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

arXiv: https://arxiv.org/abs/2003.01262 · HTML: https://mlsanity.com/paper/2003.01262/

## Key facts

- Cited by 15 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 3 Mar 2020.

## Built on by (most cited first, 15 of 15)

- [Sudden Drops in the Loss: Syntax Acquisition, Phase Transitions, and Simplicity Bias in MLMs](https://mlsanity.com/paper/2309.07311/) (2023)
- [Scientific Inference With Interpretable Machine Learning: Analyzing Models to Learn About Real-World Phenomena](https://mlsanity.com/paper/2206.05487/) (2022)
- [Trade-offs of Local SGD at Scale: An Empirical Study](https://mlsanity.com/paper/2110.08133/) (2021)
- [Modeling Category-Selective Cortical Regions with Topographic Variational Autoencoders](https://mlsanity.com/paper/2110.13911/) (2021)
- [Intraclass clustering: an implicit learning ability that regularizes DNNs](https://mlsanity.com/paper/2103.06733/) (2021)
- [Deeply Explain CNN via Hierarchical Decomposition](https://mlsanity.com/paper/2201.09205/) (2022)
- [On the special role of class-selective neurons in early training](https://mlsanity.com/paper/2305.17409/) (2023)
- [Artificial Neural Nets and the Representation of Human Concepts](https://mlsanity.com/paper/2312.05337/) (2023)
- [Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics](https://mlsanity.com/paper/2606.06533/) (2026)
- [Estimating the Generalization in Deep Neural Networks via Sparsity](https://mlsanity.com/paper/2104.00851/) (2021)
- [Large Learning Rates Simultaneously Achieve Robustness to Spurious Correlations and Compressibility](https://mlsanity.com/paper/2507.17748/) (2025)
- [Higher-order mutual information reveals synergistic sub-networks for multi-neuron importance](https://mlsanity.com/paper/2211.00416/) (2022)
- [Towards understanding deep learning with the natural clustering prior](https://mlsanity.com/paper/2203.08174/) (2022)
- [Much Easier Said Than Done: Falsifying the Causal Relevance of Linear Decoding Methods](https://mlsanity.com/paper/2211.04367/) (2022)
- [Linking in Style: Understanding learned features in deep learning models](https://mlsanity.com/paper/2409.16865/) (2024)

## Abstract (arXiv)

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 (~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.