Skip to content

Translation in progress Some entries still include text in the original language.

Nicholas Carlini Writing·Indexed by this site · Originally published

Nicholas Carlini 回顾《Towards Evaluating the Robustness of Neural Networks》五年后的反思

Reflecting on Towards Evaluating the Robustness of Neural Networks

AI summary

Nicholas Carlini 撰文回顾其五年前与他人合著的对抗样本论文《Towards Evaluating the Robustness of Neural Networks》,总结该工作做对了什么以及存在的问题。他认为这篇论文最正确的核心主张是用基于梯度下降的攻击来评估神经网络鲁棒性——当时学界流行的是定制化启发式攻击,难以迁移到带防御模型的评估中。

Read originalnicholas.carlini.com