On Generalization of Graph Autoencoders with Adversarial Training

Tianjin Huang; Yulong Pei; Vlado Menkovski; Mykola Pechenizkiy

doi:10.1007/978-3-030-86520-7_23

On Generalization of Graph Autoencoders with Adversarial Training

Tianjin Huang, Yulong Pei, Vlado Menkovski, Mykola Pechenizkiy

Data Mining

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

3 Citations (Scopus)

Abstract

Adversarial training is an approach for increasing model’s resilience against adversarial perturbations. Such approaches have been demonstrated to result in models with feature representations that generalize better. However, limited works have been done on adversarial training of models on graph data. In this paper, we raise such a question – does adversarial training improve the generalization of graph representations. We formulate L₂ and L_∞ versions of adversarial training in two powerful node embedding methods: graph autoencoder (GAE) and variational graph autoencoder (VGAE). We conduct extensive experiments on three main applications, i.e. link prediction, node clustering, graph anomaly detection of GAE and VGAE, and demonstrate that both L₂ and L_∞ adversarial training boost the generalization of GAE and VGAE.

Original language	English
Title of host publication	Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings
Editors	Nuria Oliver, Fernando Pérez-Cruz, Stefan Kramer, Jesse Read, Jose A. Lozano
Publisher	Springer
Pages	367-382
Number of pages	16
ISBN (Print)	9783030865191
DOIs	https://doi.org/10.1007/978-3-030-86520-7_23
Publication status	Published - 2021
Event	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021 - Virtual, Online Duration: 13 Sept 2021 → 17 Sept 2021

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12976 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021
City	Virtual, Online
Period	13/09/21 → 17/09/21

Bibliographical note

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

Keywords

Adversarial training
Generalization
Graph autoencoders
Node embedding
Variational graph autoencoders

Access to Document

10.1007/978-3-030-86520-7_23

Cite this

Huang, T., Pei, Y., Menkovski, V., & Pechenizkiy, M. (2021). On Generalization of Graph Autoencoders with Adversarial Training. In N. Oliver, F. Pérez-Cruz, S. Kramer, J. Read, & J. A. Lozano (Eds.), Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings (pp. 367-382). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12976 LNAI). Springer. https://doi.org/10.1007/978-3-030-86520-7_23

Huang, Tianjin ; Pei, Yulong ; Menkovski, Vlado et al. / On Generalization of Graph Autoencoders with Adversarial Training. Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings. editor / Nuria Oliver ; Fernando Pérez-Cruz ; Stefan Kramer ; Jesse Read ; Jose A. Lozano. Springer, 2021. pp. 367-382 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{f5f3ab0e3dda49a7a3352d569bda0d45,

title = "On Generalization of Graph Autoencoders with Adversarial Training",

abstract = "Adversarial training is an approach for increasing model{\textquoteright}s resilience against adversarial perturbations. Such approaches have been demonstrated to result in models with feature representations that generalize better. However, limited works have been done on adversarial training of models on graph data. In this paper, we raise such a question – does adversarial training improve the generalization of graph representations. We formulate L2 and L∞ versions of adversarial training in two powerful node embedding methods: graph autoencoder (GAE) and variational graph autoencoder (VGAE). We conduct extensive experiments on three main applications, i.e. link prediction, node clustering, graph anomaly detection of GAE and VGAE, and demonstrate that both L2 and L∞ adversarial training boost the generalization of GAE and VGAE.",

keywords = "Adversarial training, Generalization, Graph autoencoders, Node embedding, Variational graph autoencoders",

author = "Tianjin Huang and Yulong Pei and Vlado Menkovski and Mykola Pechenizkiy",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021 ; Conference date: 13-09-2021 Through 17-09-2021",

year = "2021",

doi = "10.1007/978-3-030-86520-7_23",

language = "English",

isbn = "9783030865191",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "367--382",

editor = "Nuria Oliver and Fernando P{\'e}rez-Cruz and Stefan Kramer and Jesse Read and Lozano, {Jose A.}",

booktitle = "Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings",

}

Huang, T , Pei, Y , Menkovski, V & Pechenizkiy, M 2021, On Generalization of Graph Autoencoders with Adversarial Training. in N Oliver, F Pérez-Cruz, S Kramer, J Read & JA Lozano (eds), Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12976 LNAI, Springer, pp. 367-382, European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021, Virtual, Online, 13/09/21. https://doi.org/10.1007/978-3-030-86520-7_23

On Generalization of Graph Autoencoders with Adversarial Training. / Huang, Tianjin ; Pei, Yulong ; Menkovski, Vlado et al.
Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings. ed. / Nuria Oliver; Fernando Pérez-Cruz; Stefan Kramer; Jesse Read; Jose A. Lozano. Springer, 2021. p. 367-382 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12976 LNAI).

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

TY - GEN

T1 - On Generalization of Graph Autoencoders with Adversarial Training

AU - Huang, Tianjin

AU - Pei, Yulong

AU - Menkovski, Vlado

AU - Pechenizkiy, Mykola

PY - 2021

Y1 - 2021

N2 - Adversarial training is an approach for increasing model’s resilience against adversarial perturbations. Such approaches have been demonstrated to result in models with feature representations that generalize better. However, limited works have been done on adversarial training of models on graph data. In this paper, we raise such a question – does adversarial training improve the generalization of graph representations. We formulate L2 and L∞ versions of adversarial training in two powerful node embedding methods: graph autoencoder (GAE) and variational graph autoencoder (VGAE). We conduct extensive experiments on three main applications, i.e. link prediction, node clustering, graph anomaly detection of GAE and VGAE, and demonstrate that both L2 and L∞ adversarial training boost the generalization of GAE and VGAE.

AB - Adversarial training is an approach for increasing model’s resilience against adversarial perturbations. Such approaches have been demonstrated to result in models with feature representations that generalize better. However, limited works have been done on adversarial training of models on graph data. In this paper, we raise such a question – does adversarial training improve the generalization of graph representations. We formulate L2 and L∞ versions of adversarial training in two powerful node embedding methods: graph autoencoder (GAE) and variational graph autoencoder (VGAE). We conduct extensive experiments on three main applications, i.e. link prediction, node clustering, graph anomaly detection of GAE and VGAE, and demonstrate that both L2 and L∞ adversarial training boost the generalization of GAE and VGAE.

KW - Adversarial training

KW - Generalization

KW - Graph autoencoders

KW - Node embedding

KW - Variational graph autoencoders

UR - http://www.scopus.com/inward/record.url?scp=85115721477&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-86520-7_23

DO - 10.1007/978-3-030-86520-7_23

M3 - Conference contribution

AN - SCOPUS:85115721477

SN - 9783030865191

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 367

EP - 382

BT - Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings

A2 - Oliver, Nuria

A2 - Pérez-Cruz, Fernando

A2 - Kramer, Stefan

A2 - Read, Jesse

A2 - Lozano, Jose A.

PB - Springer

T2 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021

Y2 - 13 September 2021 through 17 September 2021

ER -

Huang T , Pei Y , Menkovski V , Pechenizkiy M. On Generalization of Graph Autoencoders with Adversarial Training. In Oliver N, Pérez-Cruz F, Kramer S, Read J, Lozano JA, editors, Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings. Springer. 2021. p. 367-382. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-86520-7_23

On Generalization of Graph Autoencoders with Adversarial Training

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this