JSAI2024

Presentation information

Organized Session

Organized Session » OS-2

[3K5-OS-2b] OS-2

Thu. May 30, 2024 3:30 PM - 4:50 PM Room K (Room 44)

オーガナイザ:鈴木 健二(ソニーグループ株式会社)、原 聡(大阪大学)、谷中 瞳(東京大学)、菅原 朔(国立情報学研究所)

3:30 PM - 3:50 PM

[3K5-OS-2b-01] The Effects of Generated Data on Future Datasets

〇Ryuichiro Hataya1 (1. RIKEN)

Keywords:generative AI

Recent deep generative models can generate high-quality and realistic data from users' instructions, and the generated data are uploaded to the Internet.
Meanwhile, deep learning, including generative models, relies on large-scale datasets collected from the Internet.
They imply that datasets to train deep learning models in the future will be ``contaminated'' by such generated data.
This paper discusses the potential negative effects of generated data of large-scale generative models on datasets in the future, based on our paper [Hataya et al. 2023].

Authentication for paper PDF access

A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.

Password