Pith. sign in

REVIEW 4 major objections 4 minor 97 references

Quantum-Informed Contrastive Learning with Dynamic Mixup Augmentation for Class-Imbalanced Expert Systems

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read QCL-MixNet claims to set a new benchmark for imbalanced tabular classification by pairing a quantum-entanglement-inspired feature layer with kNN-guided dynamic mixup and a hybrid contrastive loss, reporting superior macro-F1 and recall…

desk verdict The paper's own tables contradict the central claim, so the benchmark results cannot be trusted despite a sensible combination of known components. read the letter →

arxiv 2506.13987 v2 pith:IU42IHAO submitted 2025-06-16 cs.LG

Md Abrar Jahin , Adiba Abid , M. F. Mridha This is my paper · ORCID
classification cs.LG
keywords Quantum-InspiredNeuralNetworksContrastiveLearningClassImbalanceDynamicMixupAugmentationExpertSystemsTabularDataClassificationFocalVarianceLosskNN-Guided
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

QCL-MixNet is a proposed architecture for classifying class-imbalanced tabular data — the setting of expert systems where rare cases (a fraudulent transaction, an early-stage cancer, a turbine-blade crack) are the ones that matter most. The paper tries to establish that a blend of three ideas — a quantum-entanglement-inspired feature layer, a k-nearest-neighbor-guided dynamic mixup, and a four-part hybrid loss — reliably beats twenty established machine-learning, deep-learning, and graph-based baselines on macro-F1 and recall across eighteen real-world datasets. The motivation is that standard remedies (oversampling, cost-sensitive learning, graph neural networks) each carry known weaknesses such as overfitting, label noise, poor generalization in low-density regions, or scalability limits. If the claim holds, QCL-MixNet would become a new benchmark for tabular imbalance handling in expert systems, and the framework's ablations pinpoint the dynamic mixup as the largest single contributor to the reported gains.

What carries the argument

The load-bearing machinery has three parts plus one design choice. (1) The Quantum Entanglement (QE) layer computes $x_{\mathrm{ent}} = \sigma(x^\top\sin\theta)\,(x\odot\cos\theta)$ for a learnable $\theta$: cosine scaling modulates each feature, while the sigmoid of a dot product with $\sin\theta$ gates the whole vector, creating a cheap non-linear coupling between features; two such layers sit at the input and at an intermediate level, followed by a residual self-attention recalibration. (2) kNN-Guided Dynamic Mixup interpolates an anchor with a randomly chosen one of its $k=5$ nearest neighbours in the embedding space, drawing $\lambda\sim\mathrm{Beta}(\alpha,\alpha)$ and forcing $\lambda'\ge 0.5$, then mixes features and one-hot labels; the classification loss, however, uses the original unmixed label, so the model is regularized to keep the anchor's identity under neighbourhood perturbations. (3) The hybrid loss is $\mathcal{L}=\alpha\mathcal{L}_{\mathrm{FVL}}+(1-\alpha)(\mathcal{L}_{\mathrm{SupCon}}+\mathcal{L}_{\mathrm{Triplet}})$, where $\mathcal{L}_{\mathrm{FVL}}$ combines focal loss, an intra-class compactness term $\beta\|\mathbf{e}-\mathbf{c}_y\|^2$ against learnable class centroids, and a log-sigmoid inter-centroid separation penalty. The design choice: a projection head maps representations to an 8-dimensional space used only by the contrastive and variance losses, so the classifier sees a different representation than the metric losses do.

What would settle it

Re-run the 18-dataset comparison with checkpoint selection based on a validation split, touching the test set only once at the end; if QCL-MixNet's macro-F1 lead over the baselines shrinks or reverses — especially on datasets where the paper's own Diff values are negative, such as abalone, HAR, and one-hundred-plants-texture — the central benchmark claim does not survive.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the three components work together rather than additively. The QE layer applies a learnable sinusoidal projection followed by sigmoid gating to create non-linear feature interactions; the kNN-guided mixup replaces random-pair interpolation with interpolation between each anchor and one of its five nearest neighbors in the learned embedding space, biased so the anchor keeps at least half the weight; and the hybrid loss combines focal reweighting, supervised contrastive learning, triplet margin loss (with multi-similarity mining), and centroid-based variance regularization so that embeddings of one class cluster tightly while class centroids repel. Trained for 100 epochs with AdamW on a fixed 80/20 stratified split, the full model reports the best accuracy on 14 of the 18 datasets and the best macro-F1 on 12, and its average rank of 3.55 across the four metrics is the lowest of all 21 models in the Friedman test, with the null hypothesis of equal performance rejected. An ablation attributes the largest average macro-F1 drop, about 24.5%, to removing the dynamic mixup, followed by roughly 23.2% for attention and 11.5% for the QE layers.

