We proposed flipover, an new approach not only serves as a more effective regularization technique than conventional dropout, mitigating overfitting, but also introduces adversarial perturbations to gradients, enhancing resilience against adversairal attacks.
