Invisible bias: effects on evaluation(Talk by Noa Garcia, The University of Osaka)

Noa Garcia of The University of Osaka gives an online seminar hosted by RIKEN AIP on invisible biases in computer vision evaluation: factors correlated with a model's outcome that are not visible in the data, yet systematically skew what a benchmark measures. Drawing on three recent studies, the talk covers bias introduced by camera acquisition and processing parameters, data leakage where evaluation images reappear identically or near-identically in training data and inflate reported performance, and gender bias in vision-language models driven by demographic priors rather than visual content. The talk frames bias in evaluation not only as a fairness issue but as a general threat to the construct validity of benchmarks, models and visual representations. It runs 17:00 to 18:00 JST and registration is required.