arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2201.02149cs.CV
收藏

仿生最小网提高深度网络的性能和鲁棒性

Bio-inspired Min-Nets Improve the Performance and Robustness of Deep Networks

University of Lübeck(吕贝克大学)

Philipp Grüning, Erhardt Barth

摘要:

Min-Nets的灵感来自于末端停止的皮质细胞,其单元输出两个学习过滤器中的最小值。我们将这样的Min-units插入到最先进的深度网络中,例如流行的ResNet和DenseNet,并表明由此产生的Min-Nets在Cifar-10基准测试中表现更好。此外,我们证明了Min-Nets对JPEG压缩伪影的鲁棒性更强。我们认为,最小的操作是最简单的方式来实现一个与操作对过滤器和这样的与操作引入的偏见,是适当的自然图像的统计。

英文摘要:

Min-Nets are inspired by end-stopped cortical cells with units that output the minimum of two learned filters. We insert such Min-units into state-of-the-art deep networks, such as the popular ResNet and DenseNet, and show that the resulting Min-Nets perform better on the Cifar-10 benchmark. Moreover, we show that Min-Nets are more robust against JPEG compression artifacts. We argue that the minimum operation is the simplest way of implementing an AND operation on pairs of filters and that such AND operations introduce a bias that is appropriate given the statistics of natural images.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