Pith. sign in

REVIEW 4 major objections 4 minor 36 references

QGHNN: A quantum graph Hamiltonian neural network

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Encoding a graph as a spin-lattice Hamiltonian and training the derived circuit recovers the graph with up to 99.8% cosine similarity, outperforming VQE, QAOA, and QNN baselines.

desk verdict A clearly written graph-learning ansatz that is undone by a load-bearing non-commutation error in Eq. (10), which the numeric results cannot repair. read the letter →

arxiv 2501.07986 v1 pith:EXWM7ETR submitted 2025-01-14 quant-ph

Wenxuan Wang This is my paper · ORCID
classification quant-ph
keywords quantumgraphHamiltonianlearningneuralnetworkrepresentationparameterizedcircuitNISQdevicesvariationalalgorithmadjacencymatrixencoding
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 way to represent a classical graph as the Hamiltonian of a topological (lattice) quantum system, with the adjacency matrix appearing as coupling coefficients of Pauli operators. The paper then builds a quantum graph Hamiltonian neural network (QGHNN) whose layered circuit is derived by decomposing a time-evolution operator into rotation and CNOT gates, and trains the circuit parameters by gradient descent on the expectation value of the graph-derived Hamiltonian. Experiments on graphs with four to six nodes report that the trained circuit recovers the target adjacency matrix with mean squared error as low as 0.004 and cosine similarity as high as 99.8%, and that these numbers beat the VQE, QAOA, and QNN baselines under the same noise conditions. The intended significance is that graph learning can be carried out on near-term quantum devices through Hamiltonian learning, providing a route toward quantum knowledge graphs and recommendation systems.

What carries the argument

The load-bearing object is the pair consisting of the graph-derived Hamiltonian $H_m$ and the circuit ansatz $U(\theta)$. $H_m$ encodes the graph through the adjacency matrix $A_{ij}$ in front of Pauli string operators $\sigma_x\otimes\sigma_x$, $\sigma_y\otimes\sigma_y$, $\sigma_z\otimes\sigma_z$, so that the graph structure becomes the coupling constants of a spin-lattice system. The circuit $U(\theta)$ is obtained by decomposing $\exp(-iH_c\theta/\hbar)$, with $H_c = \sum_n \sigma_n^y + \sum_n (\frac{1}{2}I - \frac{\pi}{4} \sigma_n^z\otimes\sigma_{n+1}^z) + \sum_n \sigma_n^x$, into the product $R_y(\hbar\theta)\cdot \mathrm{CNOT}\cdot R_z(2\hbar\theta/\pi)\cdot \mathrm{CNOT}\cdot R_x(\hbar\theta)$; this decomposition is what allows the method to turn Hamiltonian learning into a parameterized circuit whose angles are trained by gradient descent. The graph information is injected twice: into the target observable $H_m$ used to define the loss, and into the structure of the layered circuit that produces the intermediate state.

What would settle it

Take the two-qubit case, fix $\theta$ to a generic value, compute $\exp(-iH_c\theta/\hbar)$ directly from Eq. (8), and compute the product $R_y(\hbar\theta)\cdot \mathrm{CNOT}\cdot R_z(2\hbar\theta/\pi)\cdot \mathrm{CNOT}\cdot R_x(\hbar\theta)$ from Eq. (12); any difference beyond machine precision means Eq. (10) does not hold and the circuit does not implement the intended Hamiltonian evolution.

Watch

Extended reading notes

Core claim

The central claim, stated on the paper's own terms, is that QGHL establishes a mapping link between a graph $G=(V,A)$ and the Hamiltonian $H_m = \sum_{i,j} A_{ij}(J_x \sigma_i^x \sigma_j^x + J_y \sigma_i^y \sigma_j^y + J_z \sigma_i^z \sigma_j^z)$ of a topological quantum system, and that QGHNN learns this Hamiltonian by training the parameterized circuit $U(\theta)=R_y(\hbar\theta)\cdot \mathrm{CNOT}\cdot R_z(2\hbar\theta/\pi)\cdot \mathrm{CNOT}\cdot R_x(\hbar\theta)$ to minimize the loss $\mathrm{Loss}(\theta)=\langle\psi_t(\theta)|H_m|\psi_t(\theta)\rangle$. The paper reports that after gradient-descent training, the Hamiltonian distribution of the learned graph approaches the target graph's distribution, with the best result on a four-node graph (MSE $0.004$, cosine similarity $99.8\%$) and slightly lower but still leading accuracy on five- and six-node graphs, which it attributes to noise resilience inherited from the topological system.

