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

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/Congresprocedure › Conferentiebijdrage › Academic › peer review

3 Citaten (Scopus)

Samenvatting

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.

Originele taal-2	Engels
Titel	Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings
Redacteuren	Nuria Oliver, Fernando Pérez-Cruz, Stefan Kramer, Jesse Read, Jose A. Lozano
Uitgeverij	Springer
Pagina's	367-382
Aantal pagina's	16
ISBN van geprinte versie	9783030865191
DOI's	https://doi.org/10.1007/978-3-030-86520-7_23
Status	Gepubliceerd - 2021
Evenement	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021 - Virtual, Online Duur: 13 sep. 2021 → 17 sep. 2021

Publicatie series

Naam	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12976 LNAI
ISSN van geprinte versie	0302-9743
ISSN van elektronische versie	1611-3349

Congres

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

Bibliografische nota

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

Toegang tot document

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

Andere bestanden en links

Link naar publicatie in Scopus

Citeer dit

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 (editors), Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings (blz. 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. redacteur / Nuria Oliver ; Fernando Pérez-Cruz ; Stefan Kramer ; Jesse Read ; Jose A. Lozano. Springer, 2021. blz. 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",

address = "Germany",

}

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 (uitgave), 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, blz. 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. uitgave / Nuria Oliver; Fernando Pérez-Cruz; Stefan Kramer; Jesse Read; Jose A. Lozano. Springer, 2021. blz. 367-382 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12976 LNAI).

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/Congresprocedure › Conferentiebijdrage › 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, redacteurs, Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings. Springer. 2021. blz. 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

Samenvatting

Publicatie series

Congres

Bibliografische nota

Toegang tot document

Andere bestanden en links

Vingerafdruk

Citeer dit