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

Philipp Gruening, Erhardt Barth

Keywords: deep networks, hypercomplex cells, end-stopping, hyperselectivity

TL;DR: We insert Min-units that directly model end-stopped behavior and hyperselectivity into state-of-the art CNNs, and show that the resulting Min-Nets perform better on the Cifar-10 benchmark.

Abstract: 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.

Community Implementations: [ 3 code implementations](https://www.catalyzex.com/paper/bio-inspired-min-nets-improve-the-performance/code)