Pith. sign in

REVIEW 5 major objections 5 minor 30 references

Collaborative Filtering using Variational Quantum Hopfield Associative Memory

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A hybrid quantum Hopfield associative memory pipeline performs collaborative filtering on MovieLens 1M with ROC 0.98 and F1 0.88 in ideal simulation, and degrades gracefully under hardware-like noise.

desk verdict A plausible hybrid quantum-classical recommendation pipeline whose headline numbers are unverifiable: the QHAM circuit is never specified, and the text does not rule out a train/test leakage through k-means clustering. read the letter →

arxiv 2508.14906 v1 pith:XTKQRIKS submitted 2025-08-12 cs.IR cs.AIcs.ETcs.LG

Amir Kermanshahani , Ebrahim Ardeshir-Larijani , Rakesh Saini , Saif Al-Kuwari This is my paper · ORCID
classification cs.IRcs.AIcs.ETcs.LG
keywords QuantumMachineLearningHopfieldAssociativeMemoryCollaborativeFilteringRecommendationSystemHybridQuantum-ClassicalModelNoisyComputingMovieLens1M
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

The paper claims that a recommender system can be built on a quantum associative memory instead of on matrix factorization, and that the resulting hybrid classical-quantum pipeline is competitive with classical deep-learning recommenders on the MovieLens 1M benchmark. The pipeline compresses user rating vectors with an autoencoder, clusters users into archetypes with k-means, stores those archetypes as amplitude-encoded states in a variational quantum Hopfield associative memory, and retrieves the best-matching archetype for a new user. In ideal simulation the authors report ROC 0.9795, accuracy 0.8841, and F1-score 0.8786 on the test set; under a custom noise model with bit-flip and readout errors, the numbers become ROC 0.9177, accuracy 0.8013, and F1-score 0.7866. They also claim that updating a single randomly targeted qubit reduces qubit overhead relative to earlier QHAM implementations. If these results hold, quantum associative memories could serve as the retrieval core of practical recommendation systems on near-term noisy hardware.

What carries the argument

The load-bearing component is the variational Quantum Hopfield Associative Memory (QHAM), a trainable quantum circuit that stores user-archetype patterns as quantum states and retrieves the stored state closest to an input. Each archetype's real-valued vector is encoded into qubit amplitudes by a cosine/sine mapping, the full state is prepared with a standard uniformly-controlled-rotation procedure, and the circuit parameters are trained end-to-end with mean squared error loss. A classical autoencoder reduces each user's sparse rating vector to a compact latent code, k-means supplies the archetypes, and a final softmax layer converts the retrieved quantum state into a predicted user category. The paper's claimed novelty is updating only one randomly targeted qubit during training, which is said to cut qubit overhead compared with earlier QHAM implementations.

What would settle it

Re-run the pipeline with k-means clustering applied only to the training subset before any test user is observed, then recompute ROC, accuracy, and F1 on the untouched test split; if those metrics fall substantially below the reported 0.9795, 0.8841, and 0.8786, the central generalization claim is falsified by the data-split order alone.

Watch

Extended reading notes

Core claim

The central claim is that a hybrid architecture consisting of a deep autoencoder, k-means clustering, and a variational Quantum Hopfield Associative Memory (QHAM) can perform collaborative filtering at a level comparable to state-of-the-art classical neural recommenders, while training in only 35 epochs. User rating vectors are compressed by the encoder, archetype patterns are extracted by k-means and converted to polar form, and the QHAM stores them as amplitude-encoded quantum states. Retrieval of the archetype most similar to an encoded user is trained end-to-end under mean squared error loss. The authors report that the model achieves ROC 0.9795, accuracy 0.8841, and F1-score 0.8786 on the MovieLens 1M test set in an ideal simulator; under a simulator-based noise model incorporating bit-flip and readout errors at hardware-like probabilities, it achieves ROC 0.9177, accuracy 0.8013, and F1-score 0.7866. They further report that their one-random-qubit updating scheme reduces qubit overhead relative to prior QHAM designs.

Load-bearing premise

The reported test scores assume the user archetypes stored in the quantum memory were formed from training data only, because the paper does not state whether k-means clustering is applied before or after the train/test split, and applying it to the full dataset would make the test metrics in-sample.