Load-bearing premise

The reported rankings assume the retained checkpoint — chosen by macro-F1 computed on the held-out test set after each epoch, per Section 4.2.1 — is an unbiased model-selection rule, so if test information leaks into model selection the reported margins over the baselines are inflated.

Editorial extensions

If this is right

  • If correct, the framework establishes a new benchmark for tabular imbalance handling: on 14 of 18 datasets it reports the best accuracy, on 12 of 18 the best macro-F1, and its global average rank of 3.55 across the four metrics is the lowest of all 21 models compared.
  • The ablation ordering implies the dynamic mixup is the indispensable component: removing it costs on average 24.5% of macro-F1, so future work should treat the augmentation strategy — not the architecture alone — as the primary source of the reported gains.
  • Because the classification loss is computed with the original labels on mixed features, the framework claims to enlarge each class's decision region into the local feature manifold, which is exactly the behaviour needed when minority classes are sparse.
  • The theoretical propositions — universal approximation when QE layers are composed with an MLP, manifold proximity of kNN-guided interpolations, and piecewise smoothness of the hybrid loss — argue that the framework keeps expressiveness and stable optimization while adding the imbalance-specific structure.

Reading between the lines

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

  • A testable extension the paper does not run: replace the QE layers with a parameter-matched plain MLP and re-run the ablations; if the macro-F1 drop is small, the quantum-inspired parameterization is a capacity effect rather than a structural one.
  • The kNN-guided mixup is a generic recipe — interpolate in embedding space toward same-class neighbours under the anchor's own label — that could transfer to long-tailed image or text classification, where the same failure mode of unrealistic random interpolations has been reported.
  • The paper's own tables show negative per-dataset diffs on several datasets (abalone, HAR, one-hundred-plants-texture among them), so the 'consistently outperforms' claim is best read as an average-rank statement rather than a per-dataset guarantee.
  • Algorithm 1's SampleAwareMixup selects the interpolation label from a random permutation rather than from the chosen neighbour's label, contradicting Definition 1; re-running the training with Definition 1 implemented literally would settle which behaviour the reported results actually reflect.
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 QCL-MixNet, a framework for class-imbalanced tabular classification that combines a quantum-entanglement-inspired feature layer, kNN-guided dynamic mixup augmentation, and a hybrid loss (focal, supervised contrastive, triplet, and variance regularization). The authors report experiments on 18 binary and multi-class datasets against 20 ML, DL, and GNN baselines, together with ablations and Friedman tests, and claim that QCL-MixNet consistently and substantially outperforms all baselines in macro-F1 and recall, establishing a new benchmark for tabular imbalance handling.

Significance. If the results were reliable, QCL-MixNet would be a practically relevant method for imbalanced tabular classification, and the breadth of the evaluation (18 datasets, 20 baselines, ablations, and a statistical test) is a strength. The paper also deserves credit for attempting a systematic ablation of its components. However, the central claim rests entirely on the benchmark tables, which are internally inconsistent and were produced under a test-set-based checkpoint-selection protocol. As presented, the empirical evidence does not support the claimed consistent superiority, so the significance of the contribution cannot be assessed from this manuscript.