Load-bearing premise

The load-bearing premise is that the unitary evolution with the circuit Hamiltonian can be split into separate rotations for each Pauli term; the terms involved in Eq. (8) do not commute, so the split is mathematically unjustified.

Editorial extensions

If this is right

  • For any graph with a known adjacency matrix, QGHL gives a concrete recipe to transcribe the graph into a spin-lattice Hamiltonian and then into a trainable quantum circuit.
  • The reported loss curves converge to about $-1.0$ within 200–300 steps for all three graph sizes, indicating stable training on the tested examples.
  • QGHNN uses fewer trainable parameters than the VQE, QAOA, and QNN baselines while achieving lower error on all four evaluation metrics.
  • The accuracy gaps over baselines persist as the number of qubits increases from 4 to 6, consistent with the paper's claim of noise-resilient graph learning.

Reading between the lines

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

  • Because the claimed circuit derivation depends on the validity of Eq. (10), a direct check of that identity would show whether QGHNN is truly simulating the stated Hamiltonian or simply learning as a generic variational ansatz; the reported numbers would then need reinterpretation.
  • The graph-to-Hamiltonian mapping is independent of the particular circuit ansatz, so the same encoding could in principle be paired with a different, exactly implementable ansatz if the current decomposition fails.
  • A natural next experiment is to test the same training scheme on weighted or directed graphs, where the adjacency entries carry more information than the unweighted examples used here.
  • The method's practical value on near-term hardware will depend on how the circuit depth and measurement cost scale with the number of nodes, since the experiments here are limited to six qubits.
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 manuscript proposes a quantum graph Hamiltonian neural network (QGHNN) built on a quantum graph Hamiltonian learning method (QGHL). QGHL maps a graph G to a Hamiltonian H_m of a topological quantum system, constructs a parameterized quantum circuit from a circuit Hamiltonian H_c by factorizing exp(-iθH_c/ℏ) into rotations and CNOT gates, and then trains the circuit parameters by minimizing the expectation value of H_m. The authors report experiments on the PennyLane platform for 4-, 5-, and 6-qubit graphs and claim that QGHNN outperforms VQE, QAOA, and QNN on MSE, cosine similarity, Frobenius norm, and correlation coefficient, with the best MSE of 0.004 and cosine similarity of 99.8%.

Significance. If the construction were valid, the paper would offer an explicit quantum circuit ansatz for representing graph structure through Hamiltonian learning, with potential relevance to quantum machine learning on NISQ devices. The paper also provides concrete comparison tables and convergence plots, which is useful presentation material. However, the central derivation connecting H_c to the implementable circuit is mathematically incorrect, and the experimental evaluation is circular because the target graph is used both to construct the loss Hamiltonian and to score the learned graph. These are load-bearing issues for the paper's main claims, so the current manuscript is not publishable in its present form.

