Improved Deep Neural Network Generalization Using m-Sharpness-Aware Minimization