Editorial extensions

If this is right

  • Collaborative filtering can be implemented with a quantum associative memory as the retrieval core, without explicit matrix factorization.
  • The reported noise robustness (ROC falling from 0.9795 to 0.9177 under the custom error model) indicates the architecture may tolerate realistic hardware errors without catastrophic failure.
  • Training for 35 epochs to reach roughly 88% accuracy and 0.88 F1 suggests faster convergence than the 80-epoch classical baseline the paper compares against.
  • The model's performance on a standard benchmark positions it as a candidate for deployment on near-term quantum devices, though only as a simulated demonstration so far.

Reading between the lines

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

  • If the one-qubit-update strategy generalizes, it may reduce the gate depth of other variational quantum memory schemes, making them cheaper to run on real hardware.
  • Because the paper does not specify whether k-means clustering is applied before or after the train/test split, a re-run with clustering restricted to the training split would clarify whether the reported test metrics are truly out-of-sample.
  • A natural extension is to ablate the quantum memory by replacing it with a classical nearest-neighbor index over the same archetypes; the difference would isolate the empirical contribution of the quantum retrieval step.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a hybrid recommendation system for MovieLens 1M that combines a deep autoencoder, k-means clustering of users into archetypes, and a variational quantum Hopfield associative memory (QHAM). The encoder compresses user rating vectors; k-means produces archetype labels; the QHAM is trained to store and retrieve these archetypes; a softmax layer classifies the user. The authors report ideal simulation ROC 0.9795, accuracy 0.8841, and F1 0.8786, and noisy simulation ROC 0.9177, accuracy 0.8013, and F1 0.7866 after 35 epochs, and claim these results are comparable or superior to classical collaborative filtering baselines.

Significance. If the results are reproducible, this would be one of the first applications of quantum associative memory to collaborative filtering at the scale of MovieLens 1M and would provide evidence that variational quantum models can tolerate realistic noise. The claimed qubit-overhead reduction by updating one random qubit is also potentially interesting. However, the manuscript currently does not supply enough detail to verify these claims: the QHAM circuit is not specified, the preprocessing order is ambiguous, and no code or statistical uncertainty is given. The conceptual contribution is interesting but not yet substantiated.

major comments (5)
  1. [§2.1.2] The variational QHAM circuit is not described. The text only gives the amplitude encoding of Eq. (1) and refers to Mottonen state preparation; it does not specify the ansatz, the number of qubits, the number of layers, the parameterized gates, the update rule, or the objective function used to train the QHAM. The abstract's claim of 'updating only one random targeted qubit' appears nowhere in the methods. Without this information, the reported performance in Table 1 cannot be reproduced or evaluated.
  2. [§2.2.1] The data-preprocessing order is ambiguous and potentially circular. The sentence 'We used all the users' data for the simulation on the ideal and noisy environments' appears immediately before the train/validation/test split description and the k-means clustering description. If k-means archetypes or autoencoder training use all users before the split, then the test labels are derived from test inputs and the metrics in Table 1 are in-sample. The authors must state explicitly that all clustering and encoder fitting are performed on the training split only, and if that is not the case, all reported results must be recomputed on a properly split pipeline.
  3. [§3.4] The comparison with Bobadilla et al. [26] misstates the reported accuracy. The text says 'our model demonstrated the same accuracy' while the cited paper reports about 90% and Table 1 reports 88.41% in the ideal case and 80.13% in the noisy case. The claim should be corrected, and any performance comparison should include confidence intervals derived from multiple random seeds.
  4. [§2.2.4] The noise model is underspecified. The text references [20] and mentions bit-flip and readout errors 'with the same probabilities as in real quantum hardware', but no error probabilities, hardware target, or circuit-level noise model are given. The noisy results cannot be reproduced without these parameters.
  5. [§2.2.1] The number of archetype clusters K is never stated, even though K defines the classification task and the size of the softmax output. Without K, the reported accuracy and F1 are not meaningful, and the reader cannot tell what fraction of the reported error comes from the quantum memory versus the clustering choice.
minor comments (5)
  1. [§2.2.4] The text says the noise model 'identifies three main sources of errors' but never lists them; add the enumeration.
  2. [§2.1.2] Equation (1) should clarify the allowable range of x_i after min-max normalization and Tanh, since the encoding assumes x_i ∈ [-1,1].
  3. [§2.2.1] The description 'split the test-validation subset into validation and test subsets using the same ratio' is ambiguous; specify the exact split fractions.
  4. [§3.4] Informal phrases such as 'way below' and 'can be considered random' should be replaced with exact numeric comparisons and statistical tests.
  5. [General] The paper would benefit from a data and code availability statement; none is provided.

