Zhang, Q.; Rao, L.; Yang, Y. A Novel Visual Interpretability for Deep Neural Networks by Optimizing Activation Maps With Perturbation. AAAI 2021, 35, 3377-3384.