BriefGPT.xyz
Sep, 2020
学习难以改变的解释
Learning explanations that are hard to vary
HTML
PDF
Giambattista Parascandolo, Alexander Neitz, Antonio Orvieto, Luigi Gresele, Bernhard Schölkopf
TL;DR
本文研究在深度学习中“好的解释很难变化”的原则，指出在梯度平均时往往偏向记忆化和拼凑的解决方案而忽视了不变性，在此基础上提出了一种基于逻辑AND的简单算法并在多个真实任务上进行测试。最后使用一组合成数据集和常见正则化方法进行比较。
Abstract
In this paper, we investigate the principle that `good explanations are hard to vary' in the context of
deep learning
. We show that averaging gradients across examples -- akin to a logical OR of patterns -- can favor
me
→