Feature-level analysis and adversarial transfer in rotationally equivariant quantum machine learning

arXiv:2604.15552 · quant-ph · Submitted 2026-04-16 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: We're looking at the paper: "Feature-level analysis and adversarial transfer in rotationally equivariant quantum machine learning".

Paper summary: Kai: We've talked about what this paper is about—it’s not just about equivariance improving trainability; it’s specifically about how symmetry constraints shape adversarial robustness. The paper, "Feature-level analysis and adversarial transfer in rotationally equivariant quantum machine learning," studies this question through a feature-level analysis of these models in a transfer-attack setting.

Mira: It claims that under equivariance with an invariant readout, the predictions only depend on the group twirled input, which identifies the symmetry invariant information accessible to the model alongside some complementary uninformative subspace.

Lev: And they specialize this framework to rotationally equivariant quantum models and derive a specific characterization of what information is accessible in terms of rotation invariant image statistics distributed across distinct symmetry sectors.

Kai: They then use targeted input transformations to determine which of these statistics are actually being relied upon for classification across several different datasets.

Mira: The core finding is that the model only accesses rotationally invariant image statistics that decompose across its symmetry channels, and these accessible features are governed by circular correlations called C r,r prime delta phi.

Lev: To probe this, they introduce two ways to test which statistics are used: targeted input transformations and modifications at the readout level. They show that while the model is restricted to rotation invariant information, the particular invariant statistics it relies on change a lot depending on which dataset you use.

Kai: Then they look at robustness under transfer attacks and find that when trained on a transformation called T3, which removes ring averages, the model becomes substantially more robust than when trained on clean data.

Mira: This is interesting because even for standard datasets like STM or RotFMNIST, the clean-trained model already shows poor robustness, similar to what you'd see with a transformation called T2.

Lev: The authors suggest that ring-average information accounts for a large chunk of the discriminative signal these models use on these specific types of data.

Kai: So, their main takeaway is that suppressing the symmetry channel associated with those ring averaged intensities substantially improves robustness, and they find this effect is strongest on datasets where that feature was previously most vulnerable.

Conclusion: Mira: So to wrap up, the paper "Feature-level analysis and adversarial transfer in rotationally equivariant quantum machine learning" shows us that simply having equivariance isn't a complete solution for adversarial robustness.

Kai: They demonstrate that simple invariant features, like the ring-averaged intensities in this rotationally equivariant model, can remain highly predictive while being very non robust to transfer attacks.

Lev: The paper provides a mechanistic understanding of how these quantum models use their symmetry channels to make decisions, which helps explain properties like adversarial robustness.

Mira: The authors show that by performing a feature-level analysis and suppressing the specific symmetry channel responsible for those brittle features, we can substantially improve robustness without losing too much accuracy.

Kai: This opens up the idea that we can use this framework to get a mechanistic look at other architectures with different types of symmetry constraints.

Lev: It gives us a concrete way to understand exactly where the vulnerabilities are hidden in these quantum models, moving beyond just observing that they are fragile under attack.

Maureen Krumtülnger, Martin Sevior, Muhammad Usman

School of Physics, University of Melbourne · Data61, CSIRO

quant-ph

Submitted: 2026-04-16

Updated: 2026-04-16

Journal ref: Advanced Quantum Technologies (2026)

DOI: 10.1002/qute.70521

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 92/100

The gist: Group-equivariant quantum models are designed to exploit symmetry and can improve trainability, but it remains unclear how symmetry constraints shape their adversarial robustness The authors study…

Key concepts

Group-equivariant quantum models
These are quantum machine learning models designed to respect a specific mathematical symmetry group. The goal is to use this inherent symmetry to make the model better at learning and potentially more robust against attacks. However, the study shows that just having this symmetry isn't enough on its own.
Invariant Readout
This means the final prediction of the model only depends on information that remains unchanged when you apply a transformation from a specific symmetry group. This limits what the model can 'see' but also defines the core, symmetric information it uses for classification.
Brittle Statistics
These are specific types of features, such as ring-averaged intensities in the rotationally equivariant model, that seem important for training but are not reliable when facing adversarial attacks. They are fragile; small changes to these specific statistics can cause large prediction errors under transfer attacks.
Transfer Attacks
These are adversarial attacks where an attacker creates an input perturbation designed to fool a model trained on one dataset, and that same attack successfully fools the model when tested on a different, unseen dataset. The paper examines how symmetry affects the success of these attacks.

