Pith. sign in

REVIEW 4 major objections 4 minor 39 references

Edge-Local and Qubit-Efficient Quantum Graph Learning for the NISQ Era

T0 review · 4 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A fully quantum graph network can learn node embeddings with only O(n) qubits, independent of graph size, by processing message passing edge by edge.

desk verdict The edge-local qubit-efficient scheme is a genuine resource idea, but the message-passing layer is inert under the only readout defined, and the SNP headline is circular. read the letter →

arxiv 2602.16018 v2 pith:OXG6BCUF submitted 2026-02-17 quant-ph cs.ETcs.LG

Armin Ahmadkhaniha , Jake Doliskani This is my paper
classification quant-phcs.ETcs.LG
keywords quantumgraphneuralnetworksmessagepassingQAOA-inspiredansatzunsupervisedrepresentationlearningDeepInfomaxNISQhardwarequbitefficiencygenomicSNPdata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes a quantum graph convolutional architecture in which message passing is decomposed into pairwise, edge-local interactions rather than a global operation over all nodes. The authors claim this reduces the qubit requirement from O(Nn) to O(n) for an N-node graph with n-qubit feature registers, making the model implementable on near-term hardware regardless of graph size. They train the model for unsupervised node embedding with a mutual-information contrastive objective and report that on the Cora citation network the learned clusters align with ground-truth classes (NMI 0.51), while a hybrid quantum-classical baseline using classical message passing does not (NMI 0.06). On a 5008-node genomic SNP graph the embeddings separate five human super-populations with 98% pseudo-label classification accuracy. The broader claim is that coherent quantum message passing, not just quantum feature extraction, is what preserves semantic structure in unsupervised graph learning.

What carries the argument

The load-bearing object is the edge-local message-passing unitary U_MP,uv(γ,β) = ∏_k RX_{u_k}(2β) ∏_k e^{-iγ Z_{u_k} Z_{v_k}}, a QAOA-style alternating-operator circuit for a single edge. Each two-qubit interaction e^{-iγZ_uZ_v} is implemented as CNOT, RZ(2γ), CNOT, and the whole circuit uses only hardware-native single- and two-qubit gates. The paper applies this unitary sequentially to each edge's 2n-qubit register, then aggregates per-node outputs by summation; together with angle encoding of the extracted features and a mutual-information training objective, this converts the graph structure into a sequence of local quantum operations whose qubit cost stays O(n).

What would settle it

For a single edge and n=1, prepare the angle-encoded states RX(2h_u)|0⟩ and RX(2h_v)|0⟩, apply RX_{u}(2β) followed by e^{-iγZ_uZ_v}, and measure ⟨Z_u⟩ and ⟨Z_v⟩ in the computational basis. If the measured values are independent of γ for all β,h_u,h_v, then the message-passing unitary transmits no information under this readout, and the central claim of graph-structure processing fails.

Watch

Extended reading notes

Core claim

The central claim is that a fully quantum GCN can perform meaningful unsupervised node representation learning on NISQ hardware by replacing global, hardware-intensive message passing with an edge-local, qubit-efficient mechanism inspired by QAOA. For each edge, the model applies alternating unitaries e^{-iβH_M} and e^{-iγH_C}, with H_C a sum of Z_u Z_v couplings between corresponding feature qubits of neighboring nodes and H_M a sum of single-qubit X rotations. Because the edge unitaries factor into local terms, each edge can be processed sequentially using only 2n qubits, and the authors state that the total qubit count becomes O(n), independent of the number of nodes. Trained end-to-end w

Load-bearing premise

The load-bearing premise is that the state after edge-local quantum message passing can be measured in a way that actually conveys which neighbours a node has; the paper never names that observable, and if it is the Pauli-Z expectation used in feature extraction, the ZZ interaction is invisible to the readout.

