Title of Talk
Machine vision in social computing
Images are powerful tools to understand properties of individuals and communities at different scales. By analyzing small curated image datasets, we can study how people subjectively perceive the visual world, e.g. what makes an image beautiful, or how we judge pictures of faces. Larger ecosystems, i.e. photo sharing platforms, allow to analyze communities at scale through the lens of the milions of pictures they generate everyday. By leveraging user and image metadata, we can investigate how visual perceptions and trends vary according to users’ culture, language, and popularity dynamics. When online visual content is geolocated, computer vision tools can also be used to discover properties of the real, physical world, for example the beauty or ambiance of the streets in a city.
Large quantities of visual content can also be found in online encyclopedias, as images play a crucial role in knowledge sharing and dissemination. Although images in these environments come with rich metadata regarding content, quality, epistemic significance, cultural aspects, this precious resource has somehow been left untapped, and the large-scale study of images’ encyclopedic and epistemic value largely unexplored.
Miriam Redi is a Research Scientist at the Wikimedia Foundation and Visiting Research Fellow at King’s College London. Formerly, she worked as a Research Scientist at Yahoo Labs Barcelona, Yahoo Research London and Nokia Bell Labs in Cambridge. She received her PhD from EURECOM, Sophia Antipolis. She conducts research in social multimedia computing, working on fair, interpretable, multimodal machine learning solutions to improve knowledge equity.