Adaptive face space models with dynamic neural priors and sparse coding

Visual adaptation leads to certain changes in the perceived identity of a face depending on the previous faces an observer has been exposed to. The explicit effect of adaptation on face perception is not well understood. This work presents, for what is believed to be the first time, a mathematical model to capture the effect of adaptation on face identity perception. Our model is grounded in three assumptions that are consistent with recent neurobiological findings: (1) adaptation has the highest effect on the neural activities generated at the lowest levels of visual cortex; (2) neural activities in low-level visual areas are representatives of extracted features from a stimulus face, and as a result, encoding the stimulus face in the face space is affected by the adaptation process; and (3) a stimulus face is represented by a small number of simultaneously active neurons out of a large population (sparse coding). The rate of target identification with or without adaptation obtained by the proposed computational model resembles psyhometric functions obtained in prior studies using real subjects.