Circularity Check

1 steps flagged · score 6.0 of 10

Archetype labels are defined by the same K-Means centers stored in the QHAM, so the reported classification metrics are self-consistency scores.

  1. self definitional [Section 2.2.1 and Section 4 (Conclusion), with the pipeline diagram in Fig. 5]
    "Section 2.2.1: 'The k-means clustering algorithm segments users into archetypes based on their preferences.' Section 4: 'we used K-Means clustering to classify raw user vectors and extracted the cluster centers, which we then encoded and polarized to function as user archetypes. Next, we configured the quantum Hopfield associative memory to preserve the polarized patterns and retrieve them when new user vectors are received. Lastly, we added a Dense layer with a SoftMax activation function as the post-processing unit to determine the user category.'"

    The prediction target (the user category) is defined by the K-Means cluster centers, and those same centers are stored in the QHAM as the patterns to be retrieved. The softmax output is therefore trained to reproduce the same nearest-center assignment that the associative memory is designed to perform. The reported ROC, accuracy, and F1 consequently measure how often the pipeline regenerates its own clustering labels, not how well it predicts held-out ratings or external user-item interactions. The Section 3.4 comparison with classical collaborative filtering baselines is thus not on equal terms: the baseline metrics refer to predicting user-item interactions, while the hybrid model is evaluated on self-generated archetype labels.

full rationale

The paper contains no self-citations and its quantum state preparation follows standard external references (Mottonen et al. [17], Miller and Mukhopadhyay [16]), so self-citation is not a factor. The central circularity is in the evaluation target: user archetypes are produced by K-Means clustering, then the same archetypes are loaded into QHAM, and the softmax head predicts the user category. This makes the classification metrics a self-consistency check rather than an external prediction. There is also an unresolved split-order ambiguity in Section 2.2.1 ('We used all the users’ data for the simulation' followed by the train/test split and then the K-Means paragraph), which would make the test metrics in-sample if clustering is performed before the split; however, even with a clean split, the target definition remains circular because the ground-truth labels are the model's own clustering output. The noisy-environment degradation (ROC 0.9795 to 0.9177) shows that the pipeline is not purely tautological, which is why the score is 6 rather than higher.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper's central contribution rests on several tunable settings (cluster count, qubit count, noise rates) that are not fully reported, and on the unvalidated assumption that a custom noise model and amplitude embedding are appropriate.

free parameters (3)
  • K (number of archetype clusters) = not reported
    K-Means cluster count is chosen by experimental analysis but the value and selection criterion are not given.
  • Number of qubits in QHAM = not reported
    Selected based on experimental analysis of pattern distinguishability, per Section 2.1.1; no exact value or search procedure given.
  • Noise model error rates = not reported
    Bit-flip and readout error probabilities are asserted to match real hardware (Section 2.2.4), but the specific rates and target hardware are not stated.
assumptions (3)
  • standard math Amplitude embedding via Mottonen state preparation encodes user patterns faithfully
    The paper relies on the Mottonen algorithm [17] to prepare quantum states from classical vectors, a known but non-trivial circuit construction whose resource scaling is not discussed.
  • domain assumption The custom Qiskit AER noise model captures relevant real-hardware errors
    The noisy-environment results depend on this assumption; the paper does not validate the noise model against a specific device.
  • domain assumption K-Means archetypes are meaningful user categories for collaborative filtering
    The whole classification target is defined by k-means output; the paper does not compare to other user segmentation schemes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Collaborative Filtering using Variational Quantum Hopfield Associative Memory." pith.science (2026). https://pith.science/paper/XTKQRIKS