major comments (4)
  1. [§III-C, Eq. (10)] Equation (10) asserts an exact equality between exp(-iθH_c/ℏ) and a product of exponentials of the individual terms in H_c from Eq. (8). This equality is mathematically false because the operators in H_c do not commute: for each qubit n, [σ_y^n, σ_x^n] = 2iσ_z^n, and the single-qubit σ_y and σ_x terms do not commute with the neighboring σ_z^n⊗σ_z^{n+1} terms. The product of exponentials differs from exp(-iθH_c/ℏ) by commutator corrections of order θ². Therefore the circuit in Eq. (12) does not implement the unitary evolution generated by H_c; at best it is a first-order Trotterization. Since the claimed graph-to-Hamiltonian mapping and the QGHL circuit both rely on Eq. (10), the central quantum construction is invalid as stated.
  2. [§IV-A and §V-A, Eqs. (6) and (13)] The experimental evaluation is circular. The loss function in Eq. (13) is the expectation value of H_m, which is constructed from the target graph G_t via Eq. (6). The same target graph's adjacency matrix is used as the input Hamiltonian, and the learned graph G* is extracted from the optimized quantum state and then compared against that same target graph. There is no training/test split, no held-out graph, and no task in which the model predicts an unseen target. The reported metrics in Tables III–V therefore quantify fitting to the training label, not graph learning or generalization.
  3. [§V-A, Tables III–V] The claimed noise robustness is not supported by the experiments. No noise model, shot count, or hardware backend is described; the tables report single aggregate values without error bars or repeated runs. The abstract and Section V conclude that QGHNN 'reduces the impact of quantum noise' and has 'high robustness ability,' but no noisy simulation or statistical analysis is presented. Additionally, the comparisons to VQE, QAOA, and QNN lack controlled hyperparameter settings and repeated trials, so the statement that QGHNN 'outperforms all assessment metrics' is not substantiated as a fair comparison.
  4. [§IV-B, Algorithm 1] The gradient update step is written incorrectly as ∂Loss(θ_j)/∂θ_j = [∂Loss(θ_j+Δ_j) − ∂Loss(θ_j−Δ_j)]/(2Δ_j), which mixes derivatives with function values. The correct central difference should use Loss(θ_j+Δ_j) and Loss(θ_j−Δ_j), not ∂Loss at those points. Also, the stopping criterion Gap = |H_t − H_m| in §III-A never defines how H_t is computed from |ψ_t(θ)⟩, so the proposed learning procedure is not fully specified.
minor comments (4)
  1. [General notation] There are several typos and inconsistent labels: 'QVE' appears for VQE in Tables VI–VIII and in the text, and the rotation angle notation alternates between R_z(2ℏθ/π) and R_z(2ℏ/π θ). These should be unified.
  2. [§V-A, Fig. 5] The procedure for extracting the learned graph G* from the final quantum state |ψ_out⟩ is not described: it is unclear how measured probabilities are converted back into an adjacency matrix and how normalization by Eq. (15) is inverted.
  3. [§V-C, Tables IX–X] The comparison criteria in Tables IX and X, such as 'Noise Resistant' and 'Implementability on Quantum Computers,' are not defined operationally, so the qualitative superiority claims cannot be assessed.
  4. [Abstract] The abstract claims the method 'shows high robustness ability' and 'can reduce the impact of quantum noise,' but no noisy experiment is reported; the claim should be limited to what the evidence supports.

Circularity Check

2 steps flagged · score 7.0 of 10

QGHNN's reported graph-learning accuracy is a training fit: the loss is the expectation of a Hamiltonian constructed from the target graph, and the same target graph is used as ground truth for the reported metrics.

  1. fitted input called prediction [Section IV-A, Eq. (13); Section III-B, Eq. (6); Section V-A]
    "The loss function Loss(θ) consists of the mapping Hamiltonian Hm designed based on graph G and the vector product of intermediate quantum states |ψt(θ)⟩, as shown in Eq.(13), Loss(θ) =⟨ψt(θ)|Hm|ψt(θ)⟩. The experiment uses MSE, cosine similarity, Frobenius norm, and correlation coefficient to assess the similarity between Fig. (5)(a) and Fig. (5)(g)."

    Hm is explicitly defined from the target adjacency matrix Aij in Eq. (6). The loss being minimized (Eq. 13) is the expectation value of that same Hm under the variational state. Gradient descent therefore drives the state directly toward an eigenstate (typically the ground state) of a Hamiltonian whose construction already contains the target graph. The final graph G* is read out from that optimized state and then compared with the same target graph. The reported MSE of 0.004 and cosine similarity of 99.8% are reconstruction accuracies on the training label, not independent predictions; they are forced by the optimization objective rather than being evidence of generalization.

  2. self definitional [Section III-A]
    "The gap between the Hamiltonian Ht generated during the system evolution and the mapping Hamiltonian Hm determines whether the quantum system evolved through the quantum circuit is the final state."

