寻找射影等变网络
计算机视觉与模式识别
2023-12-21 v3
摘要
线性神经网络层的等变性已被充分研究。在本工作中,我们将等变条件放宽至仅在射影意义下成立。我们提出了一种构建射影等变神经网络的方法:先构建一个标准等变网络,其中作用于每个中间特征空间的线性群表示是射影群表示的“乘性修正提升”。通过从理论上研究射影等变与线性等变线性层之间的关系,我们表明所提方法在使用线性层构建网络时是具有最大一般性的。该理论在两个简单实验中得到展示。
引用
@article{arxiv.2209.14719,
title = {In Search of Projectively Equivariant Networks},
author = {Georg Bökman and Axel Flinth and Fredrik Kahl},
journal= {arXiv preprint arXiv:2209.14719},
year = {2023}
}
备注
v3: Another significant rewrite. Accepted for publication in TMLR. v2: Significant rewrite. The title has been changed: "neural network" -> "network". More general description of projectively equivariant linear layers, with new proposed architectures, and a completely new accompanying experiment section, as a result