Squeeze-and-Excitation Algorithm for Efficient Face Recognition

Abstract

Convolutional neural networks (CNNs) are inspired by biological vision, yet their success usually depends on deeper architectures and large-scale data. By contrast, primate face processing appears to rely on compact, low-dimensional representations that reduce redundancy while preserving useful information. Drawing on efficient-coding principles and neural evidence on face representation, this study tested whether a lightweight squeeze-and-excitation (SE) mechanism could enhance AlexNet. Across three experiments, SE-AlexNet with a compression ratio of 16 achieved behavioral performance most closely aligned with human facial-emotion judgments. The benefit depended jointly on compression strength and insertion stage, with the best performance observed when the SE module was inserted late, between FC7 and FC8. Grad-CAM visualizations and feature-occlusion tests further indicated that the optimized model preferentially relied on the eyes and mouth, diagnostically important regions in human facial emotion perception. These findings provide computational support for biologically inspired representational compression and offer a framework for lightweight, interpretable, and cognitively plausible face-recognition models.