Deep graph regularized learning for binary classification

Research output: Contribution to conferencePaper

Abstract

With growing interest in data-driven classification, deep learning is now prevalent thanks to its ability to learn feature mapping functions solely from data. For very small training sets, however, deep learning, even with traditional regularization techniques, often overfits, resulting in sub-par classification performance. In this paper, we propose a novel binary classifier deep learning method, based on an iterative quadratic programming (QP) formulation with a graph Laplacian regularizer (GLR), combining the merits of model-based and data-driven approaches. Specifically, the proposed network employs a convolutional neural network (CNN) to learn deep features, which are used to define edge weights for a graph to pose a convex QP problem. Further, we design a novel loss function to penalize samples at the class boundary during semi-supervised learning. Results demonstrate that given a small-size training dataset, our network outperforms several state-of-the-art classifiers, including CNN, model-based GLR, and dynamic graph CNN classifiers.

Conference

Conference2019 International Conference on Acoustics, Speech, and Signal Processing
Abbreviated titleICASSP 2019
CountryUnited Kingdom
CityBrighton
Period12/05/1917/05/19

Fingerprint

Classifiers
Quadratic programming
Neural networks
Supervised learning
Deep learning

Keywords

  • graph Laplacian regularization
  • binary classification
  • semi-supervised learning,
  • deep learning

Cite this

Ye, M., Stankovic, V., Stankovic, L., & Cheung, G. (2019). Deep graph regularized learning for binary classification. Paper presented at 2019 International Conference on Acoustics, Speech, and Signal Processing, Brighton, United Kingdom.
Ye, Minxiang ; Stankovic, Vladimir ; Stankovic, Lina ; Cheung, Gene. / Deep graph regularized learning for binary classification. Paper presented at 2019 International Conference on Acoustics, Speech, and Signal Processing, Brighton, United Kingdom.5 p.
@conference{a115f406ea2c40f5bafd77b5568b0541,
title = "Deep graph regularized learning for binary classification",
abstract = "With growing interest in data-driven classification, deep learning is now prevalent thanks to its ability to learn feature mapping functions solely from data. For very small training sets, however, deep learning, even with traditional regularization techniques, often overfits, resulting in sub-par classification performance. In this paper, we propose a novel binary classifier deep learning method, based on an iterative quadratic programming (QP) formulation with a graph Laplacian regularizer (GLR), combining the merits of model-based and data-driven approaches. Specifically, the proposed network employs a convolutional neural network (CNN) to learn deep features, which are used to define edge weights for a graph to pose a convex QP problem. Further, we design a novel loss function to penalize samples at the class boundary during semi-supervised learning. Results demonstrate that given a small-size training dataset, our network outperforms several state-of-the-art classifiers, including CNN, model-based GLR, and dynamic graph CNN classifiers.",
keywords = "graph Laplacian regularization, binary classification, semi-supervised learning,, deep learning",
author = "Minxiang Ye and Vladimir Stankovic and Lina Stankovic and Gene Cheung",
year = "2019",
month = "5",
day = "12",
language = "English",
note = "2019 International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2019 ; Conference date: 12-05-2019 Through 17-05-2019",

}

Ye, M, Stankovic, V, Stankovic, L & Cheung, G 2019, 'Deep graph regularized learning for binary classification' Paper presented at 2019 International Conference on Acoustics, Speech, and Signal Processing, Brighton, United Kingdom, 12/05/19 - 17/05/19, .

Deep graph regularized learning for binary classification. / Ye, Minxiang; Stankovic, Vladimir; Stankovic, Lina; Cheung, Gene.

2019. Paper presented at 2019 International Conference on Acoustics, Speech, and Signal Processing, Brighton, United Kingdom.

Research output: Contribution to conferencePaper

TY - CONF

T1 - Deep graph regularized learning for binary classification

AU - Ye, Minxiang

AU - Stankovic, Vladimir

AU - Stankovic, Lina

AU - Cheung, Gene

PY - 2019/5/12

Y1 - 2019/5/12

N2 - With growing interest in data-driven classification, deep learning is now prevalent thanks to its ability to learn feature mapping functions solely from data. For very small training sets, however, deep learning, even with traditional regularization techniques, often overfits, resulting in sub-par classification performance. In this paper, we propose a novel binary classifier deep learning method, based on an iterative quadratic programming (QP) formulation with a graph Laplacian regularizer (GLR), combining the merits of model-based and data-driven approaches. Specifically, the proposed network employs a convolutional neural network (CNN) to learn deep features, which are used to define edge weights for a graph to pose a convex QP problem. Further, we design a novel loss function to penalize samples at the class boundary during semi-supervised learning. Results demonstrate that given a small-size training dataset, our network outperforms several state-of-the-art classifiers, including CNN, model-based GLR, and dynamic graph CNN classifiers.

AB - With growing interest in data-driven classification, deep learning is now prevalent thanks to its ability to learn feature mapping functions solely from data. For very small training sets, however, deep learning, even with traditional regularization techniques, often overfits, resulting in sub-par classification performance. In this paper, we propose a novel binary classifier deep learning method, based on an iterative quadratic programming (QP) formulation with a graph Laplacian regularizer (GLR), combining the merits of model-based and data-driven approaches. Specifically, the proposed network employs a convolutional neural network (CNN) to learn deep features, which are used to define edge weights for a graph to pose a convex QP problem. Further, we design a novel loss function to penalize samples at the class boundary during semi-supervised learning. Results demonstrate that given a small-size training dataset, our network outperforms several state-of-the-art classifiers, including CNN, model-based GLR, and dynamic graph CNN classifiers.

KW - graph Laplacian regularization

KW - binary classification

KW - semi-supervised learning,

KW - deep learning

UR - https://2019.ieeeicassp.org/

M3 - Paper

ER -

Ye M, Stankovic V, Stankovic L, Cheung G. Deep graph regularized learning for binary classification. 2019. Paper presented at 2019 International Conference on Acoustics, Speech, and Signal Processing, Brighton, United Kingdom.