BriefGPT.xyz
Jun, 2021
从示例难度的角度看深度学习
Deep Learning Through the Lens of Example Difficulty
HTML
PDF
Robert J. N. Baldock, Hartmut Maennel, Behnam Neyshabur
TL;DR
本研究采用基于个例角度的视角,引入了计算预测输入的困难程度的度量-(有效的)预测深度,发现了一些有意为之而意外的、简单的关系,进一步将困难例子分为三个可解释的组,并展示了这种理解如何提高了预测准确率,揭示了文献中一些分开探讨的现象的连贯视图。
Abstract
Existing work on understanding
deep learning
often employs measures that compress all data-dependent information into a few numbers. In this work, we adopt a perspective based on the role of
individual examples
.
→