@misc{pith2026250814906,
  author       = {Pith},
  title        = {Pith review of: Collaborative Filtering using Variational Quantum Hopfield Associative Memory},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XTKQRIKS}},
  note         = {Machine review of arXiv:2508.14906}
}
read the original abstract

Quantum computing, with its ability to do exponentially faster computation compared to classical systems, has found novel applications in various fields such as machine learning and recommendation systems. Quantum Machine Learning (QML), which integrates quantum computing with machine learning techniques, presents powerful new tools for data processing and pattern recognition. This paper proposes a hybrid recommendation system that combines Quantum Hopfield Associative Memory (QHAM) with deep neural networks to improve the extraction and classification on the MovieLens 1M dataset. User archetypes are clustered into multiple unique groups using the K-Means algorithm and converted into polar patterns through the encoder's activation function. These polar patterns are then integrated into the variational QHAM-based hybrid recommendation model. The system was trained using the MSE loss over 35 epochs in an ideal environment, achieving an ROC value of 0.9795, an accuracy of 0.8841, and an F-1 Score of 0.8786. Trained with the same number of epochs in a noisy environment using a custom Qiskit AER noise model incorporating bit-flip and readout errors with the same probabilities as in real quantum hardware, it achieves an ROC of 0.9177, an accuracy of 0.8013, and an F-1 Score equal to 0.7866, demonstrating consistent performance. Additionally, we were able to optimize the qubit overhead present in previous QHAM architectures by efficiently updating only one random targeted qubit. This research presents a novel framework that combines variational quantum computing with deep learning, capable of dealing with real-world datasets with comparable performance compared to purely classical counterparts. Additionally, the model can perform similarly well in noisy configurations, showcasing a steady performance and proposing a promising direction for future usage in recommendation systems.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 9 canonical work pages

  1. [26]

    International Journal of Interactive Multimedia and Artificial Intelligence 6(1), 68–77 (2020) https://doi.org/10.9781/ijimai.2020.02.006

    Bobadilla, J., Ortega, F., Guti´ errez, A., Alonso, S.: Classification-based deep neural network architecture for collaborative filtering recommender systems. International Journal of Interactive Multimedia and Artificial Intelligence 6(1), 68–77 (2020) https://doi.org/10.9781/ijimai.2020.02.006

    work page doi:10.9781/ijimai.2020.02.006 2020

  2. [20]

    Quantum Information Process- ing 23(5) (2024) https://doi.org/10.1007/s11128-024-04384-z

    Aseguinolaza, U., Sobrino, N., Sobrino, G., Jornet-Somoza, J., Borge, J.: Error estimation in current noisy quantum computers. Quantum Information Process- ing 23(5) (2024) https://doi.org/10.1007/s11128-024-04384-z

    doi:10.1007/s11128-024-04384-z 2024

  3. [1]

    Nature 549(7671), 195–202 (2017) https://doi.org/ 10.1038/nature23474

    Biamonte, J., Wittek, P., Pancotti, N., Rebentrost, P., Wiebe, N., Lloyd, S.: Quantum machine learning. Nature 549(7671), 195–202 (2017) https://doi.org/ 10.1038/nature23474

    doi:10.1038/nature23474 2017

  4. [2]

    Nature Communications 5(1) (2014) https://doi.org/ 10.1038/ncomms5213

    Peruzzo, A., McClean, J., Shadbolt, P., Yung, M.-H., Zhou, X.-Q., Love, P.J., Aspuru-Guzik, A., O’Brien, J.L.: A variational eigenvalue solver on a pho- tonic quantum processor. Nature Communications 5(1) (2014) https://doi.org/ 10.1038/ncomms5213

    doi:10.1038/ncomms5213 2014

  5. [3]

    https://arxiv.org/abs/1411.4028

    Farhi, E., Goldstone, J., Gutmann, S.: A Quantum Approximate Optimization Algorithm (2014). https://arxiv.org/abs/1411.4028

    arXiv 2014

  6. [4]