Editorial extensions

If this is right

  • Any graph, regardless of node count, can be embedded on a device with about 2n qubits, provided edges are processed one at a time; for bounded-degree graphs the gate count stays linear in N.
  • Multi-controlled unitaries, a major NISQ bottleneck in earlier quantum GCN proposals, are avoided entirely, so compilation overhead and noise sensitivity should drop.
  • Unsupervised quantum feature extraction plus quantum message passing preserves semantic alignment (Cora NMI 0.51), whereas the same quantum features with classical sum aggregation do not (NMI 0.06), suggesting the quantum step carries structural information.
  • The model scales to 5008 nodes and 805 binary features on the SNP dataset, producing clusters aligned with five known human super-populations and pointing toward practical genomic applications.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not specify the observable used to read out the message-passing output; since e^{-iγZ_uZ_v} is diagonal in the Z basis, Pauli-Z expectation values of both registers are invariant under it. If readout uses Z expectations, the edge-local layer cannot transmit neighbor information, and the reported gains would have to come from local rotations or the aggregation step. This hidden assum
  • A clean control experiment would set γ=0, removing the ZZ interaction, and compare the learned embeddings; if the results change only through β, then graph structure is not being used by the circuit.
  • The sequential edge-by-edge scheme trades qubit count for repeated re-encoding and measurement per edge, so end-to-end runtime on real hardware scales with |E|, a practical limit the paper only partially acknowledges in its complexity analysis.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a fully quantum graph convolutional architecture for unsupervised node embedding on NISQ hardware. It combines a variational quantum feature-extraction circuit with a QAOA-inspired edge-local message-passing unitary, using the Deep Graph Infomax objective. The authors claim the qubit requirement is reduced from O(Nn) to O(n) by processing edges one at a time with 2n-qubit circuits. Experiments on Cora and a genomic SNP dataset report strong clustering and classification results, including a Cora NMI of 0.51 versus 0.06 for a hybrid baseline.

Significance. If the central mechanism were fully specified and the evaluations were non-circular, the paper would offer a meaningful step toward qubit-efficient quantum graph learning. The edge-local decomposition and the explicit complexity analysis are useful, and the code is publicly available. However, the current manuscript leaves the readout observable after message passing unspecified; with the only stated readout, the ZZ interaction transmits no neighbor information. The SNP evaluation is also circular, so the empirical evidence for the headline claims is not established.

major comments (4)
  1. [§3.2.3, Fig. 3, Alg. 1] The observable measured after U_MP,uv is never specified. The only readout defined in the paper is the Pauli-Z expectation per qubit (§3.1). With angle encoding |φ_h⟩=⊗RX(2h_k)|0⟩ and U_MP,uv containing e^{-iγZ_uZ_v}, the ZZ gate is diagonal in the Z basis. For n=1, a direct calculation gives ⟨Z_u⟩ after the edge-local circuit = cos(2β)cos(2h_u) − sin(2β)sin(2h_u)cos(2γ), independent of h_v; the neighbor-dependent signal appears only in ⟨X_u⟩ = sin(2h_u)cos(2h_v)sin(2γ). Thus, if a Pauli-Z readout is used, the message-passing layer transmits no neighbor information and degenerates into local RX rotations. The authors must specify the readout observable and show that it transfers information across the edge; as written, the central claim is underspecified.
  2. [§4.1, Table 1] The SNP evaluation is circular: k-means clusters of the learned embeddings are used as pseudo-labels, and logistic-regression accuracy on those pseudo-labels (98%) is presented as evidence that the model distinguishes the five super-populations. This only demonstrates that the clusters are linearly separable in the embedding space; it does not establish alignment with true population structure. The text states that the dataset lacks ground-truth labels, yet invokes the known biological taxonomy to justify k=5. If the 1000 Genomes super-population labels are available, they should be used; otherwise the 98% figure should be described as internal consistency, not classification accuracy. The Cora NMI of 0.51 is the only non-circular quantitative result.
  3. [§4.1] The comparison with the hybrid model does not isolate the contribution of the quantum message-passing layer. There is no ablation with quantum feature extraction alone (i.e., no message passing). Under a Pauli-Z readout, U_MP is local, so the reported Cora NMI 0.51 could be produced entirely by the feature extractor plus degree-weighted summation. To support the claim that quantum message passing preserves semantic structure, the authors should add a baseline that omits the ZZ interactions while keeping all other components identical.
  4. [§3.2.3, Eq. (6), Fig. 3] There is an inconsistency between the mathematical definition of U_MP,uv and the circuit diagram. The equation defines U_MP,uv := ∏_k RX_uk(2β) ∏_k e^{-iγZ_ukZ_vk}, i.e., a mixer rotation only on node u, whereas Figure 3 shows RX(2β) on both u and v. This matters for the calculation of expectations and for any implementation. Additionally, Algorithm 1 aggregates multiple edge embeddings classically by summation, so the 'fully quantum' characterization and the contrast with hybrid models in the introduction are overstated; the graph-level aggregation is classical.
minor comments (4)
  1. [Throughout] There are several typos: 'Prelimenaries', 'unitareis', 'feaure', 'genral', 'resutling' should be corrected.
  2. [§3.1] The feature-extraction procedure outputs a classical vector h via Pauli-Z expectations, and the message-passing step later re-encodes h via angle encoding. This two-step 'encode—measure—re-encode' pipeline should be described more clearly; the sentence 'producing an n-qubit feature encoding' in §3.2.3 is misleading.
  3. [§4.1, Table 1] The column 'Accuracy' is logistic-regression accuracy on k-means pseudo-labels, not ground-truth accuracy. The table should make this explicit, and the Cora ground-truth accuracy (0.23 for the hybrid, presumably similar for the full model) should be reported alongside the pseudo-label accuracy.
  4. [§3.3.1] Amplitude encoding generally requires O(d) multi-qubit gates, not just 'O(N d) single-qubit rotations' as stated. This affects the complexity comparison and should be corrected.

