Kamrul Hasan

This work proposes CoAtNet-DeepMoE, a Convolution-Attention hybrid architecture for rich feature extraction, further enhanced with a DeepSeek Mixture of Experts to substantially reduce the number of parameters without sacrificing accuracy, and evaluates its model on both balanced and imbalanced datasets.