Terminology

Summary

Group-equivariant quantum models are designed to exploit symmetry and can improve trainability, but it remains unclear how symmetry constraints shape their adversarial robustness The authors study this question through a feature-level analysis of equivariant quantum models in a transfer-attack setting. p. 0

Where the paper says it

Feature-level analysis and adversarial transfer in rotationally equivariant quantum machine learning Maureen Krumtünger,1, ∗ Martin Sevior,1 and Muhammad Usman1, 2 1School of Physics, University of Melbourne, Parkville, VIC 3010, Australia 2Data61, CSIRO, Clayton, 3168, Victoria, Australia Group-equivariant quantum models are designed to exploit symmetry and can improve trainability, but it remains unclear how symmetry constraints shape their adversarial robustness. We study this question through a featurelevel analysis of equivariant quantum models in a transfer-attack setting.

Under equivariance with an invariant readout, predictions depend only on the group-twirled input, which identifies the symmetry-invariant information accessible to the model together with a complementary uninformative subspace. p. 0
Where the paper says it

We study this question through a featurelevel analysis of equivariant quantum models in a transfer-attack setting. Under equivariance with an invariant readout, predictions depend only on the group-twirled input, which identifies the symmetry-invariant information accessible to the model together with a complementary uninformative subspace.

We find that equivariance alone does not guarantee transfer robustness: even within the restricted invariant feature space, the model can rely on brittle statistics, particularly ring-averaged intensities in the rotationally equivariant model, that remain vulnerable to classical transfer attacks. p. 0
Where the paper says it

Using targeted input transformations, we determine which of these statistics are actually relied upon for classification across several datasets. We find that equivariance alone does not guarantee transfer robustness: even within the restricted invariant feature space, the model can rely on brittle statistics, particularly ring-averaged intensities in the rotationally equivariant model, that remain vulnerable to classical transfer attacks.

Guided by this analysis, we show that suppressing the symmetry sector associated with the brittle feature substantially improves robustness. p. 0
Where the paper says it

We find that equivariance alone does not guarantee transfer robustness: even within the restricted invariant feature space, the model can rely on brittle statistics, particularly ring-averaged intensities in the rotationally equivariant model, that remain vulnerable to classical transfer attacks. Guided by this analysis, we show that suppressing the symmetry sector associated with the brittle feature substantially improves robustness.

Feature Accessibility and Characterization

The model's prediction depends only on the group-twirled input, which identifies the symmetry-invariant information accessible to the model together with a complementary uninformative subspace. Specializing this framework to a rotationally equivariant quantum model, we derive an explicit characterization of the accessible information in terms of rotation-invariant image statistics distributed across distinct symmetry sectors. p. 0

Where the paper says it

Under equivariance with an invariant readout, predictions depend only on the group-twirled input, which identifies the symmetry-invariant information accessible to the model together with a complementary uninformative subspace. Specializing this framework to a rotationally equivariant quantum model, we derive an explicit characterization of the accessible information in terms of rotation-invariant image statistics distributed across distinct symmetry sectors.

Using targeted input transformations, we determine which of these statistics are actually relied upon for classification across several datasets. p. 0
Where the paper says it

Specializing this framework to a rotationally equivariant quantum model, we derive an explicit characterization of the accessible information in terms of rotation-invariant image statistics distributed across distinct symmetry sectors. Using targeted input transformations, we determine which of these statistics are actually relied upon for classification across several datasets.

We show that the model only accesses rotationally invariant image statistics that decompose across its symmetry channels. p. 0
Where the paper says it

[5] introduced this architecture and established its trainability, our contribution is different in scope: we derive a twirling-based characterization of the input information the model can access, thereby obtaining a much clearer understanding of how the model works. We show that the model only accesses rotationally invariant image statistics that decompose across its symmetry channels.

The accessible information is governed by the circular correlations C r,r' (Δϕ):= N/sum over phi-1 alpha=0 x r,alpha x'r,alpha-Δϕ, which are invariant under global rotations x r, phi -> x r,(phi+g) mod Nphi.

Probing and Identifying Brittle Features

To probe which accessible statistics are used by the trained model, we introduce two complementary approaches to probe which accessible statistics are used by the trained model: (i) targeted input transformations and (ii) readout-level modifications. Transformation 1 (T1) is designed to directly test this prediction. p. 3

Where the paper says it

The characterization above yields a concrete prediction: a Z𝑁𝜙 -equivariant model should be invariant under any input transformation that leaves the twirled representation TZ𝑁𝜙 (𝜌) unchanged. Transformation 1 (T1) is designed to directly test this prediction.

Transformation 3 (T3), ring-wise mean removal, eliminates the ring averages while retaining the correlation structure up to a Δϕ-independent offset. p. 4
Where the paper says it

We subtract the ring mean from each ring, 𝑥 (3) 𝑟 = 𝑥𝑟 − 𝑥¯𝑟 1, 𝑥¯𝑟 = 1 𝑁𝜙 ∑︁ 𝜙 𝑥𝑟, 𝜙, (17) where 1 is an all-ones vector. This eliminates the ring averages while retaining the correlation structure up to a Δ𝜙- independent offset (see Appendix C).

