Selectivity considered harmful: evaluating the causal impact of class selectivity in DNNs

TL;DR

This paper investigates the causal role of class selectivity in deep neural networks, finding that reducing selectivity can improve or not harm performance, challenging its assumed importance.

Contribution

It provides causal evidence that class selectivity is neither necessary nor sufficient for DNN performance, and shows regularizing against it can enhance accuracy.

Findings

Reducing class selectivity increased test accuracy by over 2% in ResNet18.

Lowering class selectivity had minimal impact on CIFAR10 accuracy in ResNet20.

Increasing class selectivity decreased accuracy across models and datasets.

Abstract

The properties of individual neurons are often analyzed in order to understand the biological and artificial neural networks in which they're embedded. Class selectivity-typically defined as how different a neuron's responses are across different classes of stimuli or data samples-is commonly used for this purpose. However, it remains an open question whether it is necessary and/or sufficient for deep neural networks (DNNs) to learn class selectivity in individual units. We investigated the causal impact of class selectivity on network function by directly regularizing for or against class selectivity. Using this regularizer to reduce class selectivity across units in convolutional neural networks increased test accuracy by over 2% for ResNet18 trained on Tiny ImageNet. For ResNet20 trained on CIFAR10 we could reduce class selectivity by a factor of 2.5 with no impact on test accuracy, and reduce it nearly to zero with only a small ($\sim$2%) drop in test accuracy. In contrast, regularizing to increase class selectivity significantly decreased test accuracy across all models and datasets. These results indicate that class selectivity in individual units is neither sufficient nor strictly necessary, and can even impair DNN performance. They also encourage caution when focusing on the properties of single units as representative of the mechanisms by which DNNs function.