major comments (4)
  1. [Tables 3 and 5 (Section 5.2)] QCL-MixNet (FULL) on MINERALS is reported as maP 0.84, maR 0.87, maF1 0.85 in Table 3 but as maP 0.47, maR 0.47, maF1 0.43 in Table 5; SATIMAGE has maF1 0.89 in Table 3 and 0.91 in Table 5. These rows describe the same full model on the same datasets, so at least one table cannot have been produced by the described protocol. Since the Diff rows, the claimed benchmark superiority, and the Friedman test in Table 6 are all computed from these tables, the central empirical claim is not supported by the paper's own evidence. In addition, the Diff rows themselves contradict 'consistently outperforms' on several datasets (e.g., ECOLI maF1 Diff -0.02, ABALONE maF1 Diff -0.15, HAR maF1 Diff -0.10, ONE-HUNDRED-PLANTS maF1 Diff -0.02).
  2. [Section 4.2.1 and Algorithm 1] The text states that the macro-F1 was computed after each epoch on the held-out test set and the best checkpoint was retained for final evaluation, while Algorithm 1 refers to validation data and no validation split is defined in Section 4.1. Selecting the checkpoint on the test set leaks test labels into model selection and makes the reported test metrics optimistically biased; this invalidates the comparison with baselines and the claimed rankings.
  3. [Section 3.2.2 vs Algorithm 1 (lines 22-28 and 50)] Definition 1 states that mixup interpolates an anchor with a kNN and its label, but the pseudocode selects the label from a random permutation that is explicitly not the neighbor's label, and the hybrid loss is computed with the original labels y rather than the mixed labels. The core augmentation mechanism is therefore described inconsistently, and it is unclear which procedure was actually implemented.
  4. [Sections 4.2.1, 5.3 and Table 6] The evaluation uses a single fixed split and seed with no error bars or repeated runs, and the Friedman test is used only to reject a global null hypothesis. It does not show that QCL-MixNet significantly outperforms any specific baseline; in fact, SVM (SMOTE) has a better average rank on both maR and maF1 (2.8 vs 3.2-3.6). The Diff rows are computed against 'either the best-performing or second-best-performing baseline ... depending on which is more relevant in each scenario,' but the selection rule is never specified, so the margins cannot be verified. Finally, many DL/GNN baselines report values that match a degenerate majority-class predictor (e.g., 0.90/0.45/0.50/0.47 on ECOLI), which raises the concern that the baselines were not trained to a competitive standard.
minor comments (4)
  1. [Throughout] Several figure references appear as 'Figure??' in Sections 3.2, 5.1, and 5.2, but no figures are included in the manuscript; this must be fixed before any resubmission.
  2. [Section 1] The sentence 'These issues suggest to suggest to handle the class imbalance issue carefully' contains a duplicated phrase and should be rewritten.
  3. [Section 4.3.1] The evaluation section does not state how undefined per-class precision or recall were handled (e.g., zero_division setting), which is important because many reported baseline rows have identical degenerate values.
  4. [Section 3.2.1] The name 'Quantum Entanglement-inspired layer' overclaims the mechanism; the layer is a sinusoidal projection followed by a scalar sigmoid gate, and the text itself notes it is not a quantum simulation. A more neutral name would be more appropriate.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported test metrics are selected on the held-out test set, so the benchmark 'predictions' are maxima over epochs rather than independent estimates; the central empirical claim is partly fitted to its evaluation target.

  1. fitted input called prediction [Section 4.2.1 (Training details and reproducibility measures); also Algorithm 1 lines 53-56]
    "Macro F1 score was computed after each epoch on the held-out test set, and the best-performing model checkpoint was retained for final evaluation."

    The reported final macro-F1 for the deep/QI models is not an independent estimate of generalization: the checkpoint is selected by maximizing macro-F1 on the same held-out test set on which final performance is then reported. By construction, each reported number is the maximum of up to 100 test-set evaluations, so the abstract's claim that QCL-MixNet 'consistently outperforms 20 state-of-the-art ... baselines in macro-F1 and recall' is a comparison of test-selected maxima, not of predicted generalization. Since Section 4.2.1 presents this as the shared protocol for all deep and QI models, the 'Diff' rows and Friedman rankings are built from the same selected-maximum scores. This is a statistically forced evaluation artifact rather than a derivation from the method's components.

full rationale

No load-bearing self-citation, imported uniqueness theorem, or ansatz-via-citation circularity was found: the QE layer, kNN mixup, and hybrid loss are all defined in the paper and evaluated on external public datasets, and the only self-citation (Jahin et al., 2023) is used for background and future direction, not to justify the central claim. The theoretical propositions are sketches but they do not reduce a prediction to an input; Proposition 1 merely invokes the known universal-approximation closure of MLPs, and Proposition 4 cites the external SimCLR result. The one genuine circularity-adjacent step is the test-set checkpoint selection in Section 4.2.1, which makes every reported deep-model score a test-selected maximum and thereby partially constructs the claimed benchmark superiority. This is scored as 6 because one class of reported predictions (final test metrics) reduces by construction to the selection criterion. Separately, the paper contains serious internal consistency problems (e.g., QCL-MixNet FULL on MINERALS is maP 0.84 in Table 3 but 0.47 in Table 5, and SATIMAGE maF1 differs between Table 3 and Table 5); these are correctness and falsification concerns that also undermine the headline claim, but they are not themselves a circular-derivation pattern and are not counted in the circularity score beyond reinforcing caution.