    Success of QGHL is defined as closeness between the evolved Hamiltonian Ht and Hm, while Hm is itself defined from the target graph via Eq. (6). Thus the stopping criterion, the learned output, and the evaluation target all reduce to the same input graph. The claimed 'mapping link between graph and Hamiltonian' is stipulated by construction in Eq. (6), not derived or independently validated.

full rationale

The central circularity is that the target graph enters the loss function directly through Hm, and the same graph is then used as the ground truth for the reported metrics. Equation (6) builds Hm from the adjacency matrix Aij; Equation (13) sets Loss(θ) = ⟨ψt(θ)|Hm|ψt(θ)⟩; Section V-A evaluates the learned graph against the same Aij. The near-perfect MSE and cosine similarity therefore measure how well the parameterized circuit can fit the training label, not whether QGHNN predicts unseen graph structure. This is the fitted-input-called-prediction pattern. A second, related self-definitional aspect is that QGHL declares success when Ht approaches Hm, with Hm already encoding the target graph. The invalid factorization in Eq. (10) (non-commuting Pauli terms exponentiated separately) is a separate correctness flaw, not a circularity; it does not affect this score. The self-citation to Shi et al. [18] supplies generic QHL circuit techniques but is not the load-bearing source of the circularity, so it is not separately penalized. Overall, the central claim that QGHNN 'learns' the graph reduces to optimizing an objective defined by that same graph, warranting a score of 7.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on hand-chosen coupling constants and circuit coefficients, a false commutativity assumption behind Eq. (10), and an unspecified measurement-to-graph extraction. These inputs are not independently justified by the paper.

free parameters (2)
  • Jx, Jy, Jz coupling constants = not specified
    Appear in the mapping Hamiltonian Hm (Eq. 6) and determine the loss landscape; their values are never given, and the learned graph depends on them.
  • Hc coefficients (1/2 and pi/4) = 1/2 and pi/4
    Coefficients in the circuit Hamiltonian Hc (Eq. 8) are chosen ad hoc to enable a CNOT-Rz-CNOT decomposition; no derivation is provided.
assumptions (4)
  • ad hoc to paper The Pauli terms in Hc commute, so the exponential factorization in Eq. (10) is valid.
    The factorization assumes pairwise commutation of sigma_y, sigma_x, and sigma_z⊗sigma_z terms, which is false. This load-bearing error appears at Eq. (10) and invalidates the derived circuit.
  • domain assumption Minimizing the expectation of Hm (built from the target graph's adjacency matrix) yields a quantum state that encodes the target graph.
    The loss function (Eq. 13) uses Hm derived from the target graph; the paper assumes the ground state of this Hamiltonian represents the graph, without proof, and evaluates on the same graph.
  • domain assumption The output graph G* can be extracted from the final quantum state by measurement.
    The paper does not specify how the final quantum state is measured to produce the adjacency matrix of G*, yet the numerical metrics depend on this extraction.
  • domain assumption A noiseless PennyLane simulator represents a noisy intermediate-scale quantum computer.
    Experiments are run on a default simulator with no noise model, but the paper claims noise resilience on NISQ devices.

how reviews work

0 comments
Cite this review

Pith. "Pith review of QGHNN: A quantum graph Hamiltonian neural network." pith.science (2026). https://pith.science/paper/EXWM7ETR

@misc{pith2026250107986,
  author       = {Pith},
  title        = {Pith review of: QGHNN: A quantum graph Hamiltonian neural network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EXWM7ETR}},
  note         = {Machine review of arXiv:2501.07986}
}
abstract

Representing and learning from graphs is essential for developing effective machine learning models tailored to non-Euclidean data. While Graph Neural Networks (GNNs) strive to address the challenges posed by complex, high-dimensional graph data, Quantum Neural Networks (QNNs) present a compelling alternative due to their potential for quantum parallelism. However, much of the current QNN research tends to overlook the vital connection between quantum state encoding and graph structures, which limits the full exploitation of quantum computational advantages. To address these challenges, this paper introduces a quantum graph Hamiltonian neural network (QGHNN) to enhance graph representation and learning on noisy intermediate-scale quantum computers. Concretely, a quantum graph Hamiltonian learning method (QGHL) is first created by mapping graphs to the Hamiltonian of the topological quantum system. Then, QGHNN based on QGHL is presented, which trains parameters by minimizing the loss function and uses the gradient descent method to learn the graph. Experiments on the PennyLane quantum platform reveal that QGHNN outperforms all assessment metrics, achieving the lowest mean squared error of \textbf{$0.004$} and the maximum cosine similarity of \textbf{$99.8\%$}, which shows that QGHNN not only excels in representing and learning graph information, but it also has high robustness ability. QGHNN can reduce the impact of quantum noise and has significant potential application in future research of quantum knowledge graphs and recommendation systems.

Figures

Figures reproduced from arXiv: 2501.07986 by the authors.