Circularity Check

2 steps flagged · score 6.0 of 10

SNP evaluation is circular (k-means pseudo-labels scored by logistic regression on the same embeddings), and the message-passing output observable is unspecified; under the only readout defined, the ZZ interaction carries no neighbor signal, so the central graph-processing claim reduces to local rotations.

  1. fitted input called prediction [§4.1 Results (SNP dataset); Table 1]
    "Since the SNP dataset does not provide ground-truth population labels, we adopt a self-supervised evaluation strategy. ... we treat the cluster assignments obtained via k-means clustering with k=5 as pseudo-labels. We then train a standard logistic regression classifier to predict these pseudo-labels from the embeddings. ... our model achieves a logistic regression classification accuracy of 98%. This result indicates that our model does not merely recover coarse-grained clusters, but instead learns embeddings that capture fine-grained structure required to distinguish all five super-populatio"

    The reported 98% accuracy is computed against pseudo-labels produced by k-means on the very embeddings being evaluated. Logistic regression is trained to predict labels that are a deterministic function of the same vectors it sees, so the accuracy only measures whether those self-generated cluster assignments are linearly reproducible from the embeddings. It is high by construction and cannot provide evidence about correspondence to the five true super-populations. The paper explicitly frames this as internal consistency, but then uses it to support a claim about distinguishing the five super-populations.

  2. self definitional [§3.1 Eq. (3), §3.2.3 Eq. (6), Algorithm 1 line 6]
    "Finally, the classical encoding of the node x is obtained by computing the Pauli-Z expectation values on each qubit of |ψ_L(θ)⟩. ... Measurement of the output yields updated embeddings for both nodes. ... e^{-iγZ_{uk}Z_{vk}} = cnot_{uk→vk} · RZ_{vk}(2γ) · cnot_{uk→vk}."

    Equation (6) is diagonal in the Z basis, and the only readout defined in the paper is the Pauli-Z expectation vector h of §3.1. For the angle-encoded states |φ_h⟩=RX(2h)|0⟩ used in §3.2.3, direct calculation for one feature qubit gives ⟨Z_u⟩ after U_MP,uv = cos(2β)cos(2h_u) − sin(2β)cos(2γ)sin(2h_u), independent of h_v; the neighbor-dependent term lives in ⟨X_u⟩, which the local RX(2β) rotations do not rotate into Z. Thus, with the paper's stated readout, the updated embedding of each node is a local function of its own h, the ZZ edge interaction carries no information between nodes, and the message-passing layer degenerates into local single-qubit rotations. No alternative observable is specified, so the claimed graph-structure processing is not carried by the equations as written.

