Saved in:
Bibliographic Details
Main Authors: Santini, Gabriele, Paissan, Francesco, Farella, Elisabetta
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.10689
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918022057295872
author Santini, Gabriele
Paissan, Francesco
Farella, Elisabetta
author_facet Santini, Gabriele
Paissan, Francesco
Farella, Elisabetta
contents We propose a probabilistic framework for dynamic quantization of neural networks that allows for a computationally efficient input-adaptive rescaling of the quantization parameters. Our framework applies a probabilistic model to the network's pre-activations through a lightweight surrogate, enabling the adaptive adjustment of the quantization parameters on a per-input basis without significant memory overhead. We validate our approach on a set of popular computer vision tasks and models, observing only a negligible loss in performance. Our method strikes the best performance and computational overhead tradeoff compared to standard quantization strategies.
format Preprint
id arxiv_https___arxiv_org_abs_2505_10689
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A probabilistic framework for dynamic quantization
Santini, Gabriele
Paissan, Francesco
Farella, Elisabetta
Machine Learning
Computer Vision and Pattern Recognition
We propose a probabilistic framework for dynamic quantization of neural networks that allows for a computationally efficient input-adaptive rescaling of the quantization parameters. Our framework applies a probabilistic model to the network's pre-activations through a lightweight surrogate, enabling the adaptive adjustment of the quantization parameters on a per-input basis without significant memory overhead. We validate our approach on a set of popular computer vision tasks and models, observing only a negligible loss in performance. Our method strikes the best performance and computational overhead tradeoff compared to standard quantization strategies.
title A probabilistic framework for dynamic quantization
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2505.10689