arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1802.09308cs.LG
收藏

Max-Mahalanobis线性判别分析网络

Max-Mahalanobis Linear Discriminant Analysis Networks

Tsinghua University(清华大学)

Tianyu Pang, Chao Du, Jun Zhu

摘要:

深度神经网络(DNN)由从输入到特征表示的非线性变换组成,然后是常见的softmax线性分类器。虽然许多努力已经致力于设计一个合适的非线性变换的架构,很少有调查已经做了分类器的一部分。在本文中,我们表明,一个适当设计的分类器可以提高鲁棒性的对抗攻击,并导致更好的预测结果。具体来说,我们定义了一个Max-Mahalanobis分布(MMD),并从理论上证明,如果输入分布为MMD,线性判别分析(LDA)分类器将具有最好的鲁棒性对抗的例子。我们进一步提出了一种新的Max-Mahalanobis线性判别分析(MM-LDA)网络,该网络将输入空间中的复杂数据分布显式映射到潜在特征空间中的MMD,然后应用LDA进行预测。我们的研究结果表明,MM-LDA网络是显着更强大的对抗攻击,并有更好的性能,在类偏置分类。

英文摘要:

A deep neural network (DNN) consists of a nonlinear transformation from an input to a feature representation, followed by a common softmax linear classifier. Though many efforts have been devoted to designing a proper architecture for nonlinear transformation, little investigation has been done on the classifier part. In this paper, we show that a properly designed classifier can improve robustness to adversarial attacks and lead to better prediction results. Specifically, we define a Max-Mahalanobis distribution (MMD) and theoretically show that if the input distributes as a MMD, the linear discriminant analysis (LDA) classifier will have the best robustness to adversarial examples. We further propose a novel Max-Mahalanobis linear discriminant analysis (MM-LDA) network, which explicitly maps a complicated data distribution in the input space to a MMD in the latent feature space and then applies LDA to make predictions. Our results demonstrate that the MM-LDA networks are significantly more robust to adversarial attacks, and have better performance in class-biased classification.

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。

↑