This shows you the differences between two versions of the page.

Link to this comparison view

label_smoothing [2016/12/13 16:20] (current)
Line 1: Line 1:
 +https://​arxiv.org/​abs/​1606.03498v1 Improved Techniques for Training GANs
 +Label smoothing, a technique from the 1980s recently independently re-discovered by Szegedy et.
 +al [17], replaces the 0 and 1 targets for a classifier with smoothed values, like .9 or .1, and was
 +recently shown to reduce the vulnerability of neural networks to adversarial examples [18].
 +[17] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna. Rethinking the Inception Architecture for
 +Computer Vision. ArXiv e-prints, December 2015.
 +[18] David Warde-Farley and Ian Goodfellow. Adversarial perturbations of deep neural networks. In Tamir
 +Hazan, George Papandreou, and Daniel Tarlow, editors, Perturbations,​ Optimization,​ and Statistics, chapter
 +11. 2016. Book in preparation for MIT Press.