full rationale

The Hamiltonian-to-circuit part of the paper is internally consistent: Eq. (4)-(6) commute and decompose into CNOT+RZ+RX gates, and Algorithm 1's repeated 2n-qubit edge circuits genuinely give an O(n)-qubit, O(|E|log d) procedure. This part does not depend on circular reasoning, and there are no load-bearing self-citations: QAOA, DGI, GCN, and amplitude encoding are all external prior work. The circularity is concentrated in two places. First, the SNP evaluation uses k-means clusters of the learned embeddings as pseudo-labels and then reports logistic-regression accuracy on those labels; this accuracy is a tautological measure of linear reproducibility of the same clusters, not evidence about the five super-populations. Second, the message-passing output observable is never specified; the only readout defined anywhere is the Pauli-Z expectation vector of §3.1. Since e^{-iγZ_uZ_v} is Z-diagonal and the angle-encoded inputs have zero X-quadrature initially, the neighbor-dependent information appears only in quadratures that the local RX(2β) rotations do not rotate into Z; under the paper's own readout the updated embeddings are independent of the neighbor and the message-passing layer degenerates to local rotations. The Cora NMI=0.51 against ground-truth labels is a genuine external benchmark and some independent content remains, but the central graph-processing claim is not established without specifying a readout that transmits neighbor information. Because one evaluation is circular by construction and the central mechanism is definitionally inert under the stated readout, score 6.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim rests on standard QC assumptions plus several domain choices: trainable variational angles whose values are unreported, a single unspecified readout observable, a mutual-kNN graph construction, and DGI's contrastive signal. No new physical entities are introduced.

free parameters (6)
  • Variational angles θ in feature-extraction circuit = not reported
    Trained to minimize DGI loss; no values or initialization given.
  • QAOA angles β (mixer) and γ (cost) in U_MP,uv = not reported
    Trainable parameters for the message-passing unitary; values not reported.
  • Number of entangling layers L = not reported
    Assumed constant in complexity analysis but never specified in experiments.
  • Range parameter r in entangling block B = not reported
    Controls which qubit pairs are entangled; value not stated.
  • k = 5 for mutual kNN graph construction (SNP) = 5
    Chosen by hand; no sensitivity analysis.
  • k = 5 for k-means clustering (SNP) = 5
    Chosen 'to align with the known biological taxonomy' even though the paper reports Silhouette/elbow suggest k = 3 or 4 (§4.1).
assumptions (6)
  • standard math Amplitude encoding of d-dimensional features into ceil(log d) qubits is possible and efficient.
    Requires 2^n ≥ d, used in §3.1; standard in QML.
  • domain assumption Parameterized quantum circuits can be trained by classical optimization on NISQ hardware.
    Invoked throughout §3; no noise model or gradient estimation details are given.
  • domain assumption Deep Graph Infomax provides a valid unsupervised training signal for graph embeddings.
    Adopted from [32] and applied directly to the measured embeddings.
  • ad hoc to paper A single layer of pairwise ZZ interactions suffices as a message-passing step.
    The paper does not analyze how many QAOA layers or edge passes are needed to propagate information across the graph.
  • ad hoc to paper The unspecified measurement observable in the message-passing step yields embeddings that preserve neighbor information.
    Central to §3.2.3 but never stated; if Z-expectation is used, the ZZ interaction is inert.
  • ad hoc to paper Mutual kNN graph construction with reciprocal relaxation produces meaningful graph structure for the SNP data.
    Introduced in §4 for the SNP dataset; no comparison to other graph constructions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Edge-Local and Qubit-Efficient Quantum Graph Learning for the NISQ Era." pith.science (2026). https://pith.science/paper/OXG6BCUF

@misc{pith2026260216018,
  author       = {Pith},
  title        = {Pith review of: Edge-Local and Qubit-Efficient Quantum Graph Learning for the NISQ Era},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OXG6BCUF}},
  note         = {Machine review of arXiv:2602.16018}
}
abstract

Graph neural networks (GNNs) are a powerful framework for learning representations from graph-structured data, but their direct implementation on near-term quantum hardware remains challenging due to circuit depth, multi-qubit interactions, and qubit scalability constraints. In this work, we introduce a fully quantum graph convolutional architecture designed explicitly for unsupervised learning in the noisy intermediate-scale quantum (NISQ) regime. Our approach combines a variational quantum feature extraction layer with an edge-local and qubit-efficient quantum message-passing mechanism inspired by the Quantum Alternating Operator Ansatz (QAOA) framework. Unlike prior models that rely on global operations or multi-controlled unitaries, our model decomposes message passing into pairwise interactions along graph edges using only hardware-native single- and two-qubit gates. This design reduces the qubit requirement from $O(Nn)$ to $O(n)$ for a graph with $N$ nodes and $n$-qubit feature registers, enabling implementation on current quantum devices regardless of graph size. We train the model using the Deep Graph Infomax objective to perform unsupervised node representation learning. Experiments on the Cora citation network and a large-scale genomic SNP dataset demonstrate that our model remains competitive with prior quantum and hybrid approaches.

Figures

Figures reproduced from arXiv: 2602.16018 by the authors.

