And the Bit Goes Down: Revisiting the Quantization of Neural Networks - Archive ouverte HAL Access content directly
Conference Papers Year : 2020

And the Bit Goes Down: Revisiting the Quantization of Neural Networks

Abstract

In this paper, we address the problem of reducing the memory footprint of convolutional network architectures. We introduce a vector quantization method that aims at preserving the quality of the reconstruction of the network outputs rather than its weights. The principle of our approach is that it minimizes the loss reconstruction error for in-domain inputs. Our method only requires a set of unlabelled data at quantization time and allows for efficient inference on CPU by using bytealigned codebooks to store the compressed weights. We validate our approach by quantizing a high performing ResNet-50 model to a memory size of 5 MB (20 X compression factor) while preserving a top-1 accuracy of 76:1% on ImageNet object classification and by compressing a Mask R-CNN with a 26 X factor.1
Fichier principal
Vignette du fichier
1907.05686 (1).pdf (1.51 Mo) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-02434572 , version 1 (10-01-2020)

Identifiers

  • HAL Id : hal-02434572 , version 1

Cite

Pierre Stock, Armand Joulin, Rémi Gribonval, Benjamin Graham, Hervé Jégou. And the Bit Goes Down: Revisiting the Quantization of Neural Networks. ICLR 2020 - Eighth International Conference on Learning Representations, Apr 2020, Addis-Abeba, Ethiopia. pp.1-11. ⟨hal-02434572⟩
241 View
733 Download

Share

Gmail Facebook X LinkedIn More