Transplant of Perceptrons
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
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.
Keywords: perceptron, transplant, neural network
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We propose to transplant active cells into inactive cells in neural networks to improve the performance of the model.
Abstract: We propose to *transplant active cells into inactive cells* in neural networks, inspired by the concept of ``transplant'' in the field of neuroscience, where dead neurons are replaced with live ones to improve brain functions.
This is motivated by the fact that a number of major machine learning methodologies such as the perceptron and convolutional neural networks have been invented via the collaboration between neurobiology and computer science.
We theoretically discuss how transplant improves the quality of representation of perceptron layers in terms of the mutual information and the loss function with respect to the performance of the whole network. Moreover, we empirically evaluate the effectiveness of transplant in the task of supervised classification. Our proposal is simple and applicable to any neural networks which contain at least one perceptron layer.
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: 1649
Loading