Figure 1
Figure 1. Five-qubit quantum feature extraction for a single [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. General QAOA circuit with input |ψ0⟩ and parameter sets (β1, . . . , βp) and (γ1, . . . , γp). hard constraints or to respect the structure of the feasible so￾lution space. From a conceptual perspective, QAOA can be interpreted in several complementary ways. It may be viewed as a dig￾itized version of adiabatic quantum optimization, where in￾creasing the depth p yields progressively better approxima￾tions to continu… view at source ↗
Figure 3
Figure 3. The quantum message-passing unitary UMP,uv(γ, β) for an edge (u, v), where each node is represented by n = 4 qubits. The key idea behind our qubit-efficient construction is to apply the edge-local unitary UMP,uv separately for each edge (u, v) ∈ E, rather than implementing the full unitary UMP on all nodes simultaneously. To obtain embeddings for all nodes in G, we first apply the quantum feature extraction procedur… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: t-SNE embedding visualization of node representa [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 10 linked inside Pith

  1. [1]

    Towards quantum graph neural net- works: An ego-graph learning approach.arXiv preprint arXiv:2201.05158, 2022

    Xing Ai, Zhihong Zhang, Luzhe Sun, Junchi Yan, and Edwin Hancock. Towards quantum graph neural net- works: An ego-graph learning approach.arXiv preprint arXiv:2201.05158, 2022

    arXiv 2022

  2. [2]

    El- ementary gates for quantum computation.Physical re- view A, 52(5):3457, 1995

    Adriano Barenco, Charles H Bennett, Richard Cleve, David P DiVincenzo, Norman Margolus, Peter Shor, Ty- cho Sleator, John A Smolin, and Harald Weinfurter. El- ementary gates for quantum computation.Physical re- view A, 52(5):3457, 1995

    1995

  3. [3]

    Noisy intermediate-scale quantum algo- rithms.Reviews of Modern Physics, 94(1):015004, 2022

    Kishor Bharti, Alba Cervera-Lierta, Thi Ha Kyaw, Tobias Haug, Sumner Alperin-Lea, Abhinav Anand, Matthias Degroote, Hermanni Heimonen, Jakob S Kottmann, Tim Menke, et al. Noisy intermediate-scale quantum algo- rithms.Reviews of Modern Physics, 94(1):015004, 2022

    2022

  4. [4]

    Quantum machine learning.Nature, 549(7671):195–202, 2017

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Rebentrost, Nathan Wiebe, and Seth Lloyd. Quantum machine learning.Nature, 549(7671):195–202, 2017

    2017

  5. [5]

    Genomic.https://pennylane.ai/ datasets/other/genomic, 2025

    Joseph Bowles. Genomic.https://pennylane.ai/ datasets/other/genomic, 2025

    2025

  6. [6]

    Variational quantum algorithms.Nature Reviews Physics, 3(9):625–644, 2021

    Marco Cerezo, Andrew Arrasmith, Ryan Babbush, Si- mon C Benjamin, Suguru Endo, Keisuke Fujii, Jarrod R McClean, Kosuke Mitarai, Xiao Yuan, Lukasz Cincio, et al. Variational quantum algorithms.Nature Reviews Physics, 3(9):625–644, 2021

    2021

  7. [7]

    Hybrid quantum- classical graph convolutional network.arXiv preprint arXiv:2101.06189, 2021

    Samuel Yen-Chi Chen, Tzu-Chieh Wei, Chao Zhang, Haiwang Yu, and Shinjae Yoo. Hybrid quantum- classical graph convolutional network.arXiv preprint arXiv:2101.06189, 2021

    arXiv 2021

  8. [8]

    Jet discrimination with a quantum complete graph neural network.Physical Re- view D, 111(1):016020, 2025

    Yi-An Chen and Kai-Feng Chen. Jet discrimination with a quantum complete graph neural network.Physical Re- view D, 111(1):016020, 2025

    2025

Show all 39 references
  1. [9]

    A quantum approximate optimization algorithm.arXiv preprint arXiv:1411.4028, 2014

    Edward Farhi, Jeffrey Goldstone, and Sam Gutmann. A quantum approximate optimization algorithm.arXiv preprint arXiv:1411.4028, 2014

    2014 arXiv

  2. [10]

    Classification with quantum neural networks on near term processors

    Edward Farhi and Hartmut Neven. Classification with quantum neural networks on near term processors. arXiv preprint arXiv:1802.06002, 2018

