Neuroscience-Inspired Neural Network Exhibits More Human-Like Face Processing Abilities

This study introduces an updated version of AlexNet called Squeeze-and-Excitation (SE)-AlexNet inspired by neuroscience findings. We compared SE-AlexNet's performance with the original AlexNet and VGG16 models (for object-based and face-based ones), focusing on their ability to imitate human face processing. Using an adapted behavioral experimental method, we assessed the models' perceptual sensitivity and accuracy in face perception. SE-AlexNet, especially with a 16-fold compression ratio, outperformed the other models, closely matching human performance in recognizing facial emotions. These results suggest that a neuroscience-inspired CNN model can replicate human-like perception in specific computer vision tasks rather than more complicated models.