Robustness of Deep Learning Architectures with Respect to Training Data Variation

The use of deep neural networks (DNN) revolutionized machine learning tasks like image classification and segmentation [6]. However, DNNs are black box models developed with the aim to generalize a given training set and thus to process unknown data. Due to the high amount of model parameters, it is yet impossible to fully understand the decision making process. The identification of parameters with poor generalization is hard and manual improvement is impossible. Thus, robustness is subject of consideration.

[1]  Lina J. Karam,et al.  Understanding how image quality affects deep neural networks , 2016, 2016 Eighth International Conference on Quality of Multimedia Experience (QoMEX).

[2]  Sergey Ioffe,et al.  Rethinking the Inception Architecture for Computer Vision , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[3]  Michael S. Bernstein,et al.  ImageNet Large Scale Visual Recognition Challenge , 2014, International Journal of Computer Vision.

[4]  Alex Krizhevsky,et al.  Learning Multiple Layers of Features from Tiny Images , 2009 .

[5]  Jonathon Shlens,et al.  Explaining and Harnessing Adversarial Examples , 2014, ICLR.

[6]  Samy Bengio,et al.  Adversarial examples in the physical world , 2016, ICLR.

[7]  Joan Bruna,et al.  Intriguing properties of neural networks , 2013, ICLR.

[8]  P. Cochat,et al.  Et al , 2008, Archives de pediatrie : organe officiel de la Societe francaise de pediatrie.

[9]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[10]  Zhuowen Tu,et al.  Aggregated Residual Transformations for Deep Neural Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).