Figure 1
Figure 1. Graph Neural Network. The input graph data can be trained by GNN [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Framework of quantum graph Hamiltonian learning. QGHL is introduced as an approach for linking graphs with the Hamiltonian of a topological [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The example of the quantum circuits in QGHL. Quantum circuits of QGHL are made up of a 3-qubit quantum system with [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Quantum graph Hamiltonian neural network. The QGHNN model [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Experiment setting of QGHNN. QGHNN learns the target graph [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Experimental comparison results of QGHNN and other different [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Experiment result of 4-Qubits based on QGHNN model. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Experiment result of loss function. QVE QAOA QNN QGHNN 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 Parameters Box Plot of Parameters for Different Models [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Results about the use of parameters between QGHNN and other [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 34 canonical work pages

  1. [1]

    Training a quantum annealing based restricted boltzmann machine on cybersecurity data,

    V . Dixit, R. Selvarajan, T. Aldwairi, Y . Koshka, M. A. Novotny, T. S. Humble, M. A. Alam, and S. Kais, “Training a quantum annealing based restricted boltzmann machine on cybersecurity data,” IEEE Transactions on Emerging Topics in Computational Intelligence , vol. 6, no. 3, pp. 417–428, 2022

    work page 2022

  2. [2]

    Quantum reachability games,

    W. Liu, Z. Li, and Y . Li, “Quantum reachability games,” IEEE Trans- actions on Emerging Topics in Computational Intelligence , pp. 1–15, 2024

    work page 2024

  3. [3]

    Quantum advantage in learning from experiments,

    H.-Y . Huang, M. Broughton, J. Cotler, S. Chen, J. Li, M. Mohseni, H. Neven, R. Babbush, R. Kueng, J. Preskill et al., “Quantum advantage in learning from experiments,” Science, vol. 376, no. 6598, pp. 1182– 1186, 2022

    2022

  4. [4]

    Pretrained quantum- inspired deep neural network for natural language processing,

    J. Shi, T. Chen, W. Lai, S. Zhang, and X. Li, “Pretrained quantum- inspired deep neural network for natural language processing,” IEEE Transactions on Cybernetics , 2024

    work page 2024

  5. [5]

    Event-based deep reinforcement learning for quan- tum control,

    H. Yu and X. Zhao, “Event-based deep reinforcement learning for quan- tum control,” IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 8, no. 1, pp. 548–562, 2024

    work page 2024

  6. [6]

    A quantum spatial graph convolutional neural network model on quantum circuits,

    J. Zheng, Q. Gao, M. Ogorzałek, J. Lü, and Y . Deng, “A quantum spatial graph convolutional neural network model on quantum circuits,” IEEE Transactions on Neural Networks and Learning Systems , 2024

    work page 2024

  7. [7]

    Graph-based dissimilarity measurement for cluster analysis of any-type-attributed data,

    Y . Zhang and Y .-M. Cheung, “Graph-based dissimilarity measurement for cluster analysis of any-type-attributed data,” IEEE Transactions on Neural Networks and Learning Systems , vol. 34, no. 9, pp. 6530–6544, 2022

    work page 2022

  8. [8]

    Strong social graph based trust-oriented graph pattern matching with multiple constraints,

    G. Liu, Y . Wang, B. Zheng, Z. Li, and K. Zheng, “Strong social graph based trust-oriented graph pattern matching with multiple constraints,” IEEE Transactions on Emerging Topics in Computational Intelligence , vol. 4, no. 5, pp. 675–685, 2020

    work page 2020

Show all 36 references
  1. [9]

    Graph structure enhanced pre-training language model for knowledge graph completion,

    H. Zhu, D. Xu, Y . Huang, Z. Jin, W. Ding, J. Tong, and G. Chong, “Graph structure enhanced pre-training language model for knowledge graph completion,” IEEE Transactions on Emerging Topics in Compu- tational Intelligence, vol. 8, no. 4, pp. 2697–2708, 2024

    2024

  2. [10]

    A branch and bound irredundant graph algorithm for large-scale mlcs problems,

    C. Wang, Y . Wang, and Y . Cheung, “A branch and bound irredundant graph algorithm for large-scale mlcs problems,” Pattern Recognition , vol. 119, p. 108059, 2021

    2021

  3. [11]

    Qksan: A quantum kernel self-attention network,

    R.-X. Zhao, J. Shi, and X. Li, “Qksan: A quantum kernel self-attention network,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 2024

    2024

  4. [12]

    Quantum machine learning,

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, “Quantum machine learning,” Nature, vol. 549, no. 7671, pp. 195–202, 2017. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 12 TABLE X ADVANTAGES OF QGHNN Indicators Models Du’s [34] Guerresch...

