arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2007.11259cs.LGcs.CVstat.ML
收藏

对抗性训练减少信息并提高可转移性

Adversarial Training Reduces Information and Improves Transferability

University of Padova(帕多瓦大学);University of California, Los Angeles(加利福尼亚大学洛杉矶分校)

Matteo Terzi, Alessandro Achille, Marco Maggipinto, Gian Antonio Susto

摘要:

最近的研究结果表明,用于分类的逆向训练网络的特征除了具有鲁棒性外,还具有理想的特性,如可逆性。后一个属性可能看起来违反直觉,因为社区广泛接受分类模型应该只捕获任务所需的最小信息(特征)。基于这种差异,我们研究了对抗训练和信息论之间的双重关系。我们表明,对抗训练可以提高新任务的线性可转移性,从而在表示的可转移性和源任务的准确性之间产生新的权衡。我们使用在CIFAR-10,CIFAR-100和ImageNet上训练的强大网络在几个数据集上验证了我们的结果。此外,我们表明对抗训练减少了关于输入和任务权重的表示的Fisher信息,并且我们提供了一个理论论据,解释了确定性网络的可逆性,而不违反最小性原则。最后,我们利用我们的理论见解,通过反演显着提高重建图像的质量。

英文摘要:

Recent results show that features of adversarially trained networks for classification, in addition to being robust, enable desirable properties such as invertibility. The latter property may seem counter-intuitive as it is widely accepted by the community that classification models should only capture the minimal information (features) required for the task. Motivated by this discrepancy, we investigate the dual relationship between Adversarial Training and Information Theory. We show that the Adversarial Training can improve linear transferability to new tasks, from which arises a new trade-off between transferability of representations and accuracy on the source task. We validate our results employing robust networks trained on CIFAR-10, CIFAR-100 and ImageNet on several datasets. Moreover, we show that Adversarial Training reduces Fisher information of representations about the input and of the weights about the task, and we provide a theoretical argument which explains the invertibility of deterministic networks without violating the principle of minimality. Finally, we leverage our theoretical insights to remarkably improve the quality of reconstructed images through inversion.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