Lateralized Learning for Multi-class Visual Classification Tasks

Abstract

The majority of computer vision algorithms fail to find higher-order (abstract) patterns in an image so they are not robust against adversarial attacks. Deep learning considers each input pixel in a homogeneous manner such that different parts of a locality-sensitive hashing table are often not connected, meaning higher-order patterns are not discovered. Hence, these systems are sensitive to noisy, irrelevant, and redundant data, leading to wrong predictions with high confidence. Adversarial attacks exploit this vulnerability by generating deceptive inputs that mislead AI systems. In contrast, human vision is rarely susceptible to adversarial attacks. Vertebrate brains afford heterogeneous knowledge representation through lateralization, enabling modular learning at different levels of abstraction. This work aims to verify the effectiveness, scalability, and robustness of a lateralized approach to real-world problems that contain noisy, irrelevant, and redundant data. Two well-known and widely used adversarial attacks, the Fast Gradient Sign Method and the Iterative Adversarial Technique, are applied to generate corrupted test images. The experimental results on multi-class (200 classes) image classification tasks demonstrate that the proposed system effectively captures hierarchical knowledge representations, enhancing its robustness. Crucially, the lateralized system outperformed four state-of-the-art deep learning-based systems for the classification of normal and adversarial images by 19.05% − 41.02% and 1.36% − 49.22%, respectively.