Reassessing Number-Detector Units in Convolutional Neural Networks
Keywords: Feature Selection, Pruning, RSA, CNN, Numerosity
TL;DR: Pruning CNNs indicates that number-detector units are not crucial for representing numerosity at the population level, thereby challenging their suggested significance in visual tasks.
Abstract: Convolutional neural networks (CNNs), including CORnet, have become essential models for predicting neural activity and behavior in visual tasks. However, their ability to capture complex cognitive functions, such as numerosity discrimination, remains a topic of debate. Numerosity—the ability to perceive and estimate the number of items in a visual scene—is believed to be represented by specialized 'number-detector' units within CNNs. In this study, we utilize CORnet, a specialized type of CNN inspired by brain anatomy, which also effectively captures the variance in human behavioral data, to address the limitations of classical representational similarity analysis (RSA), which assumes equal importance for all features. We apply pruning, a feature selection technique that identifies and retains the most behaviorally relevant units. Our results demonstrate that number-detector units are not critical for population-level representations of numerosity, challenging their proposed significance in previous studies. These results can have implications for both machine learning and neuroscience.
Primary Area: applications to neuroscience & cognitive science
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9862
Loading