The role of variability in appearance in dynamic face learning conditions

We recognize familiar faces even in unpredictable situations where appearance and viewing conditions differ. In lab-based studies, such variations are termed "within-person variability" and seem crucial for face learning. Research with uncontrolled stimuli shows that learning faces with greater within-person variability generally enhances recognition performance but the learning mechanism itself is unclear. Moreover, recent data suggest that in the real world, stability in appearance specifically facilitates learning during early stages. To reconcile these data, we propose a dynamic and cost-efficiency face learning mechanism through which diagnostic facial information would be stored following a coarse-to-fine encoding mechanism. Exposure to stable facial information would quickly yield coarse and efficient representations, while exposure to increased variations would encourage a more costly encoding of finer diagnostic details.
To test this hypothesis in ecological conditions, we used a new video database showing dynamic faces in controlled changing conditions. Participants learned four women’s faces, with two displaying a stable appearance and the other two a variable appearance (i.e., appearance condition) and were assigned to one of four learning groups, exposed to 3, 6, 9, or 12 videos per identity. Using this setup, we conducted a series of four online studies. We found that recognition performance was consistently higher under stable conditions, especially when the test images were similar to the learning materials. No interaction was observed between learning groups and appearance conditions. These findings provide valuable insights into how face learning occurs under ecological but highly controlled conditions and enrich theoretical models of face recognition.