Proceedings of the Annual Meeting of the Cognitive Science Society

Current face-comparison tests use a fixed set of stimuli, such that task difficulty is not tailored to the participant’s abilityto perform face matching, which varies greatly across people. Here, we create an adaptive version of the Glasgow FaceMatching Test (GFMT). To accomplish this, we make use of recent advances in machine learning that can encode pho-tographs into a learned face space and then generate photorealistic morphs that interpolate between mid-level features ofthe depicted individuals. In particular, we first use the StyleGAN neural-network architecture to generate challenging vari-ants of the GFMT. We then use QUEST+, a Bayesian adaptive psychometric testing procedure, to estimate the observer’ssensitivity to appearance changes during face matching. The resulting test, the adaptive GFMT (aGFMT), aims to moreefficiently estimate a participant’s face-matching ability.