Image aesthetic scorers prioritize fidelity over demographic bias, study finds

A new audit of image aesthetic scoring models reveals that these systems primarily prioritize image fidelity rather than demographic attributes. Researchers found that while some scorers showed a preference for darker skin tones on synthetic images, this trend reversed when tested on real-world images. The study highlights the limitations of using synthetic data for bias audits and emphasizes the need for causal isolation on real data to accurately assess demographic bias. AI

影响 Highlights potential biases in AI models used for data filtering and generation, emphasizing the need for robust auditing on real-world data.

排序理由 Academic paper detailing a new audit methodology for AI models. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

• alphaXiv
• CatalyzeX
• CIELAB color space
• DagsHub
• Gotit.pub
• HPSv2
• Hugging Face
• ImageReward
• LAION-Aesthetics
• PickScore
• ScienceCast

• 论文
• 安全

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →