Example of WikiArtVectors retrieving similar art works to Picasso's 'The old blind guitarist.' Left: similar artworks by style. Right: similar artworks by color. (Image: Desikan, Shimao, and Miton, 2022)

We’ve all seen art made from data, but what about data from art? In a feature paper in Entropy, Bhargav Srinivasa Desikan (École Polytechnique Fédérale de Lausanne), Hajime Shimao (McGill University, former SFI Postdoctoral Fellow), and SFI Complexity Postdoctoral Fellow Helena Miton released a novel dataset for indexing, searching, retrieving, organizing, and analyzing 68,094 works of art by more than 1600 historically significant artists. Using state-of-the-art machine learning, the authors were able to extract both style representations and color distributions, which can be used to query stylistic periods for an artist or a movement (eg, Picasso’s “blue” phase). Their dataset, WikiArtVectors, aims to make computational data approaches available to art historians and cultural analysts, to help discover and understand patterns of cultural evolution.