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

Re-assign community