    Nature 414(6862), 413–418 (2001) https://doi.org/10.1038/35106500

    Duan, L.-M., Lukin, M.D., Cirac, J.I., Zoller, P.: Long-distance quantum com- munication with atomic ensembles and linear optics. Nature 414(6862), 413–418 (2001) https://doi.org/10.1038/35106500

    doi:10.1038/35106500 2001

  7. [5]

    Physics Reports 831, 1–57 (2019) https: //doi.org/10.1016/j.physrep.2019.10.001

    Schlosshauer, M.: Quantum decoherence. Physics Reports 831, 1–57 (2019) https: //doi.org/10.1016/j.physrep.2019.10.001

    doi:10.1016/j.physrep.2019.10.001 2019

  8. [6]

    Journal of Modern Optics 60(18), 1519–1537 (2013) https://doi.org/10.1080/09500340.2013.856482

    Bussi` eres, F., Sangouard, N., Afzelius, M., Riedmatten, H., Simon, C., Tittel, W.: Prospective applications of optical quantum memories. Journal of Modern Optics 60(18), 1519–1537 (2013) https://doi.org/10.1080/09500340.2013.856482

    arXiv 2013

Show all 30 references
  1. [7]

    Proceedings of the National Academy of Sciences of the United States of America 79, 2554–8 (1982) https://doi.org/10.1073/pnas.79.8

    Hopfield, J.: Neural networks and physical systems with emergent collective com- putational abilities. Proceedings of the National Academy of Sciences of the United States of America 79, 2554–8 (1982) https://doi.org/10.1073/pnas.79.8. 2554

    1982 doi

  2. [8]

    https://arxiv

    Ventura, D., Martinez, T.: Quantum Associative Memory (1998). https://arxiv. org/abs/quant-ph/9807053 17

    1998 arXiv

  3. [9]

    Quantum 2, 79 (2018) https://doi.org/10.22331/q-2018-08-06-79

    Preskill, J.: Quantum computing in the nisq era and beyond. Quantum 2, 79 (2018) https://doi.org/10.22331/q-2018-08-06-79

    2018 doi

  4. [10]

    248–255 (2009)

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., Li, F.-F.: Imagenet: a large- scale hierarchical image database, pp. 248–255 (2009). https://doi.org/10.1109/ CVPR.2009.5206848

    2009

  5. [11]

    Quantum 4, 340 (2020) https://doi

    Mari, A., Bromley, T.R., Izaac, J., Schuld, M., Killoran, N.: Transfer learning in hybrid classical-quantum neural networks. Quantum 4, 340 (2020) https://doi. org/10.22331/q-2020-10-09-340

    2020 doi

  6. [12]

    International Journal of Theoretical Physics 58, 2235–2243 (2019)

    Wang, X., Wang, R., Li, D., Adu-Gyamfi, D., Zhu, Y.: Qcf: quantum collaborative filtering recommendation algorithm. International Journal of Theoretical Physics 58, 2235–2243 (2019)

    2019

  7. [13]

    Quantum Information Pro- cessing 20(5), 1–32 (2021) https://doi.org/10.1007/s11128-021-03131-y

    Wang, X., Gu, L., Lee, H., Zhang, G.: Quantum context-aware recommendation systems based on tensor singular value decomposition. Quantum Information Pro- cessing 20(5), 1–32 (2021) https://doi.org/10.1007/s11128-021-03131-y . Funding Information: This research is supported in ...

    2021 doi

  8. [14]

    https://arxiv.org/abs/2301

    Brahmachari, S., Lumbreras, J., Tomamichel, M.: Quantum contextual bandits and recommender systems for quantum data (2023). https://arxiv.org/abs/2301. 13524

    2023

  9. [15]

    New Journal of Physics18(2), 023023 (2016) https://doi.org/10.1088/1367-2630/18/2/023023

    McClean, J.R., Romero, J., Babbush, R., Aspuru-Guzik, A.: The theory of varia- tional hybrid quantum-classical algorithms. New Journal of Physics18(2), 023023 (2016) https://doi.org/10.1088/1367-2630/18/2/023023

    2016 doi

  10. [16]

