BriefGPT.xyz
Feb, 2017
具有高斯输入的 ConvNet 的全局最优梯度下降
Globally Optimal Gradient Descent for a ConvNet with Gaussian Inputs
HTML
PDF
Alon Brutzkus, Amir Globerson
TL;DR
在神经网络模型中,使用Gradient descent算法时,当输入分布满足高斯分布时,使用Convolutional neural network和ReLU activations的神经网络模型可以在多项式时间内收敛于全局最优点。但是,我们证明了这种情况下学习是NP完全问题。
Abstract
deep learning
models are often successfully trained using
gradient descent
, despite the worst case hardness of the underlying non-convex optimization problem. The key question is then under what conditions can on
→