Overall, while the model is restricted to rotation-invariant information, the particular invariant statistics it relies on vary substantially across datasets. p. 5
Where the paper says it

At the same time, the relatively weak performance even on T3 indicates that these features alone do not fully account for the discriminative signal present in the clean data. Overall, while the model is restricted to rotation-invariant information, the particular invariant statistics it relies on vary substantially across datasets.

Robustness under Transfer Attacks

The model becomes substantially more robust when trained on T3 than when trained on clean data, although this comes at the cost of lower clean-test accuracy. By contrast, models trained on T2 are consistently less robust. p. 6

Where the paper says it

The model becomes substantially more robust when trained on T3 than when trained on clean data, although this comes at the cost of lower clean-test accuracy. By contrast, models trained on T2 are consistently less robust.

For STM, RotFMNIST, and CIFAR, the clean-trained model already exhibits poor robustness, at a level similar to T2. p. 6
Where the paper says it

By contrast, models trained on T2 are consistently less robust. Moreover, for STM, RotFMNIST, and CIFAR, the clean-trained model already exhibits poor robustness, at a level similar to T2.

This suggests that ring-average information accounts for a large fraction of the discriminative signal used by the model on these datasets. p. 5
Where the paper says it

Moreover, models trained on T2 generalize to clean test data slightly better than T3-trained models, and substantially better in the case of STM. Together, these results suggest that ring-mean information accounts for a large fraction of the discriminative signal used by the model on these datasets.

Improving Robustness through Suppression

The strongest improvement is observed for the datasets that were previously most vulnerable to the ring-average feature. Projecting out the m = 0 mode, which carries the ring-averaged intensities, increases the robustness of the model without any substantial loss in accuracy. p. 8

Where the paper says it

At the same time, the equivariant structure—especially through its irreducible decomposition—provides a principled and targeted handle for controlling specific invariant features. In particular, we showed that projecting out the 𝑚 = 0 sector, which carries the ring-averaged intensities, increases the robustness of the model without any substantial loss in accuracy.

This second strategy is particularly motivated by the fact that training on T3 also improves robustness, but at a substantial cost in clean-test accuracy. p. 7
Where the paper says it

tectural intervention: training on clean data while projecting out the 𝑚 = 0 channel, corresponding to the ring-averaged intensities, which Section III identified as a likely source of non-robustness. This second strategy is particularly motivated by the fact that training on T3 also improves robustness, but at a substantial cost in clean-test accuracy; see Table I (b).

Overall, this makes the m = 0 projection a particularly effective method in our setting. p. 8
Where the paper says it

For CNN- and ResNet18-based attacks, however, it yields a larger robustness improvement than adversarial training on the LC-generated examples. Overall, this makes the 𝑚 = 0 projection a particularly effective method in our setting.

Implications and Outlook

Our results suggest that this is not the case in general, although equivariance renders the model insensitive to a large non-invariant subspace of perturbations. In particular, we find that simple invariant features, such as the ring-averaged intensities of the rotationally equivariant model, can remain highly predictive while being strongly non-robust. p. 8

Where the paper says it

As a result, brittle and robust features can still coexist within the restricted class of rotation-invariant statistics. In particular, we find that simple invariant features, such as the ring-averaged intensities of the rotationally equivariant model, can remain highly predictive while being strongly non-robust, whereas richer invariant structures, such as higher-order angular correlation statistics, can be substantially more robust.

More broadly, this idea could be explored further by investigating in greater detail the signal contribution of the remaining symmetry channels. p. 8
Where the paper says it

In particular, we showed that projecting out the 𝑚 = 0 sector, which carries the ring-averaged intensities, increases the robustness of the model without any substantial loss in accuracy. More broadly, this idea could be explored further by investigating in greater detail the signal contribution of the remaining symmetry channels, and by studying architectural strategies that either preserve them as separate feature streams or recombine them only at a later stage.

We show how equivariance can enable a mechanistic, feature-level understanding of quantum models and thereby help explain properties such as adversarial robustness. p. 8
Where the paper says it

Thus, this framework could be similarly used for a mechanistic analysis of other equivariant architectures. Overall, our study shows how equivariance can enable a mechanistic, feature-level understanding of quantum models and thereby help explain properties such as adversarial robustness.

The framework could be similarly used for a mechanistic analysis of other equivariant architectures. p. 8
Where the paper says it

[5], the same twirling-based reasoning extends to other strictly equivariant quantum models; what becomes case-specific is the interpretation of the resulting symmetry channels in terms of input-space features for a given encoding. Thus, this framework could be similarly used for a mechanistic analysis of other equivariant architectures.

