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 2001.04878cs.LGstat.ML
收藏

深度神经网络的凸性与层宽的关系

On the Convex Behavior of Deep Neural Networks in Relation to the Layers' Width

Tel Aviv University(特拉维夫大学);Facebook AI Research(脸书人工智能研究院)

Etai Littwin, Lior Wolf

摘要:

神经网络的Hessian可以分解为两个矩阵的和:(i)半正定广义高斯-牛顿矩阵G,和(ii)包含负特征值的矩阵H。我们观察到,对于更宽的网络,通过在训练开始和结束时具有正曲率的表面以及在其间接近零曲率的表面,使用梯度下降优化策略来最小化损失。换句话说,似乎在训练过程的关键部分,宽网络中的Hessian由分量G主导。为了解释这一现象,我们证明了当使用普通方法初始化时,过参数化网络的梯度近似正交于H,使得损失曲面的曲率在梯度方向上严格为正。

英文摘要:

The Hessian of neural networks can be decomposed into a sum of two matrices: (i) the positive semidefinite generalized Gauss-Newton matrix G, and (ii) the matrix H containing negative eigenvalues. We observe that for wider networks, minimizing the loss with the gradient descent optimization maneuvers through surfaces of positive curvatures at the start and end of training, and close to zero curvatures in between. In other words, it seems that during crucial parts of the training process, the Hessian in wide networks is dominated by the component G. To explain this phenomenon, we show that when initialized using common methodologies, the gradients of over-parameterized networks are approximately orthogonal to H, such that the curvature of the loss surface is strictly positive in the direction of the gradient.

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。

↑