Assumptions & free parameters 7 free parameters · 4 assumptions · 1 invented entities

The central empirical claim rests on many hand-selected hyperparameters and a test-set-based checkpoint selection rule. The only invented conceptual entity, the QE layer, is a simple classical feature modulation with no quantum content and no independent evidence. The theoretical sections rely on standard universal approximation facts and unproven manifold assumptions.

free parameters (7)
  • mixup Beta alpha = not reported
    Controls the Beta distribution for interpolation; the value is not stated in the paper.
  • k (number of neighbors) = 5
    Neighborhood size for kNN-guided mixup (Section 3.2.2).
  • beta1 (intra-class compactness weight) = 0.8
    Weight of ||e - c_y||^2 in the focal variance loss (Section 3.2.3).
  • gamma (focal loss focusing parameter) = 3.0
    Focal loss exponent (Section 3.2.3).
  • tau (SupCon temperature) = 0.2
    Temperature in supervised contrastive loss (Section 3.2.3).
  • triplet margin m = 0.5
    Margin in triplet loss (Section 3.2.3).
  • alpha_loss (hybrid weighting) = 0.5
    Weight between focal variance loss and contrastive/triplet losses (Section 3.2.3, Eq. 14).
assumptions (4)
  • standard math Feed-forward networks with non-linear activations are universal approximators, and composing them with a continuous feature transformation preserves universality.
    Used in Proposition 1 (Section 3.3.1), but the QE layer is a fixed continuous map, so the MLP alone already provides universality.
  • domain assumption Interpolating an anchor with one of its k-nearest neighbors in the current learned feature space keeps synthetic samples close to the data manifold.
    Section 3.2.2 and Proposition 2; this is an unproven assumption that depends on the learned representation.
  • domain assumption The Friedman test with N = 4 metrics as repeated measures is a valid test of model differences.
    Section 5.3.2 uses the four evaluation metrics as replications, which is statistically questionable; the test is designed for repeated runs or multiple datasets.
  • ad hoc to paper Selecting the checkpoint with the highest macro-F1 on the held-out test set gives unbiased performance estimates.
    Section 4.2.1; this leaks test information into model selection and inflates the reported scores.
invented entities (1)
  • Quantum Entanglement-inspired layer (QE layer)
    purpose: Provides feature modulation via cosine scaling and a scalar sigmoid gate; claimed to model complex feature interactions and disentangle representations.
    Despite the name, it is a deterministic classical transformation with no quantum mechanical content; no falsifiable prediction outside this paper is attached to it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum-Informed Contrastive Learning with Dynamic Mixup Augmentation for Class-Imbalanced Expert Systems." pith.science (2026). https://pith.science/paper/IU42IHAO

@misc{pith2026250613987,
  author       = {Pith},
  title        = {Pith review of: Quantum-Informed Contrastive Learning with Dynamic Mixup Augmentation for Class-Imbalanced Expert Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IU42IHAO}},
  note         = {Machine review of arXiv:2506.13987}
}
read the original abstract

Expert systems often operate in domains characterized by class-imbalanced tabular data, where detecting rare but critical instances is essential for safety and reliability. While conventional approaches, such as cost-sensitive learning, oversampling, and graph neural networks, provide partial solutions, they suffer from drawbacks like overfitting, label noise, and poor generalization in low-density regions. To address these challenges, we propose QCL-MixNet, a novel Quantum-Informed Contrastive Learning framework augmented with k-nearest neighbor (kNN) guided dynamic mixup for robust classification under imbalance. QCL-MixNet integrates three core innovations: (i) a Quantum Entanglement-inspired layer that models complex feature interactions through sinusoidal transformations and gated attention, (ii) a sample-aware mixup strategy that adaptively interpolates feature representations of semantically similar instances to enhance minority class representation, and (iii) a hybrid loss function that unifies focal reweighting, supervised contrastive learning, triplet margin loss, and variance regularization to improve both intra-class compactness and inter-class separability. Extensive experiments on 18 real-world imbalanced datasets (binary and multi-class) demonstrate that QCL-MixNet consistently outperforms 20 state-of-the-art machine learning, deep learning, and GNN-based baselines in macro-F1 and recall, often by substantial margins. Ablation studies further validate the critical role of each architectural component. Our results establish QCL-MixNet as a new benchmark for tabular imbalance handling in expert systems. Theoretical analyses reinforce its expressiveness, generalization, and optimization robustness.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

