Hi!

Thanks again for maintaining this valuable resource.

We have three new NeurIPS papers on (provably) robust learning on graphs
and wanted to ask if we could include them.

In this paper, we revisit the problem of proving robustness for graphs from a perspective of geometric machine learning.
Among other things, we show that proving robustness w.r.t. graph edit distance is actually not as hard as one might think.

(Provable) Adversarial Robustness for Group Equivariant Tasks: Graphs, Point Clouds, Molecules, and More
https://openreview.net/forum?id=mLe63bAYc7
Code will be made available here: https://www.cs.cit.tum.de/daml/equivariance-robustness/

In this paper, we propose a procedure to prove robustness to adversaries that only control a limited number of nodes and only a limited fraction of their features / edges.

Hierarchical Randomized Smoothing
https://openreview.net/forum?id=6IhNHKyuJO
Code will be made available here: https://www.cs.cit.tum.de/daml/hierarchical-smoothing

In this paper, we revisit adversarial training for graph neural networks.

Adversarial Training for Graph Neural Networks: Pitfalls, Solutions, and New Directions
https://openreview.net/forum?id=GPtroppvUM
Code will be made available here: https://www.cs.cit.tum.de/daml/adversarial-training/

Thanks again,
Jan

Robustness of Graph Neural Networks at Scale

Covering two scalable attacks and one defense.

https://www.in.tum.de/daml/robustness-of-gnns-at-scale/

@inproceedings{geisler2021_robustness_of_gnns_at_scale,
title = {Robustness of Graph Neural Networks at Scale},
author = {Tobias Schmidt, Hakan \c{S}irin, DanielZ"ugner, Aleksandar Bojchevski, and StephanG"unnemann},
booktitle={Neural Information Processing Systems, {NeurIPS}},
year = {2021}
}

Generalization of Neural Combinatorial Solvers Through the Lens of Adversarial Robustness

Adversarial attacks on neural solvers for TSP and SAT.

https://arxiv.org/abs/2110.10942

Thanks :-)

Hi! Thanks a lot for curating this very helpful collection of graph adversarial robustness papers.

I wanted to ask if you could add the following two certificate papers from our group (the second one can be applied to various tasks, but is especially effective for graph neural networks).

https://www.cs.cit.tum.de/daml/interception-smoothing

 @inproceedings{scholten2022interception_smoothing,
    title = {Randomized Message-Interception Smoothing: Gray-box Certificates for Graph Neural Networks},
    author = {Scholten, Yan and Schuchardt, Jan and Geisler, Simon and Bojchevski, Aleksandar and G{\"u}nnemann, Stephan},
    booktitle={Neural Information Processing Systems, {NeurIPS}},
    year = {2022}
    }

and

https://openreview.net/forum?id=-k7Lvk0GpBl

   @inproceedings{schuchardt2023localized_smoothing,
    title = {Localized Randomized Smoothing for Collective Robustness Certification},
    author = {Schuchardt, Jan and Wollschl\"ager, Tom and Bojchevski, Aleksandar and G{\"u}nnemann, Stephan},
    booktitle={International Conference on Learning Representations, {ICLR}},
    year = {2023}
    }