MOGNET: A Mux-residual quantized Network leveraging Online-Generated weights

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Nguyen, Van Thien, Guicquero, William, Sicard, Gilles
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866909457897750528
author Nguyen, Van Thien
Guicquero, William
Sicard, Gilles
author_facet Nguyen, Van Thien
Guicquero, William
Sicard, Gilles
contents This paper presents a compact model architecture called MOGNET, compatible with a resource-limited hardware. MOGNET uses a streamlined Convolutional factorization block based on a combination of 2 point-wise (1x1) convolutions with a group-wise convolution in-between. To further limit the overall model size and reduce the on-chip required memory, the second point-wise convolution's parameters are on-line generated by a Cellular Automaton structure. In addition, MOGNET enables the use of low-precision weights and activations, by taking advantage of a Multiplexer mechanism with a proper Bitshift rescaling for integrating residual paths without increasing the hardware-related complexity. To efficiently train this model we also introduce a novel weight ternarization method favoring the balance between quantized levels. Experimental results show that given tiny memory budget (sub-2Mb), MOGNET can achieve higher accuracy with a clear gap up to 1% at a similar or even lower model size compared to recent state-of-the-art methods.
format Preprint
id arxiv_https___arxiv_org_abs_2501_09531
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MOGNET: A Mux-residual quantized Network leveraging Online-Generated weights
Nguyen, Van Thien
Guicquero, William
Sicard, Gilles
Machine Learning
Hardware Architecture
This paper presents a compact model architecture called MOGNET, compatible with a resource-limited hardware. MOGNET uses a streamlined Convolutional factorization block based on a combination of 2 point-wise (1x1) convolutions with a group-wise convolution in-between. To further limit the overall model size and reduce the on-chip required memory, the second point-wise convolution's parameters are on-line generated by a Cellular Automaton structure. In addition, MOGNET enables the use of low-precision weights and activations, by taking advantage of a Multiplexer mechanism with a proper Bitshift rescaling for integrating residual paths without increasing the hardware-related complexity. To efficiently train this model we also introduce a novel weight ternarization method favoring the balance between quantized levels. Experimental results show that given tiny memory budget (sub-2Mb), MOGNET can achieve higher accuracy with a clear gap up to 1% at a similar or even lower model size compared to recent state-of-the-art methods.
title MOGNET: A Mux-residual quantized Network leveraging Online-Generated weights
topic Machine Learning
Hardware Architecture
url https://arxiv.org/abs/2501.09531