AbstractIn this work, we study how the performance and evaluation of generative image models are impacted by racial composition their training datasets. By examining controlling distributions in various datasets, able to observe impacts different on generated quality images. Our results show that compositions images successfully preserve data. However, truncation, a technique used generate high...

This paper uses text analysis to measure gender bias in cities through the use of street names. Focusing on case Spain, we collect data 15 million names analyze inequality urban toponyms. We calculate for each Spanish municipality and year from 2001 2020 a variable measuring percentage streets with female over total number male Our results reveal strong imbalance cities: named after women is on...