97 extracted references · 72 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    and Huang, C

    Ando, S. and Huang, C. Y. (2017). Deep Over -sampling Framework for Classifying Imbalanced Data . In Ceci, M., Hollmén, J., Todorovski, L., Vens, C., and Džeroski, S., editors, Machine Learning and Knowledge Discovery in Databases , pages 770--785, Cham. Springer International Publishing

    2017

  3. [3]

    Araf, I., Idri, A., and Chairi, I. (2024). Cost-sensitive learning for imbalanced medical data: a review. Artificial Intelligence Review , 57(4):80

    2024

  4. [4]

    Asuncion, A., Newman, D., et al. (2007). Uci machine learning repository

    2007

  5. [5]

    S., Garcı́a, V., and Rangel, E

    Barandela, R., Sánchez, J. S., Garcı́a, V., and Rangel, E. (2003). Strategies for learning in class imbalance problems. Pattern Recognition , 36(3):849--851

    2003

  6. [6]

    Batista, G. E. A. P. A., Prati, R. C., and Monard, M. C. (2004). A study of the behavior of several methods for balancing machine learning training data. ACM SIGKDD Explorations Newsletter , 6(1):20--29

    2004

  7. [7]

    Bengio, Y., Courville, A., and Vincent, P. (2013). Representation Learning : A Review and New Perspectives . IEEE Transactions on Pattern Analysis and Machine Intelligence , 35(8):1798--1828

    2013

  8. [8]

    Branco, P., Torgo, L., and Ribeiro, R. P. (2016). A Survey of Predictive Modeling on Imbalanced Domains . ACM Comput. Surv. , 49(2):31:1--31:50

    2016

Show all 97 references
  1. [9]

    Buda, M., Maki, A., and Mazurowski, M. A. (2018). A systematic study of the class imbalance problem in convolutional neural networks. Neural Networks , 106:249--259

    2018

  2. [10]

    A., Yones, C., Milone, D

    Bugnon, L. A., Yones, C., Milone, D. H., and Stegmayer, G. (2020). Deep Neural Architectures for Highly Imbalanced Data in Bioinformatics . IEEE Transactions on Neural Networks and Learning Systems , 31(8):2857--2867

    2020

  3. [11]

    Cao, Y., Jia, M., Zhao, X., Yan, X., and Feng, K. (2024). Cost-sensitive learning considering label and feature distribution consistency: A novel perspective for health prognosis of rotating machinery with imbalanced data. Expert Systems with Applications , 250:123930

    2024

  4. [12]

    C., Huang, H.-Y., Cerezo, M., Sharma, K., Sornborger, A., Cincio, L., and Coles, P

    Caro, M. C., Huang, H.-Y., Cerezo, M., Sharma, K., Sornborger, A., Cincio, L., and Coles, P. J. (2022). Generalization in quantum machine learning from few training data. Nature Communications , 13(1):4919. Publisher: Nature Publishing Group

    2022

  5. [13]

    J., and Brás, S

    Carvalho, M., Pinho, A. J., and Brás, S. (2025). Resampling approaches to handle class imbalance: a review from a data perspective. Journal of Big Data , 12(1):71

    2025

  6. [14]

    Y.-C., Wei, T.-C., Zhang, C., Yu, H., and Yoo, S

    Chen, S. Y.-C., Wei, T.-C., Zhang, C., Yu, H., and Yoo, S. (2022). Quantum convolutional neural networks for high energy physics data analysis. Physical Review Research , 4(1):013231. Publisher: American Physical Society

    2022

  7. [15]

    Chen, T., Kornblith, S., Norouzi, M., and Hinton, G. (2020). A Simple Framework for Contrastive Learning of Visual Representations . In Proceedings of the 37th International Conference on Machine Learning , pages 1597--1607. PMLR. ISSN: 2640-3498

    2020

  8. [16]

    D., Pontil, M., Rocchetto, A., Severini, S., and Wossnig, L

    Ciliberto, C., Herbster, M., Ialongo, A. D., Pontil, M., Rocchetto, A., Severini, S., and Wossnig, L. (2018). Quantum machine learning: a classical perspective. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , 474(2209):20170551. Publisher:...

    2018

  9. [17]

    and Weston, J

    Collobert, R. and Weston, J. (2008). A unified architecture for natural language processing: deep neural networks with multitask learning. In Proceedings of the 25th International Conference on Machine Learning , ICML '08, pages 160--167, New York, NY, USA. Association for Com...

