Computer Vision News 32 "Datasets through the LookingGlass" is a webinar series focused on introspecting into the datarelated facets of Machine Learning (ML) methods. Our goal is to build a community of enthusiastic researchers interested who care about understanding the impact that data and ML methods could have in our society. The webinar is part of “Making MetaDataCount” project and is organized by Veronika Cheplygina (left in the picture) and Amelia Jiménez-Sánchez (right in the picture) at IT University of Copenhagen. We had four successful editions so far (in February, June, September, and December 2023) with 14 speakers in total, the videos are available on our YouTube playlist. Datasets through the L king-Glass In our last webinar, we covered several topics about spurious correlations or shortcuts, fairness, out-of-distribution data and augmenting annotations of publicly available medical image datasets. Jessica Schrouff is a research scientist at Google DeepMind, working on responsible AI through a causal perspective. Jessica discussed in her talk how a model could learn sensitive characteristics due to various correlations in a dataset. They hypothesize that when a model learns a shortcut, increasing the encoding of the sensitive attribute affects the (un)fairness metrics. She uses the term unfairness as disparities in model output across demographic groups. Their method correctly detects shortcut learning when a spurious correlation is engineered, and they could vary the amount
RkJQdWJsaXNoZXIy NTc3NzU=