References

[1] M. Larocca, F. p. 9

Where the paper says it

The authors acknowledge the computational resources provided by the University of Melbourne’s Research Computing Service. Larocca, F.

Sauvage, F. p. 9
Where the paper says it

Larocca, F. Sauvage, F.

M. Sbahi, G. Verdon, P. J. Coles, and M. p. 9
Where the paper says it

Sauvage, F. Coles, and M.

Cerezo, Group-invariant quantum machine learning, PRX Quantum 3, 030341 (2022)

[2] J. J. Meyer, M. Mularski, E. p. 9

Where the paper says it

Cerezo, Group-invariant quantum machine learning, PRX Quantum 3, 030341 (2022). Mularski, E.

Gil-Fuster, A. p. 9
Where the paper says it

Mularski, E. Gil-Fuster, A.

A. Mele, F. Arzani, A. Wilms, and J. p. 9
Where the paper says it

Gil-Fuster, A. Wilms, and J.

Eisert, Exploiting symmetry in variational quantum machine learning, PRX Quantum 4, 010328 (2023)

[3] Q. T. Nguyen, L. Schatzki, P. p. 9

Where the paper says it

Eisert, Exploiting symmetry in variational quantum machine learning, PRX Quantum 4, 010328 (2023). Schatzki, P.

Braccia, M. p. 9
Where the paper says it

Schatzki, P. Braccia, M.

Ragone, P. J. Coles, F. Sauvage, M. p. 9
Where the paper says it

Larocca, F. Sauvage, F.

Larocca, and M. p. 9
Where the paper says it

The authors acknowledge the computational resources provided by the University of Melbourne’s Research Computing Service. Larocca, F.

Cerezo, Theory for equivariant quantum neural networks, PRX Quantum 5, 020328 (2024)

[4] L. Schatzki, M. p. 9

Where the paper says it

Eisert, Exploiting symmetry in variational quantum machine learning, PRX Quantum 4, 010328 (2023). Schatzki, P.

Larocca, Q. p. 9
Where the paper says it

The authors acknowledge the computational resources provided by the University of Melbourne’s Research Computing Service. Larocca, F.

T. Nguyen, F. Sauvage, and M. p. 9
Where the paper says it

Larocca, F. Sauvage, F.

Cerezo, Theoretical guarantees for permutation-equivariant quantum neural networks, npj Quantum Information 10, 10. p. 9
Where the paper says it

Sauvage, and M. Cerezo, Theoretical guarantees for permutation-equivariant quantum neural networks, npj Quantum Information 10, 10.

1038/s41534-024-00804-1 (2024)

[5] M. T. West, J. Heredge, M. p. 9

Where the paper says it

1038/s41534-024-00804-1 (2024). Heredge, M.

Sevior, and M. p. 0
Where the paper says it

Feature-level analysis and adversarial transfer in rotationally equivariant quantum machine learning Maureen Krumtünger,1, ∗ Martin Sevior,1 and Muhammad Usman1, 2 1School of Physics, University of Melbourne, Parkville, VIC 3010, Australia 2Data61, CSIRO, Clayton, 3168, Victoria, Australia Group-equivariant quantum models are designed to exploit symmetry and can improve trainability, but it remains unclear how symmetry constraints shape their adversarial robustness. We study this question through a featurelevel analysis of equivariant quantum models in a transfer-attack setting.

p. 0
Where the paper says it

Feature-level analysis and adversarial transfer in rotationally equivariant quantum machine learning Maureen Krumtünger,1, ∗ Martin Sevior,1 and Muhammad Usman1, 2 1School of Physics, University of Melbourne, Parkville, VIC 3010, Australia 2Data61, CSIRO, Clayton, 3168, Victoria, Australia Group-equivariant quantum models are designed to exploit symmetry and can improve trainability, but it remains unclear how symmetry constraints shape their adversarial robustness. We study this question through a featurelevel analysis of equivariant quantum models in a transfer-attack setting.

Usman, Provably trainable rotationally equivariant quantum machine learning, PRX Quantum 5, 030320 (2024)

[6] M. T West, M. Sevior, and M. Usman, Reflection equivariant quantum neural networks for enhanced image classification, Machine Learning: Science and Technology 4, 035027 (2023)

[7] X. Yuan, P. He, Q. Zhu, and X. p. 9

Where the paper says it

Usman, Reflection equivariant quantum neural networks for enhanced image classification, Machine Learning: Science and Technology 4, 035027 (2023). Zhu, and X.

Li, Adversarial examples: Attacks and defenses for deep learning (2018), arXiv:1712. p. 9
Where the paper says it

Zhu, and X. Li, Adversarial examples: Attacks and defenses for deep learning (2018), arXiv:1712.

