Face processing emerges from object-trained convolutional neural networks

2024-05-29

arXiv.org

AI Summary

Whether face processing relies on domain-specific, innate mechanisms or can spontaneously emerge from general-purpose object recognition systems remains a fundamental question in cognitive neuroscience. Method: We systematically evaluated the zero-shot face recognition capability of standard convolutional neural networks (e.g., ResNet, VGG) pretrained exclusively on object-centric datasets (e.g., ImageNet), with no exposure to face stimuli. Using representational similarity analysis (RSA) and neural response modeling, we compared model behavior against human psychophysical and neuroimaging data. Contribution/Results: Object-pretrained models achieve near-human performance in face identification, identity discrimination, and pareidolic face detection. Critically, they replicate the functional selectivity of the fusiform face area (FFA)—including category-specific activation patterns—despite lacking face-specific training. This provides the first systematic evidence that robust, human-like face processing can self-organize purely from object-driven statistical regularities, challenging the long-standing domain-specificity hypothesis of face perception.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Whether face processing depends on unique, domain-specific neurocognitive mechanisms or domain-general object recognition mechanisms has long been debated. Directly testing these competing hypotheses in humans has proven challenging due to extensive exposure to both faces and objects. Here, we systematically test these hypotheses by capitalizing on recent progress in convolutional neural networks (CNNs) that can be trained without face exposure (i.e., pre-trained weights). Domain-general mechanism accounts posit that face processing can emerge from a neural network without specialized pre-training on faces. Consequently, we trained CNNs solely on objects and tested their ability to recognize and represent faces as well as objects that look like faces (face pareidolia stimuli).... Due to the character limits, for more details see in attached pdf

Problem

• Does face processing require specialized mechanisms or general object recognition
• Can object-trained CNNs recognize faces without face-specific training
• Testing domain-general vs domain-specific hypotheses for face processing

Innovation

• Object-trained CNNs for face processing
• Testing domain-general recognition mechanisms
• Using face pareidolia stimuli for validation