Towards Measuring Bias in Image Classification

Convolutional Neural Networks (CNN) have become de facto state-of-the-art for the main computer vision tasks. However, due to complex underlying structure their decisions are hard understand which limits use in some context of industrial world. A common and detect challenge machine learning (ML) tasks is data bias. In this work, we present a systematic approach uncover bias by means attribution maps. For purpose, first an artificial dataset with known created used train intentionally biased CNNs. The networks’ then inspected using Finally, meaningful metrics measure maps’ representativeness respect proposed study shows that map techniques highlight presence better than others can support identification