EUVIMEDEUVIMED entdecken
Crossref

Exploiting Cultural Biases via Homoglyphs in Text-to-Image Synthesis

Lukas Struppek, Dom Hintersdorf, Felix Friedrich, Manuel Br, Patrick Schramowski, Kristian Kersting

Journal of Artificial Intelligence Research · 18.12.2023 · 78 · 1017-1068

DOI: 10.1613/jair.1.15388 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Artificial Intelligence Research, online erschienen am 18.12.2023. Verfasst von Lukas Struppek und 5 weiteren Autor:innen. Die Arbeit wurde bisher 35-mal zitiert (Crossref).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Models for text-to-image synthesis, such as DALL-E 2 and Stable Diffusion, have recently drawn a lot of interest from academia and the general public. These models are capable of producing high-quality images that depict a variety of concepts and styles when conditioned on textual descriptions. However, these models adopt cultural characteristics associated with specific Unicode scripts from their vast amount of training data, which may not be immediately apparent. We show that by simply inserting single non-Latin characters in the textual description, common models reflect cultural biases in their generated images. We analyze this behavior both qualitatively and quantitatively and identify a model’s text encoder as the root cause of the phenomenon. Such behavior can be interpreted as a model feature, offering users a simple way to customize the image generation and reflect their own cultural background. Yet, malicious users or service providers may also try to intentionally bias the image generation. One goal might be to create racist stereotypes by replacing Latin characters with similarly-looking characters from non-Latin scripts, so-called homoglyphs. To mitigate such unnoticed script attacks, we propose a novel homoglyph unlearning method to fine-tune a text encoder, making it robust against homoglyph manipulations.

Zitieren

APA

Struppek, L., Hintersdorf, D., Friedrich, F., Br, M., Schramowski, P., & Kersting, K. (2023). Exploiting Cultural Biases via Homoglyphs in Text-to-Image Synthesis. Journal of Artificial Intelligence Research, 78, 1017-1068. https://doi.org/10.1613/jair.1.15388

Vancouver

Struppek L, Hintersdorf D, Friedrich F, Br M, Schramowski P, Kersting K. Exploiting Cultural Biases via Homoglyphs in Text-to-Image Synthesis. Journal of Artificial Intelligence Research. 2023;78:1017-1068. doi:10.1613/jair.1.15388

BibTeX
@article{Struppek2023,
  author = {Struppek, Lukas and Hintersdorf, Dom and Friedrich, Felix and Br, Manuel and Schramowski, Patrick and Kersting, Kristian},
  title = {Exploiting Cultural Biases via Homoglyphs in Text-to-Image Synthesis},
  journal = {Journal of Artificial Intelligence Research},
  year = {2023},
  volume = {78},
  pages = {1017--1068},
  doi = {10.1613/jair.1.15388}
}
Kostenlos starten →