QL-Net: Quantized-by-LookUp CNN

Kamila Abdiyeva, Kim Hui Yap, Wang Gang, Narendra Ahuja, Martin Lukac

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Convolutional Neural Networks (CNNs) have achieved a state-of-the-art performance in the different computer vision tasks. However, CNN algorithms are computationally and power intensive, which makes them difficult to run on wearable and embedded systems. One way to address this constraint is to reduce the number of computational operations performed. Recently, several approaches addressed the problem of the computational complexity in the CNNs. Most of these methods, however, require a dedicated hardware. We propose a new method for the computation reduction in CNNs that substitutes Multiply and Accumulate (MAC) operations with a codebook lookup and can be executed on the generic hardware. The proposed method called QL-Net combines several concepts: (i) a codebook construction, (ii) a layer-wise retraining strategy, and (iii) a substitution of the MAC operations with the lookup of the convolution responses at inference time. The proposed QL-Net achieves a 98.6% accuracy on the MNIST dataset with a 5.8x reduction in runtime, when compared to MAC-based CNN model that achieved a 99.2% accuracy.

Original languageEnglish
Title of host publication2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages413-418
Number of pages6
ISBN (Electronic)9781538695821
DOIs
Publication statusPublished - Dec 18 2018
Event15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018 - Singapore, Singapore
Duration: Nov 18 2018Nov 21 2018

Other

Other15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018
CountrySingapore
CitySingapore
Period11/18/1811/21/18

Fingerprint

Accumulate
Neural Networks
Neural networks
Multiplication
Codebook
Hardware
Network Algorithms
Substitute
Neural Network Model
Embedded Systems
Computer Vision
Substitution
Convolution
Computational Complexity
Embedded systems
Computer vision
Computational complexity
Substitution reactions

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Vision and Pattern Recognition
  • Control and Optimization

Cite this

Abdiyeva, K., Yap, K. H., Gang, W., Ahuja, N., & Lukac, M. (2018). QL-Net: Quantized-by-LookUp CNN. In 2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018 (pp. 413-418). [8581119] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICARCV.2018.8581119

QL-Net : Quantized-by-LookUp CNN. / Abdiyeva, Kamila; Yap, Kim Hui; Gang, Wang; Ahuja, Narendra; Lukac, Martin.

2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018. Institute of Electrical and Electronics Engineers Inc., 2018. p. 413-418 8581119.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abdiyeva, K, Yap, KH, Gang, W, Ahuja, N & Lukac, M 2018, QL-Net: Quantized-by-LookUp CNN. in 2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018., 8581119, Institute of Electrical and Electronics Engineers Inc., pp. 413-418, 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018, Singapore, Singapore, 11/18/18. https://doi.org/10.1109/ICARCV.2018.8581119
Abdiyeva K, Yap KH, Gang W, Ahuja N, Lukac M. QL-Net: Quantized-by-LookUp CNN. In 2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018. Institute of Electrical and Electronics Engineers Inc. 2018. p. 413-418. 8581119 https://doi.org/10.1109/ICARCV.2018.8581119
Abdiyeva, Kamila ; Yap, Kim Hui ; Gang, Wang ; Ahuja, Narendra ; Lukac, Martin. / QL-Net : Quantized-by-LookUp CNN. 2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018. Institute of Electrical and Electronics Engineers Inc., 2018. pp. 413-418
@inproceedings{63012039785144e4ba722a88d12713b4,
title = "QL-Net: Quantized-by-LookUp CNN",
abstract = "Convolutional Neural Networks (CNNs) have achieved a state-of-the-art performance in the different computer vision tasks. However, CNN algorithms are computationally and power intensive, which makes them difficult to run on wearable and embedded systems. One way to address this constraint is to reduce the number of computational operations performed. Recently, several approaches addressed the problem of the computational complexity in the CNNs. Most of these methods, however, require a dedicated hardware. We propose a new method for the computation reduction in CNNs that substitutes Multiply and Accumulate (MAC) operations with a codebook lookup and can be executed on the generic hardware. The proposed method called QL-Net combines several concepts: (i) a codebook construction, (ii) a layer-wise retraining strategy, and (iii) a substitution of the MAC operations with the lookup of the convolution responses at inference time. The proposed QL-Net achieves a 98.6{\%} accuracy on the MNIST dataset with a 5.8x reduction in runtime, when compared to MAC-based CNN model that achieved a 99.2{\%} accuracy.",
author = "Kamila Abdiyeva and Yap, {Kim Hui} and Wang Gang and Narendra Ahuja and Martin Lukac",
year = "2018",
month = "12",
day = "18",
doi = "10.1109/ICARCV.2018.8581119",
language = "English",
pages = "413--418",
booktitle = "2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
address = "United States",

}

TY - GEN

T1 - QL-Net

T2 - Quantized-by-LookUp CNN

AU - Abdiyeva, Kamila

AU - Yap, Kim Hui

AU - Gang, Wang

AU - Ahuja, Narendra

AU - Lukac, Martin

PY - 2018/12/18

Y1 - 2018/12/18

N2 - Convolutional Neural Networks (CNNs) have achieved a state-of-the-art performance in the different computer vision tasks. However, CNN algorithms are computationally and power intensive, which makes them difficult to run on wearable and embedded systems. One way to address this constraint is to reduce the number of computational operations performed. Recently, several approaches addressed the problem of the computational complexity in the CNNs. Most of these methods, however, require a dedicated hardware. We propose a new method for the computation reduction in CNNs that substitutes Multiply and Accumulate (MAC) operations with a codebook lookup and can be executed on the generic hardware. The proposed method called QL-Net combines several concepts: (i) a codebook construction, (ii) a layer-wise retraining strategy, and (iii) a substitution of the MAC operations with the lookup of the convolution responses at inference time. The proposed QL-Net achieves a 98.6% accuracy on the MNIST dataset with a 5.8x reduction in runtime, when compared to MAC-based CNN model that achieved a 99.2% accuracy.

AB - Convolutional Neural Networks (CNNs) have achieved a state-of-the-art performance in the different computer vision tasks. However, CNN algorithms are computationally and power intensive, which makes them difficult to run on wearable and embedded systems. One way to address this constraint is to reduce the number of computational operations performed. Recently, several approaches addressed the problem of the computational complexity in the CNNs. Most of these methods, however, require a dedicated hardware. We propose a new method for the computation reduction in CNNs that substitutes Multiply and Accumulate (MAC) operations with a codebook lookup and can be executed on the generic hardware. The proposed method called QL-Net combines several concepts: (i) a codebook construction, (ii) a layer-wise retraining strategy, and (iii) a substitution of the MAC operations with the lookup of the convolution responses at inference time. The proposed QL-Net achieves a 98.6% accuracy on the MNIST dataset with a 5.8x reduction in runtime, when compared to MAC-based CNN model that achieved a 99.2% accuracy.

UR - http://www.scopus.com/inward/record.url?scp=85060781423&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85060781423&partnerID=8YFLogxK

U2 - 10.1109/ICARCV.2018.8581119

DO - 10.1109/ICARCV.2018.8581119

M3 - Conference contribution

AN - SCOPUS:85060781423

SP - 413

EP - 418

BT - 2018 15th International Conference on Control, Automation, Robotics and Vision, ICARCV 2018

PB - Institute of Electrical and Electronics Engineers Inc.

ER -