    Scientific Reports 11(1) (2021) https: //doi.org/10.1038/s41598-021-02866-z

    Miller, N.E., Mukhopadhyay, S.: A quantum hopfield associative memory imple- mented on an actual quantum processor. Scientific Reports 11(1) (2021) https: //doi.org/10.1038/s41598-021-02866-z

    2021 doi

  11. [17]

    https://arxiv.org/ abs/quant-ph/0407010

    Mottonen, M., Vartiainen, J.J., Bergholm, V., Salomaa, M.M.: Transformation of quantum states using uniformly controlled rotations (2004). https://arxiv.org/ abs/quant-ph/0407010

    2004 arXiv

  12. [18]

    https://arxiv.org/abs/1811.04968

    Bergholm, V., Izaac, J., Schuld, M., Gogolin, C., Ahmed, S., Ajith, V., Alam, M.S., Alonso-Linaje, G., AkashNarayanan, B., Asadi, A., Arrazola, J.M., Azad, U., Banning, S., Blank, C., Bromley, T.R., Cordier, B.A., Ceroni, J., Delgado, A., Matteo, O.D., Dusko, A., Garg, T., Gua...

    2022 arXiv

  13. [19]

    https://arxiv

    Cao, Y., Guerreschi, G.G., Aspuru-Guzik, A.: Quantum Neuron: an elementary building block for machine learning on quantum computers (2017). https://arxiv. org/abs/1711.11240

    2017 arXiv

  14. [21]

    https://qiskit.github.io/qiskit-aer/apidocs/aer noise.html Accessed 2024/12/10

    IBM: Qiskit AER Noise Models. https://qiskit.github.io/qiskit-aer/apidocs/aer noise.html Accessed 2024/12/10

    2024

  15. [22]

    https://doi.org/10.48550/arXiv

    Javadi-Abhari, A., Treinish, M., Krsulich, K., Wood, C.J., Lishman, J., Gacon, J., Martiel, S., Nation, P.D., Bishop, L.S., Cross, A.W., Johnson, B.R., Gambetta, J.M.: Quantum computing with Qiskit (2024). https://doi.org/10.48550/arXiv. 2405.08810

    2024 doi arXiv

  16. [23]

    https://qiskit.github.io/qiskit-aer/ apidocs/aer provider.html Accessed 2024/12/10

    IBM: Qiskit AER Simulator Provider. https://qiskit.github.io/qiskit-aer/ apidocs/aer provider.html Accessed 2024/12/10

    2024

  17. [24]

    https://arxiv.org/abs/1708.05031

    He, X., Liao, L., Zhang, H., Nie, L., Hu, X., Chua, T.-S.: Neural Collaborative Filtering (2017). https://arxiv.org/abs/1708.05031

    2017 arXiv

  18. [25]

    https://arxiv.org/abs/1701.04783

    Zheng, L., Noroozi, V., Yu, P.S.: Joint Deep Modeling of Users and Items Using Reviews for Recommendation (2017). https://arxiv.org/abs/1701.04783

    2017 arXiv

  19. [27]

    Applied Sciences 10(7) (2020) https: //doi.org/10.3390/app10072441

    Bobadilla, J., Alonso, S., Hernando, A.: Deep learning architecture for collab- orative filtering recommender systems. Applied Sciences 10(7) (2020) https: //doi.org/10.3390/app10072441

    2020 doi

  20. [28]

    ACM Trans

    Herlocker, J.L., Konstan, J.A., Terveen, L.G., Riedl, J.T.: Evaluating collabora- tive filtering recommender systems. ACM Trans. Inf. Syst. 22(1), 5–53 (2004) https://doi.org/10.1145/963770.963772

    2004

  21. [29]

    IEEE Internet Computing 7(1), 76–80 (2003) https://doi

    Linden, G., Smith, B., York, J.: Amazon.com recommendations: item-to-item 19 collaborative filtering. IEEE Internet Computing 7(1), 76–80 (2003) https://doi. org/10.1109/MIC.2003.1167344

    2003 arXiv

  22. [30]

    Cambridge University Press, ??? (2012) 20

    Busemeyer, J.R., Bruza, P.D.: Quantum Models of Cognition and Decision. Cambridge University Press, ??? (2012) 20

    2012

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.