    2018 arXiv

  3. [11]

    Large-scale learnable graph convolutional networks

    Hongyang Gao, Zhengyang Wang, and Shuiwang Ji. Large-scale learnable graph convolutional networks. pages 1416–1424, 2018

    2018

  4. [12]

    Neural message pass- ing for quantum chemistry

    Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message pass- ing for quantum chemistry. InInternational conference on machine learning, pages 1263–1272. Pmlr, 2017

    2017

  5. [13]

    From the quantum approximate optimization algorithm to a quantum alternating operator ansatz.Algorithms, 12(2):34, 2019

    Stuart Hadfield, Zhihui Wang, Bryan O’gorman, Eleanor G Rieffel, Davide Venturelli, and Rupak Biswas. From the quantum approximate optimization algorithm to a quantum alternating operator ansatz.Algorithms, 12(2):34, 2019

    2019

  6. [14]

    Induc- tive representation learning on large graphs.Advances in neural information processing systems, 30, 2017

    Will Hamilton, Zhitao Ying, and Jure Leskovec. Induc- tive representation learning on large graphs.Advances in neural information processing systems, 30, 2017

    2017

  7. [15]

    Supervised learning with quantum- enhanced feature spaces.Nature, 567(7747):209–212, 2019

    Vojt ˇech Havl´ıˇcek, Antonio D C´orcoles, Kristan Temme, Aram W Harrow, Abhinav Kandala, Jerry M Chow, and Jay M Gambetta. Supervised learning with quantum- enhanced feature spaces.Nature, 567(7747):209–212, 2019

    2019

  8. [16]

    On the design of quantum graph convolu- tional neural network in the nisq-era and beyond

    Zhirui Hu, Jinyang Li, Zhenyu Pan, Shanglin Zhou, Lei Yang, Caiwen Ding, Omer Khan, Tong Geng, and Wei- wen Jiang. On the design of quantum graph convolu- tional neural network in the nisq-era and beyond. pages 290–297, 2022

    2022

  9. [17]

    Highly accurate protein structure pre- diction with alphafold.nature, 596(7873):583–589, 2021

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ronneberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin ˇZ´ıdek, Anna Potapenko, et al. Highly accurate protein structure pre- diction with alphafold.nature, 596(7873):583–589, 2021

    2021

  10. [18]

    Semi-supervised clas- sification with graph convolutional networks.arXiv preprint arXiv:1609.02907, 2016

    Thomas N Kipf and Max Welling. Semi-supervised clas- sification with graph convolutional networks.arXiv preprint arXiv:1609.02907, 2016

    2016 arXiv

  11. [19]

    Imagenet classification with deep convolutional neural networks.Advances in neural information pro- cessing systems, 25, 2012

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hin- ton. Imagenet classification with deep convolutional neural networks.Advances in neural information pro- cessing systems, 25, 2012

    2012

  12. [20]

    Robust data encodings for quantum classifiers.arXiv preprint arXiv:2003.01695, 2020

    Ryan LaRose and Brian Coyle. Robust data encodings for quantum classifiers.arXiv preprint arXiv:2003.01695, 2020

    2003 arXiv

  13. [21]

    Equivariant quantum graph circuits

    P ´eter Mernyei, Konstantinos Meichanetzidis, and Is- mail Ilkan Ceylan. Equivariant quantum graph circuits. InInternational conference on machine learning, pages 15401–15420. PMLR, 2022

    2022

  14. [22]

    Quantum computing in the nisq era and beyond.Quantum, 2:79, 2018

    John Preskill. Quantum computing in the nisq era and beyond.Quantum, 2:79, 2018

    2018

  15. [23]

    Improving language understanding by generative pre-training

    Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al. Improving language understanding by generative pre-training. 2018

    2018

  16. [24]

    Train on classical, deploy on quantum: scal- ing generative quantum machine learning to a thousand qubits.arXiv preprint arXiv:2503.02934, 2025

    Erik Recio-Armengol, Shahnawaz Ahmed, and Joseph Bowles. Train on classical, deploy on quantum: scal- ing generative quantum machine learning to a thousand qubits.arXiv preprint arXiv:2503.02934, 2025. 10

    2025

  17. [25]

    Quantum and classical graph convolutional neural networks for pro- tein ligand dissociation constant prediction.bioRxiv, pages 2025–11, 2025