    2017

  5. [13]

    Learning graph convolutional networks based on quantum vertex information propagation,

    L. Bai, Y . Jiao, L. Cui, L. Rossi, Y . Wang, S. Y . Philip, and E. R. Hancock, “Learning graph convolutional networks based on quantum vertex information propagation,” IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 2, pp. 1747–1760, 2021

    2021

  6. [14]

    Quantum- based subgraph convolutional neural networks,

    Z. Zhang, D. Chen, J. Wang, L. Bai, and E. R. Hancock, “Quantum- based subgraph convolutional neural networks,” Pattern Recognition , vol. 88, pp. 38–49, 2019

    2019

  7. [15]

    Quantum walk neural networks for graph-structured data,

    S. Dernbach, A. Mohseni-Kabir, S. Pal, and D. Towsley, “Quantum walk neural networks for graph-structured data,” in Complex Networks and Their Applications VII , L. M. Aiello, C. Cherifi, H. Cherifi, R. Lam- biotte, P. Lió, and L. M. Rocha, Eds. Cham: Springer International Pu...

    2019

  8. [16]

    Quantum hamiltonian learning using imperfect quantum resources,

    N. Wiebe, C. Granade, C. Ferrie, and D. Cory, “Quantum hamiltonian learning using imperfect quantum resources,”Physical Review A, vol. 89, no. 4, p. 042314, 2014

    2014

  9. [17]

    Practical hamiltonian learning with unitary dynamics and gibbs states,

    A. Gu, L. Cincio, and P. J. Coles, “Practical hamiltonian learning with unitary dynamics and gibbs states,” Nature Communications , vol. 15, no. 1, p. 312, 2024

    2024

  10. [18]

    Parameterized hamil- tonian learning with quantum circuit,

    J. Shi, W. Wang, X. Lou, S. Zhang, and X. Li, “Parameterized hamil- tonian learning with quantum circuit,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 5, pp. 6086–6095, 2022

    2022

  11. [19]

    Quantum-probabilistic hamiltonian learning for generative modeling and anomaly detection,

    J. Y . Araz and M. Spannowsky, “Quantum-probabilistic hamiltonian learning for generative modeling and anomaly detection,” Physical Review A, vol. 108, no. 6, p. 062422, 2023

    2023

  12. [20]

    Adversarial hamiltonian learning of quantum dots in a minimal kitaev chain,

    R. Koch, D. Van Driel, A. Bordin, J. L. Lado, and E. Greplova, “Adversarial hamiltonian learning of quantum dots in a minimal kitaev chain,” Physical Review A , vol. 20, no. 4, p. 044081, 2023

    2023

  13. [21]

    Quantum simulation of open quantum systems using a unitary decomposition of operators,

    A. W. Schlimgen, K. Head-Marsden, L. M. Sager, P. Narang, and D. A. Mazziotti, “Quantum simulation of open quantum systems using a unitary decomposition of operators,” Physical Review L, vol. 127, no. 27, p. 270503, 2021

    2021

  14. [22]

    Learning quantum systems,

    V . Gebhart, R. Santagati, A. A. Gentile, E. M. Gauger, D. Craig, N. Ares, L. Banchi, F. Marquardt, L. Pezzè, and C. Bonato, “Learning quantum systems,” Nature Reviews Physics , vol. 5, no. 3, pp. 141–156, 2023

    2023

  15. [24]

    Mdg: A multi-task dynamic graph generation framework for multivariate time series forecasting,

    L. Huang, J. Yuan, S. Chen, and X. Li, “Mdg: A multi-task dynamic graph generation framework for multivariate time series forecasting,” IEEE Transactions on Emerging Topics in Computational Intelligence , vol. 8, no. 2, pp. 1337–1349, 2024

    2024

  16. [25]

    Kgcna: Knowledge graph collaborative neighbor awareness network for recommendation,