07107 [cs.LG]

[8] O. Ibitoye, R. p. 9

Where the paper says it

Li, Adversarial examples: Attacks and defenses for deep learning (2018), arXiv:1712. Ibitoye, R.

Abou-Khamis, M. p. 9
Where the paper says it

Ibitoye, R. Abou-Khamis, M.

el Shehaby, A. p. 9
Where the paper says it

Abou-Khamis, M. el Shehaby, A.

Matrawy, and M. p. 9
Where the paper says it

el Shehaby, A. Matrawy, and M.

O. Shafiq, The threat of adversarial attacks on machine learning in network security – a survey (2023), arXiv:1911. p. 9
Where the paper says it

Matrawy, and M. Shafiq, The threat of adversarial attacks on machine learning in network security – a survey (2023), arXiv:1911.

02621 [cs.CR]

[9] I. J. Goodfellow, J. p. 9

Where the paper says it

Shafiq, The threat of adversarial attacks on machine learning in network security – a survey (2023), arXiv:1911. Goodfellow, J.

Shlens, and C. p. 9
Where the paper says it

Goodfellow, J. Shlens, and C.

Szegedy, Explaining and harnessing adversarial examples (2015), arXiv:1412. p. 9
Where the paper says it

Shlens, and C. Szegedy, Explaining and harnessing adversarial examples (2015), arXiv:1412.

6572 [stat. p. 9
Where the paper says it

Szegedy, Explaining and harnessing adversarial examples (2015), arXiv:1412. 6572 [stat.

ML]

[10] D. Tsipras, S. p. 9

Where the paper says it

6572 [stat. Tsipras, S.

Santurkar, L. p. 9
Where the paper says it

Tsipras, S. Santurkar, L.

Engstrom, A. p. 9
Where the paper says it

Santurkar, L. Engstrom, A.

Turner, and A. p. 9
Where the paper says it

Engstrom, A. Turner, and A.

Madry, Robustness may be at odds with accuracy, in International Conference on Learning Representations (2019) [stat. p. 9
Where the paper says it

Turner, and A. Madry, Robustness may be at odds with accuracy, in International Conference on Learning Representations (2019).

ML]

[11] A. Ilyas, S. Santurkar, D. p. 9

Where the paper says it

Tsipras, S. Santurkar, L.

Tsipras, L. p. 9
Where the paper says it

6572 [stat. Tsipras, S.

Engstrom, B. p. 9
Where the paper says it

Santurkar, L. Engstrom, A.

Tran, and A. p. 9
Where the paper says it

Engstrom, B. Tran, and A.

Madry, Adversarial examples are not bugs, they are features (2019), arXiv:1905. p. 9
Where the paper says it

Tran, and A. Madry, Adversarial examples are not bugs, they are features (2019), arXiv:1905.

02175 [stat. p. 9
Where the paper says it

Madry, Adversarial examples are not bugs, they are features (2019), arXiv:1905. 02175 [stat.

ML]

[12] A. Demontis, M. p. 9

Where the paper says it

02175 [stat. Demontis, M.

Melis, M. Pintor, M. Jagielski, B. p. 9
Where the paper says it

Demontis, M. Jagielski, B.

Biggio, A. Oprea, C. Nita-Rotaru, and F. p. 9
Where the paper says it

Jagielski, B. Nita-Rotaru, and F.

Roli, Why do adversarial attacks transfer? p. 9
Where the paper says it

Nita-Rotaru, and F. Roli, Why do adversarial attacks transfer?

explaining transferability of evasion and poisoning attacks (2019), arXiv:1809. p. 9
Where the paper says it

Roli, Why do adversarial attacks transfer? explaining transferability of evasion and poisoning attacks (2019), arXiv:1809.

02861 [cs.LG]

[13] A. Madry, A. Makelov, L. p. 9

Where the paper says it

explaining transferability of evasion and poisoning attacks (2019), arXiv:1809. Makelov, L.

Schmidt, D. p. 9
Where the paper says it

Makelov, L. Schmidt, D.

Tsipras, and A. p. 9
Where the paper says it

6572 [stat. Tsipras, S.

Vladu, Towards deep learning models resistant to adversarial attacks, in International Conference on Learning Representations (2018) [stat. p. 9
Where the paper says it

Tsipras, and A. Vladu, Towards deep learning models resistant to adversarial attacks, in International Conference on Learning Representations (2018).

ML]

[14] M. Usman, Quantum Robustness in Artificial Intelligence: Principles and Applications, Quantum Science and Technology (Springer Nature Switzerland, 2026)

[15] S. Lu, L.-M. Duan, and D. p. 9

Where the paper says it

Usman, Quantum Robustness in Artificial Intelligence: Principles and Applications, Quantum Science and Technology (Springer Nature Switzerland, 2026). Duan, and D.

p. 9
Where the paper says it

Usman, Quantum Robustness in Artificial Intelligence: Principles and Applications, Quantum Science and Technology (Springer Nature Switzerland, 2026). Duan, and D.

