The Functional Neuroanatomy of Face Processing: Insights from Neuroimaging and Implications for Deep Learning

Grill-Spector, Kalanit; Kay, Kendrick; Weiner, Kevin S.

doi:10.1007/978-3-319-61657-5_1

Cited by 5 publications

(4 citation statements)

References 173 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…FFA has shown consistent preference to central vs. peripheral vision (Levy et al, 2001;Hasson et al, 2002). In addition, most population receptive fields (pRFs) in FFA cover the fovea for varying pRF sizes (Kay et al, 2015;Grill-Spector et al, 2017a;Gomez et al, 2018), and thus while foveal stimuli are likely to activate most of the FFA, stimuli at eccentricities beyond central vision (in our case the 4° and 8° conditions) are likely to activate a smaller part of the FFA. Therefore, the observed rapid reduction in FFA's BOLD magnitude with eccentricity may have resulted from reduced activation in the activated voxels and/or from a reduction in the number of activated voxels.…”

Section: Limitationsmentioning

confidence: 67%

Parafoveal vision reveals qualitative differences between FFA and PPA

Kreichman¹,

Gilaie-Dotan²

2023

Preprint

View full text Add to dashboard Cite

The center-periphery visual field axis guides early visual system organization with enhanced resources devoted to central vision leading to reduced peripheral performance relative to that of central vision (i.e., behavioral eccentricity effect) for most visual functions. The center-periphery organization extends to high-order visual cortex where for example the well-studied face-sensitive fusiform face area (FFA) shows sensitivity to central vision and place-sensitive parahippocampal place area (PPA) shows sensitivity to peripheral vision. As we have recently found that face perception is more sensitive to eccentricity than place perception, here we examined whether these behavioral findings reflect differences in FFA and PPA's sensitivities to eccentricity. We assumed FFA would show higher sensitivity to eccentricity than PPA would, but that both regions' modulation by eccentricity would be invariant to the viewed category. We parametrically investigated (fMRI, n=32) how FFA's and PPA's activations are modulated by eccentricity (less than or equal to 8 degrees) and category (upright/inverted faces/houses) while keeping stimulus size constant. As expected, FFA showed an overall higher sensitivity to eccentricity than PPA. However, both regions' activation modulations by eccentricity were dependent on the viewed category. In FFA a reduction of activation with growing eccentricity ("BOLD eccentricity effect") was found (with different amplitudes) for all categories. In PPA however, there were qualitative modulations of the BOLD eccentricity effect with mild BOLD eccentricity effect for houses but a reverse BOLD eccentricity effect for faces and no modulation for inverted faces. Our results emphasize that peripheral vision investigations are critical to further our understanding of visual processing both quantitatively and qualitatively.

show abstract

Section: Limitationsmentioning

confidence: 67%

Parafoveal vision reveals qualitative differences between FFA and PPA

Kreichman¹,

Gilaie-Dotan²

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…FFA has shown consistent preference to central versus peripheral vision (Hasson et al, 2002 ; Levy et al, 2001 ). In addition, most population receptive fields (pRFs) in FFA cover the fovea for varying pRF sizes (Gomez et al, 2018 ; Grill‐Spector, Kay, et al, 2017 ; Kay et al, 2015 ), and thus, while foveal stimuli are likely to activate most of the FFA, stimuli at eccentricities beyond central vision (in our case the 4° and 8° conditions) are likely to activate a smaller part of the FFA. Therefore, the observed rapid reduction in FFA's BOLD magnitude with eccentricity may have resulted from reduced activation in the activated voxels and/or from a reduction in the number of activated voxels.…”

Section: Discussionmentioning

confidence: 88%

Parafoveal vision reveals qualitative differences between fusiform face area and parahippocampal place area

Kreichman,

Gilaie‐Dotan

2024

Human Brain Mapping

View full text Add to dashboard Cite

The center‐periphery visual field axis guides early visual system organization with enhanced resources devoted to central vision leading to reduced peripheral performance relative to that of central vision (i.e., behavioral eccentricity effect) for many visual functions. The center‐periphery organization extends to high‐order visual cortex where, for example, the well‐studied face‐sensitive fusiform face area (FFA) shows sensitivity to central vision and the place‐sensitive parahippocampal place area (PPA) shows sensitivity to peripheral vision. As we have recently found that face perception is more sensitive to eccentricity than place perception, here we examined whether these behavioral findings reflect differences in FFA's and PPA's sensitivities to eccentricity. We assumed FFA would show higher sensitivity to eccentricity than PPA would, but that both regions' modulation by eccentricity would be invariant to the viewed category. We parametrically investigated (fMRI, n = 32) how FFA's and PPA's activations are modulated by eccentricity (≤8°) and category (upright/inverted faces/houses) while keeping stimulus size constant. As expected, FFA showed an overall higher sensitivity to eccentricity than PPA. However, both regions' activation modulations by eccentricity were dependent on the viewed category. In FFA, a reduction of activation with growing eccentricity (“BOLD eccentricity effect”) was found (with different amplitudes) for all categories. In PPA however, qualitatively different BOLD eccentricity effect modulations were found (e.g., at 8° mild BOLD eccentricity effect for houses but a reverse BOLD eccentricity effect for faces and no modulation for inverted faces). Our results emphasize that peripheral vision investigations are critical to further our understanding of visual processing.

show abstract

“…Our results also show that a higher degree of invariance can be achieved through learning, as shown by the reduced bias for highly familiar faces. This finding highlights that to increase biological plausibility of models of vision, differences in eccentricity and receptive field size should be taken into account (Poggio et al, 2014), as well as more dynamic effects such as changes induced by learning and attention (Grill-Spector et al, 2017a).…”

Section: Discussionmentioning

confidence: 97%

“…Even the models that currently provide better fits to neural activity in IT such as hierarchical, convolutional neural networks (Yamins et al, 2014; Kriegeskorte, 2015; Yamins and DiCarlo, 2016) use weight sharing in convolutional layers to achieve position invariance (LeCun et al, 2015; Schmidhuber, 2015; Goodfellow et al, 2016). While this reduces complexity by limiting the number of parameters to be fitted, neuroimaging and behavioral experiments have shown that translational invariance in IT is preserved only for small displacements (DiCarlo and Maunsell, 2003; Kay et al, 2015; Silson et al, 2016; for a review see Kravitz et al, 2008), with varying receptive field sizes and eccentricities (Grill-Spector et al, 2017a). Our results highlight the limited position invariance for high-level judgments such as identity, and add to the known spatial heterogeneity for gender and age judgments (Afraz et al, 2010).…”

Section: Discussionmentioning

confidence: 99%

Idiosyncratic, retinotopic bias in face identification modulated by familiarity

Castello

Taylor

Cavanagh

et al. 2018

Preprint

View full text Add to dashboard Cite

34The perception of gender and age of unfamiliar faces is reported to vary 35 idiosyncratically across retinal locations such that, for example, the same 36 androgynous face may appear to be male at one location but female at another. 37Here we test spatial heterogeneity for the recognition of the identity of personally 38 familiar faces. We found idiosyncratic biases that were stable within subjects 39 and that varied more across locations for low as compared to high familiar faces.

show abstract

The Functional Neuroanatomy of Face Processing: Insights from Neuroimaging and Implications for Deep Learning

Cited by 5 publications

References 173 publications

Parafoveal vision reveals qualitative differences between FFA and PPA

Parafoveal vision reveals qualitative differences between FFA and PPA

Parafoveal vision reveals qualitative differences between fusiform face area and parahippocampal place area

Idiosyncratic, retinotopic bias in face identification modulated by familiarity

Contact Info

Product

Resources

About