BriefGPT.xyz
Feb, 2019
身份危机：极端过参数化下的记忆和泛化
Identity Crisis: Memorization and Generalization under Extreme Overparameterization
HTML
PDF
Chiyuan Zhang, Samy Bengio, Moritz Hardt, Yoram Singer
TL;DR
研究过参数化网络在单个训练样本和恒等映射任务中，记忆与泛化之间的相互作用。通过正式表征单层FCNs和CNNs中的泛化，我们展示了不同架构的归纳偏见。我们的工作有助于量化和可视化感性偏见对架构选择（如深度，内核宽度和通道数量）的敏感性。
Abstract
We study the interplay between
memorization
and
generalization
of overparametrized networks in the extreme case of a single training example. The learning task is to predict an output which is as similar as possi
→