Abstract :
[en] Humans can be remarkably proficient at recognising familiar faces, yet learning new identities remains surprisingly difficult. Although laboratory studies have consistently suggested that exposure to variability facilitates face learning, observations from everyday life and our previous theoretical work point to a more nuanced picture. Building on the cost-efficient theory of face learning, we propose that facial representations develop through adaptive allocation of encoding resources: learners initially rely on coarse diagnostic information, while finer facial details are encoded only when variability makes these coarse representations insufficient. In this talk, I will present a research programme designed to test this framework. First, I will introduce DIVA-Face, a highly controlled yet ecologically valid face database that independently manipulates facial appearance, viewing conditions, and facial dynamics, providing a unique tool to investigate the mechanisms underlying face learning. Using this database, we revisited one of the field's central assumptions. Across multiple preregistered experiments, variability in appearance did not consistently improve learning. Instead, its effects depended on factors such as the learning procedure, the correspondence between learning and test images, the amount of exposure or the number of items to learn, suggesting that memory demands and resource allocation constrain the benefits of variability. Finally, I will show that learners also strategically allocate encoding resources according to expected future utility, preferentially learning identities that are more likely to be encountered again. Together, these findings suggest that variability is neither inherently beneficial nor detrimental to learning. Rather, its effects are shaped by the interaction between attentional and memory resources, learning strategies, and task demands. I will argue that a cost-efficient framework provides a unified account of these findings and offers new research avenues on how robust face representations emerge through experience.
