Can overfitted deep neural networks in adversarial training generalize? -- An approximation viewpoint

Open in new window