    Azamat Salamatov and Gowtham Atluri. Quantum and classical graph convolutional neural networks for pro- tein ligand dissociation constant prediction.bioRxiv, pages 2025–11, 2025

    2025

  18. [26]

    The graph neural network model.IEEE transactions on neural net- works, 20(1):61–80, 2008

    Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph neural network model.IEEE transactions on neural net- works, 20(1):61–80, 2008

    2008

  19. [27]

    Modeling relational data with graph convolutional net- works

    Michael Schlichtkrull, Thomas N Kipf, Peter Bloem, Rianne Van Den Berg, Ivan Titov, and Max Welling. Modeling relational data with graph convolutional net- works. pages 593–607, 2018

    2018

  20. [28]

    Circuit-centric quantum classifiers

    Maria Schuld, Alex Bocharov, Krysta M Svore, and Nathan Wiebe. Circuit-centric quantum classifiers. Physical Review A, 101(3):032308, 2020

    2020

  21. [29]

    Supervised learning with quantum computers.Quantum science and technology, 17, 2018

    Maria Schuld and Francesco Petruccione. Supervised learning with quantum computers.Quantum science and technology, 17, 2018

    2018

  22. [30]

    Springer, 2021

    Maria Schuld and Francesco Petruccione.Machine learning with quantum computers, volume 676. Springer, 2021

    2021

  23. [31]

    Hy- brid quantum-classical graph neural networks for parti- cle track reconstruction.Quantum Machine Intelligence, 3(2):29, 2021

    Cenk T ¨uys¨uz, Carla Rieger, Kristiane Novotny, Bilge Demirk¨oz, Daniel Dobos, Karolos Potamianos, Sofia Vallecorsa, Jean-Roch Vlimant, and Richard Forster. Hy- brid quantum-classical graph neural networks for parti- cle track reconstruction.Quantum Machine Intelligence, 3(2)...

    2021

  24. [32]

    Deep graph infomax.arXiv preprint arXiv:1809.10341, 2018

    Petar Veli ˇckovi´c, William Fedus, William L Hamilton, Pietro Li`o, Yoshua Bengio, and R Devon Hjelm. Deep graph infomax.arXiv preprint arXiv:1809.10341, 2018

    2018 arXiv

  25. [33]

    Quantum graph neural networks.arXiv preprint arXiv:1909.12264, 2019

    Guillaume Verdon, Trevor McCourt, Enxhell Luzhnica, Vikash Singh, Stefan Leichenauer, and Jack Hidary. Quantum graph neural networks.arXiv preprint arXiv:1909.12264, 2019

    1909 arXiv

  26. [34]

    Cam- bridge university press, 2018

    John Watrous.The theory of quantum information. Cam- bridge university press, 2018

    2018

  27. [35]

    How powerful are graph neural networks? arXiv preprint arXiv:1810.00826, 2018

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? arXiv preprint arXiv:1810.00826, 2018

    2018 arXiv

  28. [36]

    Quantum graph convo- lutional networks based on spectral methods.arXiv preprint arXiv:2503.06447, 2025

    Zi Ye, Kai Yu, and Song Lin. Quantum graph convo- lutional networks based on spectral methods.arXiv preprint arXiv:2503.06447, 2025

    2025 arXiv

  29. [37]

    Creating artifi- cial human genomes using generative neural networks

    Burak Yelmen, Aur ´elien Decelle, Linda Ongaro, Davide Marnetto, Corentin Tallec, Francesco Montinaro, Cyril Furtlehner, Luca Pagani, and Flora Jay. Creating artifi- cial human genomes using generative neural networks. PLoS genetics, 17(2):e1009303, 2021

    2021

  30. [38]

    Quantum graph convolutional neural networks

    Jin Zheng, Qing Gao, and Yanxuan L ¨u. Quantum graph convolutional neural networks. In2021 40th Chinese Control Conference (CCC), pages 6335–6340. IEEE, 2021

    2021

  31. [39]

    A quantum spatial graph convolutional neural network model on quantum circuits.IEEE Transactions on Neural Networks and Learning Systems, 36(3):5706–5720, 2024

    Jin Zheng, Qing Gao, Maciej Ogorzałek, Jinhu L ¨u, and Yue Deng. A quantum spatial graph convolutional neural network model on quantum circuits.IEEE Transactions on Neural Networks and Learning Systems, 36(3):5706–5720, 2024. 11

    2024

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.