-L. Deng, Quantum adversarial machine learning, Physical Review Research 2, 10. p. 9
Where the paper says it

Duan, and D. Deng, Quantum adversarial machine learning, Physical Review Research 2, 10.

1103/physrevresearch. p. 9
Where the paper says it

Deng, Quantum adversarial machine learning, Physical Review Research 2, 10. 1103/physrevresearch.

p. 9
Where the paper says it

Deng, Quantum adversarial machine learning, Physical Review Research 2, 10. 1103/physrevresearch.

2.033212 (2020)

[16] W. Gong and D. p. 9

Where the paper says it

033212 (2020). Gong and D.

-L. Deng, Universal adversarial examples and perturbations for quantum classifiers, National Science Review 10. p. 9
Where the paper says it

Gong and D. Deng, Universal adversarial examples and perturbations for quantum classifiers, National Science Review 10.

1093/nsr/nwab130 (2021)

[17] M. T. West, S. M. Erfani, C. Leckie, M. Sevior, L. C. L. Hollenberg, and M. p. 9

Where the paper says it

1093/nsr/nwab130 (2021). Hollenberg, and M.

Usman, Benchmarking adversarially robust quantum machine learning at scale, Physical Review Research 5, 10. p. 9
Where the paper says it

Hollenberg, and M. Usman, Benchmarking adversarially robust quantum machine learning at scale, Physical Review Research 5, 10.

1103/physrevresearch.5.023186 (2023)

[18] M. T. West, S.-L. Tsang, J. S. Low, C. D Hill, C Leckie, L C L Hollenberg, S M Erfani, and M Usman, Towards quantum enhanced adversarial robustness in machine learning (Nature Machine Intelligence 5)

[19] N. Dowling, M. p. 9

Where the paper says it

Usman, Towards quantum enhanced adversarial robustness in machine learning, Nature Machine Intelligence 5, 581–589 (2023). Dowling, M.

T. West, A Southwell, A C Nakhl, M Sevior, M Usman, and K Modi, Adversarial robustness guarantees for quantum classifiers (npj Quantum Information 12)

[20] M. Ragone, P. Braccia, Q. p. 9

Where the paper says it

Schatzki, P. Braccia, M.

T. Nguyen, L Schatzki, P J Coles, F Sauvage, M Larocca, and M Cerezo, Representation theory for geometric quantum machine learning (2023), arXiv:2210. p. 9
Where the paper says it

Larocca, and M. Cerezo, Representation theory for geometric quantum machine learning (2023), arXiv:2210.

07980 [quant-ph]

[21] Y. Lecun, L. Bottou, Y. Bengio, and P. p. 9

Where the paper says it

07980 [quant-ph]. Bengio, and P.

Haffner, Gradient-based learning applied to document recognition (Proceedings of the IEEE 86)

[22] H. Xiao, K. Rasul, and R. p. 9

Where the paper says it

Haffner, Gradient-based learning applied to document recognition, Proceedings of the IEEE 86, 2278 (1998). Rasul, and R.

Vollgraf, Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms (2017), arXiv:1708. p. 9
Where the paper says it

Rasul, and R. Vollgraf, Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms (2017), arXiv:1708.

p. 9
Where the paper says it

Rasul, and R. Vollgraf, Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms (2017), arXiv:1708.

07747 [cs.LG]

[23] A. Krizhevsky, Learning multiple layers of features from tiny images (2009)

[24] K. He, X. Zhang, S. Ren, and J. p. 9

Where the paper says it

Krizhevsky, Learning multiple layers of features from tiny images (2009). Ren, and J.

Sun, Deep residual learning for image recognition (CVPR 2016)

[27] M. Zhao, L Zhang, J Ye, H Lu, B Yin, and X Wang, Adversarial training: A survey (2024), arXiv:2410. p. 9

Where the paper says it

Yin, and X. Wang, Adversarial training: A survey (2024), arXiv:2410.

15042 [cs.LG]

[28] T. Bai, J. Luo, J Zhao, B Wen, and Q Wang, Recent advances in adversarial training for adversarial robustness (2021), arXiv:2102. p. 9

Where the paper says it

Wen, and Q. Wang, Recent advances in adversarial training for adversarial robustness (2021), arXiv:2102.

01356 [cs.LG]

[29] A. Raghunathan, S. p. 9

Where the paper says it

Wang, Recent advances in adversarial training for adversarial robustness (2021), arXiv:2102. Raghunathan, S.

