Light Differentiable Logic Gate Networks

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Rüttgers, Lukas, Aczel, Till, Plesner, Andreas, Wattenhofer, Roger
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914072875761664
author Rüttgers, Lukas
Aczel, Till
Plesner, Andreas
Wattenhofer, Roger
author_facet Rüttgers, Lukas
Aczel, Till
Plesner, Andreas
Wattenhofer, Roger
contents Differentiable logic gate networks (DLGNs) exhibit extraordinary efficiency at inference while sustaining competitive accuracy. But vanishing gradients, discretization errors, and high training cost impede scaling these networks. Even with dedicated parameter initialization schemes from subsequent works, increasing depth still harms accuracy. We show that the root cause of these issues lies in the underlying parametrization of logic gate neurons themselves. To overcome this issue, we propose a reparametrization that also shrinks the parameter size logarithmically in the number of inputs per gate. For binary inputs, this already reduces the model size by 4x, speeds up the backward pass by up to 1.86x, and converges in 8.5x fewer training steps. On top of that, we show that the accuracy on CIFAR-100 remains stable and sometimes superior to the original parametrization.
format Preprint
id arxiv_https___arxiv_org_abs_2510_03250
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Light Differentiable Logic Gate Networks
Rüttgers, Lukas
Aczel, Till
Plesner, Andreas
Wattenhofer, Roger
Machine Learning
Performance
Differentiable logic gate networks (DLGNs) exhibit extraordinary efficiency at inference while sustaining competitive accuracy. But vanishing gradients, discretization errors, and high training cost impede scaling these networks. Even with dedicated parameter initialization schemes from subsequent works, increasing depth still harms accuracy. We show that the root cause of these issues lies in the underlying parametrization of logic gate neurons themselves. To overcome this issue, we propose a reparametrization that also shrinks the parameter size logarithmically in the number of inputs per gate. For binary inputs, this already reduces the model size by 4x, speeds up the backward pass by up to 1.86x, and converges in 8.5x fewer training steps. On top of that, we show that the accuracy on CIFAR-100 remains stable and sometimes superior to the original parametrization.
title Light Differentiable Logic Gate Networks
topic Machine Learning
Performance
url https://arxiv.org/abs/2510.03250