HENP: Dynamic Pruning via Neuron Entropy
Keywords: Pruning, Network Compression, Dying Neurons, Network Architecture Search
Abstract: We introduce a novel framework for analyzing neural networks based on the concepts of \textit{dynamic} and \textit{static} neurons, which describe the stability of neuron activation under specific inputs. From these concepts, we propose \textit{neuron entropy} as a metric to quantify network expressiveness. Our analysis reveals that better generalization correlates with diverse activation patterns and higher neuron entropy. Building on this, we propose our HENP method, a dynamic pruning technique that regulates dying neurons and sparsifies the network during training. Experimental results demonstrate that our HENP improves both network sparsity and performance, offering a new approach to efficient neural network optimization.
Primary Area: other topics in machine learning (i.e., none of the above)
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: 460
Loading