Abstract
We recently proposed a shape-based computational model of face discrimination in the FFA, in which a particular face is represented by a sparse population code over highly selective face neurons tuned to different face shapes. This model has been quantitatively supported by data from both behavioral and fMRI adaptation experiments (Jiang et al., 2006). In contrast, recent human fMRI (Loffler et al., 2005) and monkey electrophysiology studies (Leopold et al., 2006) have suggested an alternative conceptual “norm-based” model, in which face perception is mediated by neurons that are tuned around an “average” face. However, by design such studies involve the repeated presentation of stimuli similar to the “average” face, which could lead to the selective adaptation of neurons selective for the “average” face, creating the impression of a neuronal population that responds little to the “average” face and increasingly more to faces different from the “average”. To directly test this hypothesis, we created four different faces spaces, centered either around an average face (as in Leopold et al., 2006), or around a specific individual face (i.e., belonging to a particular individual instead of the “average”), for both genders. This design equalized adaptation effects in both cases, allowing us to test if the “average” face had a special status, as predicted by the “norm-based” theories. For all four face spaces, we then tested discrimination performance with a 2AFC behavioral paradigm (n=10), and the selectivity of FFA neurons with a short-block fMRI adaptation paradigm (n=5). The data from both experiments revealed a significant effect of face shape difference (“morph steps”, M2/4/6/8/10 for behavioral, M0/3/6/9 for fMRI testing), but no significant difference between the “average” and the “individual” face spaces, and no interaction, suggesting that prior reports of a “norm-based” face representation might have been confounded by adaptation effects.
This study is supported by NIMH grant R01 MH076281