Understanding deep learning requires rethinking generalization. Zhang, C., Bengio, S, Hardt, M, Recht, B, & Vinyals, O. ArXiv e-prints, November, 2016.
Understanding deep learning requires rethinking generalization [link]Paper  bibtex   
author = {Zhang, Chiyuan and Bengio, S and Hardt, M and Recht, B and Vinyals, Oriol},
title = {{Understanding deep learning requires rethinking generalization}},
journal = {ArXiv e-prints},
year = {2016},
volume = {cs.LG},
month = nov,
annote = {Just read the contributions should be fine. Essentially, many theories on generalization don't work on deep learning models.

Eric Xing's comments on this paper, from his Facebook:

Not surprisingly, most statistical learning theories on generalization errors cannot be applied on deep learning, and DL is actually not learning any patterns -- they are just memorizing the data.
keywords = {deep learning},
read = {Yes},
rating = {4},
date-added = {2017-02-14T03:28:49GMT},
date-modified = {2017-02-28T18:35:28GMT},
url = {http://arxiv.org/abs/1611.03530},
local-url = {file://localhost/Users/yimengzh/Documents/Papers3_revised/Library.papers3/Articles/2016/Zhang/arXiv%202016%20Zhang.pdf},
file = {{arXiv 2016 Zhang.pdf:/Users/yimengzh/Documents/Papers3_revised/Library.papers3/Articles/2016/Zhang/arXiv 2016 Zhang.pdf:application/pdf}},
uri = {\url{papers3://publication/uuid/D566AE84-4CA7-4A74-9FCB-637EA4CA0854}}

Downloads: 0