Efficient Pre-Designed Convolutional Front-End for Deep Learning

Hamza Baali; Abdesselam Bouzerdoum

doi:10.1109/MLSP.2019.8918881

Efficient Pre-Designed Convolutional Front-End for Deep Learning

Hamza Baali, Abdesselam Bouzerdoum

HBKU College of Science and Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

This paper introduces a hierarchical learning paradigm based on a predesigned directional filter bank front-end analogous to the energy model for complex cells. The filter bank front-end is designed to extract common primitive features such as orientations and edges. Each energy response is subjected to a shunting inhibition operator to enhance contrast and reduce the effects of illumination variations. This is followed by a divisive-normalization, which bounds the responses of the feature maps. The normalized responses are then propagated through a two-layer convolutional neural network (CNN) back-end for classification. The efficiency of the proposed approach is demonstrated using the CIFAR-10 dataset, and its performance is compared against that of the DTCWT ScaterNet front-end.

Original language	English
Title of host publication	2019 IEEE 29th International Workshop on Machine Learning for Signal Processing, MLSP 2019
Publisher	IEEE Computer Society
ISBN (Electronic)	9781728108247
DOIs	https://doi.org/10.1109/MLSP.2019.8918881
Publication status	Published - Oct 2019
Event	29th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2019 - Pittsburgh, United States Duration: 13 Oct 2019 → 16 Oct 2019

Publication series

Name	IEEE International Workshop on Machine Learning for Signal Processing, MLSP
Volume	2019-October
ISSN (Print)	2161-0363
ISSN (Electronic)	2161-0371

Conference

Conference	29th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2019
Country/Territory	United States
City	Pittsburgh
Period	13/10/19 → 16/10/19

Keywords

Convolutional neural network
Energy model
directional filters

Access to Document

10.1109/MLSP.2019.8918881

Cite this

Baali, H., & Bouzerdoum, A. (2019). Efficient Pre-Designed Convolutional Front-End for Deep Learning. In 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing, MLSP 2019 Article 8918881 (IEEE International Workshop on Machine Learning for Signal Processing, MLSP; Vol. 2019-October). IEEE Computer Society. https://doi.org/10.1109/MLSP.2019.8918881

@inproceedings{76657f59e2d84ad2a6030d54034773c0,

title = "Efficient Pre-Designed Convolutional Front-End for Deep Learning",

abstract = "This paper introduces a hierarchical learning paradigm based on a predesigned directional filter bank front-end analogous to the energy model for complex cells. The filter bank front-end is designed to extract common primitive features such as orientations and edges. Each energy response is subjected to a shunting inhibition operator to enhance contrast and reduce the effects of illumination variations. This is followed by a divisive-normalization, which bounds the responses of the feature maps. The normalized responses are then propagated through a two-layer convolutional neural network (CNN) back-end for classification. The efficiency of the proposed approach is demonstrated using the CIFAR-10 dataset, and its performance is compared against that of the DTCWT ScaterNet front-end.",

keywords = "Convolutional neural network, Energy model, directional filters",

author = "Hamza Baali and Abdesselam Bouzerdoum",

note = "Publisher Copyright: {\textcopyright} 2019 IEEE.; 29th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2019 ; Conference date: 13-10-2019 Through 16-10-2019",

year = "2019",

month = oct,

doi = "10.1109/MLSP.2019.8918881",

language = "English",

series = "IEEE International Workshop on Machine Learning for Signal Processing, MLSP",

publisher = "IEEE Computer Society",

booktitle = "2019 IEEE 29th International Workshop on Machine Learning for Signal Processing, MLSP 2019",

address = "United States",

}

Baali, H & Bouzerdoum, A 2019, Efficient Pre-Designed Convolutional Front-End for Deep Learning. in 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing, MLSP 2019., 8918881, IEEE International Workshop on Machine Learning for Signal Processing, MLSP, vol. 2019-October, IEEE Computer Society, 29th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2019, Pittsburgh, United States, 13/10/19. https://doi.org/10.1109/MLSP.2019.8918881

Efficient Pre-Designed Convolutional Front-End for Deep Learning. / Baali, Hamza; Bouzerdoum, Abdesselam.
2019 IEEE 29th International Workshop on Machine Learning for Signal Processing, MLSP 2019. IEEE Computer Society, 2019. 8918881 (IEEE International Workshop on Machine Learning for Signal Processing, MLSP; Vol. 2019-October).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Efficient Pre-Designed Convolutional Front-End for Deep Learning

AU - Baali, Hamza

AU - Bouzerdoum, Abdesselam

PY - 2019/10

Y1 - 2019/10

N2 - This paper introduces a hierarchical learning paradigm based on a predesigned directional filter bank front-end analogous to the energy model for complex cells. The filter bank front-end is designed to extract common primitive features such as orientations and edges. Each energy response is subjected to a shunting inhibition operator to enhance contrast and reduce the effects of illumination variations. This is followed by a divisive-normalization, which bounds the responses of the feature maps. The normalized responses are then propagated through a two-layer convolutional neural network (CNN) back-end for classification. The efficiency of the proposed approach is demonstrated using the CIFAR-10 dataset, and its performance is compared against that of the DTCWT ScaterNet front-end.

AB - This paper introduces a hierarchical learning paradigm based on a predesigned directional filter bank front-end analogous to the energy model for complex cells. The filter bank front-end is designed to extract common primitive features such as orientations and edges. Each energy response is subjected to a shunting inhibition operator to enhance contrast and reduce the effects of illumination variations. This is followed by a divisive-normalization, which bounds the responses of the feature maps. The normalized responses are then propagated through a two-layer convolutional neural network (CNN) back-end for classification. The efficiency of the proposed approach is demonstrated using the CIFAR-10 dataset, and its performance is compared against that of the DTCWT ScaterNet front-end.

KW - Convolutional neural network

KW - Energy model

KW - directional filters

UR - http://www.scopus.com/inward/record.url?scp=85077709737&partnerID=8YFLogxK

U2 - 10.1109/MLSP.2019.8918881

DO - 10.1109/MLSP.2019.8918881

M3 - Conference contribution

AN - SCOPUS:85077709737

T3 - IEEE International Workshop on Machine Learning for Signal Processing, MLSP

BT - 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing, MLSP 2019

PB - IEEE Computer Society

T2 - 29th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2019

Y2 - 13 October 2019 through 16 October 2019

ER -

Efficient Pre-Designed Convolutional Front-End for Deep Learning

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this