[Q1.1] On similar approaches

Thank you for the suggestions. Unlike the traditional poisoning task, adversarial training data can be used in a good way. DeepConfuse before the data release. Thank you for the suggestions. Top row: Weights trained on clean training data.