The loss surface of deep neural networks has recently attracted interest in the optimization and machine learning communities as a prime example of high-dimensional non-convex problem. Some insights were recently gained using spin glass models and mean-field approximations, but at the expense of strongly simplifying the nonlinear nature of the model. In this work, we do not make any such assumption and study conditions on the data distribution and model architecture that prevent the existence of bad local minima. Our theoretical work quantifies and formalizes two important \emph{folklore} facts: (i) the landscape of deep linear networks has a radically different topology from that of deep half-rectified ones, and (ii) that the energy landscape in the non-linear case is fundamentally controlled by the interplay between the smoothness of the data distribution and model over-parametrization. These results are in accordance with empirical practice and recent literature. %Together with %recent results that rigorously establish that no gradient descent can %get stuck on saddle points, we conclude that gradient descent converges %to a global optimum in deep rectified networks. The conditioning of gradient descent is the next challenge we address. We study this question through the geometry of the level sets, and we introduce an algorithm to efficiently estimate the regularity of such sets on large-scale networks. Our empirical results show that these level sets remain connected throughout all the learning phase, suggesting a near convex behavior, but they become exponentially more curvy as the energy level decays, in accordance to what is observed in practice with very low curvature attractors.

本文研究深度神经网络优化问题中的高维非凸性质，通过对数据分布和模型进行分析得出深度线性网络与半修正网络拓扑结构差异明显、非线性问题基于数据分布平滑程度和模型过度参数化的相互影响，通过证明半修正单层网络的渐进连通性，以及通过分析水平面的几何特征来研究梯度下降的调节。实验结果显示，虽然吸引子很小，但这些水平面在所有的学习阶段都保持连通。

半整流网络优化的拓扑和几何