    G. He, Z. Zhang, H. Wu, S. Luo, and Y . Liu, “Kgcna: Knowledge graph collaborative neighbor awareness network for recommendation,” IEEE Transactions on Emerging Topics in Computational Intelligence , vol. 8, no. 4, pp. 2736–2748, 2024

    2024

  17. [26]

    Robust rank-constrained sparse learning: A graph-based framework for single view and multiview clustering,

    Q. Wang, R. Liu, M. Chen, and X. Li, “Robust rank-constrained sparse learning: A graph-based framework for single view and multiview clustering,” IEEE Transactions on Cybernetics , vol. 52, no. 10, pp. 10 228–10 239, 2021

    2021

  18. [27]

    Learn- ing relationship-enhanced semantic graph for fine-grained image–text matching,

    X. Liu, Y . He, Y .-M. Cheung, X. Xu, and N. Wang, “Learn- ing relationship-enhanced semantic graph for fine-grained image–text matching,” IEEE Transactions on Cybernetics , vol. 54, no. 2, pp. 948– 961, 2022

    2022

  19. [28]

    Reconstructing heterogeneous networks via compressive sensing and clustering,

    Y . Zhang, C. Yang, K. Huang, M. Jusup, Z. Wang, and X. Li, “Reconstructing heterogeneous networks via compressive sensing and clustering,” IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 5, no. 6, pp. 920–930, 2020

    2020

  20. [29]

    Relation- aggregated cross-graph correlation learning for fine-grained image–text retrieval,

    S.-J. Peng, Y . He, X. Liu, Y .-m. Cheung, X. Xu, and Z. Cui, “Relation- aggregated cross-graph correlation learning for fine-grained image–text retrieval,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 2, pp. 2194–2207, 2022

    2022

  21. [30]

    Sagn: Semantic-aware graph network for remote sensing scene classification,

    Y . Yang, X. Tang, Y .-M. Cheung, X. Zhang, and L. Jiao, “Sagn: Semantic-aware graph network for remote sensing scene classification,” IEEE Transactions on Image Processing, vol. 32, pp. 1011–1025, 2023

    2023

  22. [31]

    Reinforcement learning and transformer for fast magnetic resonance imaging scan,

    Y . Liu, Y . Pang, R. Jin, Y . Hou, and X. Li, “Reinforcement learning and transformer for fast magnetic resonance imaging scan,” IEEE Transactions on Emerging Topics in Computational Intelligence , 2024

    2024

  23. [32]

    Decouple graph neural networks: Train multiple simple gnns simultaneously instead of one,

    H. Zhang, Y . Zhu, and X. Li, “Decouple graph neural networks: Train multiple simple gnns simultaneously instead of one,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2024

    2024

  24. [33]

    Qgrl: Quaternion graph representation learning for heterogeneous feature data clustering,

    J. Chen, Y . Ji, R. Zou, Y . Zhang, and Y .-m. Cheung, “Qgrl: Quaternion graph representation learning for heterogeneous feature data clustering,” in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2024, pp. 297–306

    2024

  25. [34]

    Quantum circuit architecture search for variational quantum algorithms,

    Y . Du, T. Huang, S. You, M.-H. Hsieh, and D. Tao, “Quantum circuit architecture search for variational quantum algorithms,” npj Quantum Information, vol. 8, no. 1, p. 62, 2022

    2022

  26. [35]

    Qaoa for max-cut requires hundreds of qubits for quantum speed-up,

    G. G. Guerreschi and A. Y . Matsuura, “Qaoa for max-cut requires hundreds of qubits for quantum speed-up,” Scientific reports , vol. 9, no. 1, p. 6903, 2019

    2019

  27. [36]

    Experimental quantum hamiltonian learning using a silicon photonic chip and a nitrogen-vacancy electron spin in diamond,

    S. Paesani, J. Wang, R. Santagati, S. Knauer, A. A. Gentile, N. Wiebe, M. Petruzzella, A. Laing, J. G. Rarity, J. L. O’Brienet al., “Experimental quantum hamiltonian learning using a silicon photonic chip and a nitrogen-vacancy electron spin in diamond,” in European Quantum El...

    2017

  28. [37]

    A hybrid quantum-classical hamiltonian learning algorithm,

    Y . Wang, G. Li, and X. Wang, “A hybrid quantum-classical hamiltonian learning algorithm,” Science China-Information Sciences, vol. 66, no. 2, 2023

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.