---
DOAJ_listed: '1'
OA_place: publisher
OA_type: gold
_id: '18234'
abstract:
- lang: eng
  text: Convolutional Neural Networks (CNNs) are very popular in many fields including
    computer vision, speech recognition, natural language processing, etc. Though
    deep learning leads to groundbreaking performance in those domains, the networks
    used are very computationally demanding and are far from being able to perform
    in real-time applications even on a GPU, which is not power efficient and therefore
    does not suit low power systems such as mobile devices. To overcome this challenge,
    some solutions have been proposed for quantizing the weights and activations of
    these networks, which accelerate the runtime significantly. Yet, this acceleration
    comes at the cost of a larger error unless spatial adjustments are carried out.
    The method proposed in this work trains quantized neural networks by noise injection
    and a learned clamping, which improve accuracy. This leads to state-of-the-art
    results on various regression and classification tasks, e.g., ImageNet classification
    with architectures such as ResNet-18/34/50 with as low as 3 bit weights and activations.
    We implement the proposed solution on an FPGA to demonstrate its applicability
    for low-power real-time applications. The quantization code will become publicly
    available upon acceptance.
article_number: '2144'
article_processing_charge: No
article_type: original
arxiv: 1
author:
- first_name: Chaim
  full_name: Baskin, Chaim
  last_name: Baskin
- first_name: Evgenii
  full_name: Zheltonozhkii, Evgenii
  last_name: Zheltonozhkii
- first_name: Tal
  full_name: Rozen, Tal
  last_name: Rozen
- first_name: Natan
  full_name: Liss, Natan
  last_name: Liss
- first_name: Yoav
  full_name: Chai, Yoav
  last_name: Chai
- first_name: Eli
  full_name: Schwartz, Eli
  last_name: Schwartz
- first_name: Raja
  full_name: Giryes, Raja
  last_name: Giryes
- first_name: Alexander
  full_name: Bronstein, Alexander
  id: 58f3726e-7cba-11ef-ad8b-e6e8cb3904e6
  last_name: Bronstein
  orcid: 0000-0001-9699-8730
- first_name: Avi
  full_name: Mendelson, Avi
  last_name: Mendelson
citation:
  ama: 'Baskin C, Zheltonozhkii E, Rozen T, et al. NICE: Noise Injection and Clamping
    Estimation for neural network quantization. <i>Mathematics</i>. 2021;9(17). doi:<a
    href="https://doi.org/10.3390/math9172144">10.3390/math9172144</a>'
  apa: 'Baskin, C., Zheltonozhkii, E., Rozen, T., Liss, N., Chai, Y., Schwartz, E.,
    … Mendelson, A. (2021). NICE: Noise Injection and Clamping Estimation for neural
    network quantization. <i>Mathematics</i>. MDPI. <a href="https://doi.org/10.3390/math9172144">https://doi.org/10.3390/math9172144</a>'
  chicago: 'Baskin, Chaim, Evgenii Zheltonozhkii, Tal Rozen, Natan Liss, Yoav Chai,
    Eli Schwartz, Raja Giryes, Alex M. Bronstein, and Avi Mendelson. “NICE: Noise
    Injection and Clamping Estimation for Neural Network Quantization.” <i>Mathematics</i>.
    MDPI, 2021. <a href="https://doi.org/10.3390/math9172144">https://doi.org/10.3390/math9172144</a>.'
  ieee: 'C. Baskin <i>et al.</i>, “NICE: Noise Injection and Clamping Estimation for
    neural network quantization,” <i>Mathematics</i>, vol. 9, no. 17. MDPI, 2021.'
  ista: 'Baskin C, Zheltonozhkii E, Rozen T, Liss N, Chai Y, Schwartz E, Giryes R,
    Bronstein AM, Mendelson A. 2021. NICE: Noise Injection and Clamping Estimation
    for neural network quantization. Mathematics. 9(17), 2144.'
  mla: 'Baskin, Chaim, et al. “NICE: Noise Injection and Clamping Estimation for Neural
    Network Quantization.” <i>Mathematics</i>, vol. 9, no. 17, 2144, MDPI, 2021, doi:<a
    href="https://doi.org/10.3390/math9172144">10.3390/math9172144</a>.'
  short: C. Baskin, E. Zheltonozhkii, T. Rozen, N. Liss, Y. Chai, E. Schwartz, R.
    Giryes, A.M. Bronstein, A. Mendelson, Mathematics 9 (2021).
date_created: 2024-10-08T12:57:24Z
date_published: 2021-09-02T00:00:00Z
date_updated: 2024-10-15T07:37:39Z
day: '02'
doi: 10.3390/math9172144
extern: '1'
external_id:
  arxiv:
  - '1810.00162'
fulldoi: https://doi.org/10.3390/math9172144
intvolume: '         9'
issue: '17'
language:
- iso: eng
month: '09'
oa_version: Published Version
publication: Mathematics
publication_identifier:
  issn:
  - 2227-7390
publication_status: published
publisher: MDPI
quality_controlled: '1'
scopus_import: '1'
status: public
title: 'NICE: Noise Injection and Clamping Estimation for neural network quantization'
type: journal_article
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
volume: 9
year: '2021'
...