館學科館員 (Social Sciences Library)

開放取用是從使用者角度提升資訊取用性的社會運動,應用在學術研究上是透過將研究著作公開供使用者自由取閱,以促進學術傳播及因應期刊訂購費用逐年攀升。同時可加速研究發展、提升研究影響力,NTU Scholars即為本校的開放取用典藏(OA Archive)平台。(點選深入了解OA)

  • 請確認所上傳的全文是原創的內容,若該文件包含部分內容的版權非匯入者所有,或由第三方贊助與合作完成,請確認該版權所有者及第三方同意提供此授權。
    Please represent that the submission is your original work, and that you have the right to grant the rights to upload.
  • 若欲上傳已出版的全文電子檔,可使用Open policy finder網站查詢,以確認出版單位之版權政策。
    Please use Open policy finder to find a summary of permissions that are normally given as part of each publisher's copyright transfer agreement.
  • 網站簡介 (Quickstart Guide)
  • 使用手冊 (Instruction Manual)
  • 線上預約服務 (Booking Service)
  • 方案一:臺灣大學計算機中心帳號登入
    (With C&INC Email Account)
  • 方案二:ORCID帳號登入 (With ORCID)
  • 方案一:定期更新ORCID者,以ID匯入 (Search for identifier (ORCID))
  • 方案二:自行建檔 (Default mode Submission)
  • 方案三:學科館員協助匯入 (Email worklist to subject librarians)

Built with DSpace-CRIS software - Extension maintained and optimized by 4Science