Lachlan E. MacDonald, Sameera Ramasinghe, Simon Lucey
TL;DR本文介绍了一种使用李群上的卷积(group convolutions over Lie groups)来实现任何形变的不变性的严谨数学框架,经实验证明在具有仿射不变性的分类任务中,我们的方法比传统CNN提高了30%的准确性,同时优于大多数CapsNets。
Abstract
Although provably robust to translational perturbations, convolutional neural networks (CNNs) are known to suffer from extreme performance degradation when presented at test time with more general geometric transformations of inputs. Recently, this limitation has motivated a shift in f