M Xie, F Yang, J Duchi, and P Liang, Understanding and mitigating the tradeoff between robustness and accuracy (2020), arXiv:2002. p. 9
Where the paper says it

Duchi, and P. Liang, Understanding and mitigating the tradeoff between robustness and accuracy (2020), arXiv:2002.

10716 [cs.LG]

[30] T. Pang, M. Lin, X Yang, J Zhu, and S Yan, Robustness and accuracy could be reconcilable by (proper) definition (2022), arXiv:2202. p. 9

Where the paper says it

Zhu, and S. Yan, Robustness and accuracy could be reconcilable by (proper) definition (2022), arXiv:2202.

10103 [cs.LG]

[31] L Wang, I. I Uddin, K Santosh, C Zhang, X Qin, and Y Zhou, Bridging symmetry and robustness: On the role of equivariance in enhancing adversarial robustness (The Thirty-ninth Annual Conference on Neural Information Processing Systems (2025))

[32] M. Usman, Y. Z. Wong, C D Hill, et al. p. 9

Where the paper says it

Zhou, Bridging symmetry and robustness: On the role of equivariance in enhancing adversarial robustness, in The Thirty-ninth Annual Conference on Neural Information Processing Systems (2025). Hill, et al.

, Framework for atomiclevel characterisation of quantum computer arrays by machine learning, npj Computational Materials 6, 19 (2020)

[33] M. T West and M Usman, Framework for donor-qubit spatial metrology in silicon with depths approaching the bulk limit, Phys. p. 9

Where the paper says it

West and M. Usman, Framework for donor-qubit spatial metrology in silicon with depths approaching the bulk limit, Phys.

Rev. Appl. 17, 024070 (2022)

[34] M. Usman, B Voisin, J Salfi, S Rogge, and L C L Hollenberg, Towards visualisation of central-cell-effects in scanning tunnelling microscope images of subsurface dopant qubits in silicon (Nanoscale 9)

[35] K. Chen and L. p. 9

Where the paper says it

Hollenberg, Towards visualisation of central-cell-effects in scanning tunnelling microscope images of subsurface dopant qubits in silicon, Nanoscale 9, 17013–17019 (2017). Chen and L.

p. 9
Where the paper says it

Hollenberg, Towards visualisation of central-cell-effects in scanning tunnelling microscope images of subsurface dopant qubits in silicon, Nanoscale 9, 17013–17019 (2017). Chen and L.

Liu, Privacy preserving data classification with rotation perturbation, in Proceedings of the Fifth IEEE International Conference on Data Mining, ICDM ’05 (IEEE Computer Society, USA, 2005) p. p. 9
Where the paper says it

Chen and L. Liu, Privacy preserving data classification with rotation perturbation, in Proceedings of the Fifth IEEE International Conference on Data Mining, ICDM ’05 (IEEE Computer Society, USA, 2005) p.

589–592

[36] K. Chen and L. Liu, Geometric data perturbation for privacy preserving outsourced data mining, Knowl. p. 9

Where the paper says it

Chen and L. Liu, Geometric data perturbation for privacy preserving outsourced data mining, Knowl.

Inf. Syst. 29, 657–695 (2011)

[37] K. Liu, H Kargupta, and J Ryan, Random projection-based multiplicative data perturbation for privacy preserving distributed data mining, IEEE Trans. p. 9

Where the paper says it

Chen and L. Liu, Privacy preserving data classification with rotation perturbation, in Proceedings of the Fifth IEEE International Conference on Data Mining, ICDM ’05 (IEEE Computer Society, USA, 2005) p.

on Knowl. and Data Eng. p. 0
Where the paper says it

However, recent studies have shown that quantum models can exhibit increased resilience to transfer attacks crafted on classical surrogate models [17–19]. One possible explanation is that quantum and classical learners rely on different predictive features of the same data, as evidenced by the qualitatively different perturbation patterns observed for attacks optimized on classical versus quantum models [17].

18, 92–106 (2006)

[38] C Dwork, F McSherry, K Nissim, and A Smith, Calibrating noise to sensitivity in private data analysis (Springer-Verlag, Berlin, Heidelberg, 2006) p. 265–284

[39] M T West and M Usman, Framework for atomiclevel characterisation of quantum computer arrays by machine learning (npj Computational Materials 6)

[40] L Schatzki, M Larocca, Q T Nguyen, F Sauvage, and M Cerezo, Representation theory for geometric quantum machine learning (2023), arXiv:2210.07980 [quant-ph]

[41] Y Lecun, L. Bottou, Y Bengio, and P Haffner, Gradient-based learning applied to document recognition (Proceedings of the IEEE 86)

[42] H Xiao, K Rasul, and R. Vollgraf, Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms (2017), arXiv:1708.07747 [cs.LG]

[43] A Krizhevsky, Learning multiple layers of features from tiny images (2009)

