Parallel Deep Neural Networks Have Zero Duality Gap
Parallel Deep Neural Networks Have Zero Duality Gap
复制标题
并行深度神经网络具有零对偶间隙
DOI:
--
复制
发表时间:
2021
期刊:
影响因子:
--
通讯作者:
Mert Pilanci
中科院分区:
文献类型:
--
作者:
Yifei Wang;Tolga Ergen;Mert Pilanci
Training deep neural networks is a challenging non-convex optimization problem. Recent work has proven that the strong duality holds (which means zero duality gap) for regularized finite-width two-layer ReLU networks and consequently provided an equivalent convex training problem. However, extending this result to deeper networks remains to be an open problem. In this paper, we prove that the duality gap for deeper linear networks with vector outputs is non-zero. In contrast, we show that the zero duality gap can be obtained by stacking standard deep networks in parallel, which we call a parallel architecture, and modifying the regularization. Therefore, we prove the strong duality and existence of equivalent convex problems that enable globally optimal training of deep networks. As a by-product of our analysis, we demonstrate that the weight decay regularization on the network parameters explicitly encourages low-rank solutions via closed-form expressions. In addition, we show that strong duality holds for three-layer standard ReLU networks given rank-1 data matrices.
登录
查看更多内容
DOI:
10.48550/arxiv.2205.08078
发表时间:
2022-05
期刊:
ArXiv
影响因子:
--
作者:
Arda Sahiner;Tolga Ergen;Batu Mehmet Ozturkler;J. Pauly;M. Mardani;Mert Pilanci
通讯作者:
Arda Sahiner;Tolga Ergen;Batu Mehmet Ozturkler;J. Pauly;M. Mardani;Mert Pilanci
DOI:
--
发表时间:
2020-07
期刊:
ArXiv
影响因子:
--
作者:
E. Moroshko;Suriya Gunasekar;Blake E. Woodworth;J. Lee;N. Srebro;Daniel Soudry
通讯作者:
E. Moroshko;Suriya Gunasekar;Blake E. Woodworth;J. Lee;N. Srebro;Daniel Soudry
影响因子:
6
作者:
Ergen, T.;Pilanci, M.
通讯作者:
Pilanci, M.
DOI:
--
发表时间:
2022
期刊:
International Conference on Learning Representations
影响因子:
--
作者:
Ergen, T.;Sahiner, A.;Ozturkler, B.;Pauly, J.;Mardani, M.;Pilanci, M.
通讯作者:
Pilanci, M.
DOI:
--
发表时间:
2020-12
期刊:
ArXiv
影响因子:
--
作者:
Arda Sahiner;Tolga Ergen;J. Pauly;Mert Pilanci
通讯作者:
Arda Sahiner;Tolga Ergen;J. Pauly;Mert Pilanci