[44] K He, X. Zhang, S Ren, and J Sun, Deep residual learning for image recognition (CVPR 2016)

[47] M T West, S M Erfani, C Leckie, M Sevior, L C L Hollenberg, and M Usman, Benchmarking adversarially robust quantum machine learning at scale (Physical Review Research 5)

[48] N Dowling, M T West, A Southwell, A C Nakhl, M Sevior, M Usman, and K Modi, Adversarial robustness guarantees for quantum classifiers (npj Quantum Information 12)

[49] S Lu, L.-M. Duan, and D.-L. Deng, Quantum adversarial machine learning (Physical Review Research 2)

[50] W Gong and D.-L. Deng, Universal adversarial examples and perturbations for quantum classifiers (National Science Review)

How the training signal is built

The rotationally equivariant quantum model is trained for 10 epochs using a depth-100 circuit. All reported results are averaged over five independent training runs. p. 4

Where the paper says it

For each training-set variant, the rotationally equivariant model is trained for 10 epochs using a depth-100 circuit. All reported results are averaged over five independent training runs.

We consider a mixed adversarial training set by selecting 50% of the original training samples uniformly at random, while leaving the remaining samples unchanged. p. 4
Where the paper says it

For each dataset, we consider the original training set (clean) together with transformed variants obtained by applying each transformation from Sec. III B 1 to all training inputs while leaving the labels unchanged; these variants are denoted T1, T2, and T3.

For each selected sample, we replace it with an adversarial counterpart using a FGSM attack on the LC surrogate model. p. 4
Where the paper says it

Adversarial examples. To evaluate transfer robustness, we generate adversarial examples using FGSM and PGD with four classical surrogate models: a linear classifier (LC),

The perturbations were projected onto the corresponding l-infinity ball. p. 2
Where the paper says it

PGD is a multi-step variant of the FGSM attack [13]. Starting from an initial perturbation 𝑥 0 , it iteratively updates the adversarial example as 𝑥 𝑡+1 = ΠB𝜀 (𝑥) 𝑥 𝑡 + 𝛼 sign∇𝑥𝐿(𝜽 ∗ , 𝑥𝑡 , 𝑦) , (6) where ΠB𝜀 (𝑥) denotes the projection onto the ℓ∞-ball B𝜀 (𝑥) = {𝑥 ′ : ∥𝑥 ′ − 𝑥∥∞ ≤ 𝜀} around the original input 𝑥, 𝛼 is the step size, and 𝐿 is the loss function of the model.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Feature-level analysis for robustness assessment: The system can perform a mechanistic understanding of which input features are actually accessible to a group equivariant quantum model by deriving a twirling-based characterization of input information. This allows for targeted input transformations, such as Transformation 1, to test the model's reliance on specific rotation-invariant statistics.

  2. Robustness enhancement via feature suppression: The system can improve robustness by suppressing the symmetry sector associated with brittle features. Specifically, projecting out the m equals zero Fourier mode increases robustness without a substantial loss in accuracy across all datasets when compared to adversarial training alone.

  3. Dataset-dependent feature reliance mapping: The system can map which rotation-invariant statistics are utilized by a model on a dataset-specific basis. For example, it can determine that for STM, RotFMNIST, and CIFAR, the model exhibits a strong reliance on the information retained in T2, while for MNIST and RotMNIST, it retains more useful information under T3.

  4. Practical data obfuscation: The system can implement a practical obfuscation scheme using Transformation 1 for privacy-preserving data sharing. Training on scrambled images preserves the rotation-invariant correlations accessible to the model, allowing utility comparable to training on clean inputs without exposing original pixel-level content directly.

Abstract

Group-equivariant quantum models are designed to exploit symmetry and can improve trainability, but it remains unclear how symmetry constraints shape their adversarial robustness. We study this question through a feature-level analysis of equivariant quantum models in a transfer-attack setting. Under equivariance with an invariant readout, predictions depend only on the group-twirled input, which identifies the symmetry-invariant information accessible to the model together with a complementary uninformative subspace. Specializing this framework to a rotationally equivariant quantum model, we derive an explicit characterization of the accessible information in terms of rotation-invariant image statistics distributed across distinct symmetry sectors. Using targeted input transformations, we determine which of these statistics are actually relied upon for classification across several datasets. We find that equivariance alone does not guarantee transfer robustness: even within the restricted invariant feature space, the model can rely on brittle statistics, particularly ring-averaged intensities in the rotationally equivariant model, that remain vulnerable to classical transfer attacks. Guided by this analysis, we show that suppressing the symmetry sector associated with the brittle feature substantially improves robustness. These results establish a systematic mechanism to exploit symmetry-dependent features for adversarial robustness in future quantum machine learning models.

Sources

Related papers