    2008

  10. [18]

    Dai, Q., Liu, J.-w., and Shi, Y.-h. (2023). Class-overlap undersampling based on Schur decomposition for Class -imbalance problems. Expert Systems with Applications , 221:119735

    2023

  11. [19]

    Dai, Q., Zhou, X., Yang, J.-p., Du, T., and Chen, L.-f. (2025). A mutually supervised heterogeneous selective ensemble learning framework based on matrix decomposition for class imbalance problem. Expert Systems with Applications , 271:126728

    2025

  12. [20]

    K., and Purkayastha, B

    Devi, D., Biswas, S. K., and Purkayastha, B. (2020). A Review on Solution to Class Imbalance Problem : Undersampling Approaches . In 2020 International Conference on Computational Performance Evaluation ( ComPE ) , pages 626--631

    2020

  13. [21]

    and Gandotra, E

    Dhalaria, M. and Gandotra, E. (2024). MalDetect : A classifier fusion approach for detection of android malware. Expert Systems with Applications , 235:121155

    2024

  14. [22]

    and Lenssen, J

    Fey, M. and Lenssen, J. E. (2019). Fast Graph Representation Learning with PyTorch Geometric . arXiv:1903.02428 [cs]

    2019 arXiv

  15. [23]

    Gan, D., Shen, J., An, B., Xu, M., and Liu, N. (2020). Integrating TANBN with cost sensitive classification algorithm for imbalanced data in medical diagnosis. Computers & Industrial Engineering , 140:106266

    2020

  16. [24]

    Ghosh, K., Bellinger, C., Corizzo, R., Branco, P., Krawczyk, B., and Japkowicz, N. (2024). The class imbalance problem in deep learning. Machine Learning , 113(7):4845--4901

    2024

  17. [25]

    E., Köksal, O., Babur, O., and Tekinerdogan, B

    Giray, G., Bennin, K. E., Köksal, O., Babur, O., and Tekinerdogan, B. (2023). On the use of deep learning in software defect prediction. Journal of Systems and Software , 195:111537

    2023

  18. [26]

    Gori, M., Monfardini, G., and Scarselli, F. (2005). A new model for learning in graph domains. In Proceedings. 2005 IEEE international joint conference on neural networks, 2005. , volume 2, pages 729--734. IEEE

    2005

  19. [27]

    H., and Nakajima, K

    Goto, T., Tran, Q. H., and Nakajima, K. (2021). Universal Approximation Property of Quantum Machine Learning Models in Quantum - Enhanced Feature Spaces . Physical Review Letters , 127(9):090506. Publisher: American Physical Society

    2021

  20. [28]

    and Huang, H

    Guo, C. and Huang, H. (2025). Enhancing camouflaged object detection through contrastive learning and data augmentation techniques. Engineering Applications of Artificial Intelligence , 141:109703

    2025

  21. [29]

    F., Abreu, P

    Guo, J., Chong, C. F., Abreu, P. H., Mao, C., Li, J., Lam, C.-T., and Ng, B. K. (2025). Reparameterization convolutional neural networks for handling imbalanced datasets in solar panel fault classification. Engineering Applications of Artificial Intelligence , 150:110541

    2025

  22. [30]

    Hamilton, W., Ying, Z., and Leskovec, J. (2017). Inductive representation learning on large graphs. In Guyon, I., Luxburg, U. V., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., and Garnett, R., editors, Advances in Neural Information Processing Systems , volume 30. Cur...

    2017

  23. [31]

    Harrow, A. W. and Montanaro, A. (2017). Quantum computational supremacy. Nature , 549(7671):203--209. Publisher: Nature Publishing Group

    2017

  24. [32]

    D., Temme, K., Harrow, A

    Havlíček, V., Córcoles, A. D., Temme, K., Harrow, A. W., Kandala, A., Chow, J. M., and Gambetta, J. M. (2019). Supervised learning with quantum-enhanced feature spaces. Nature , 567(7747):209--212. Publisher: Nature Publishing Group

    2019

  25. [33]

    He, K., Zhang, X., Ren, S., and Sun, J. (2016). Deep Residual Learning for Image Recognition . In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 770--778

    2016

  26. [34]

    Hong, Y.-Y., Rioflorido, C. L. P. P., and Zhang, W. (2024). Hybrid deep learning and quantum-inspired neural network for day-ahead spatiotemporal wind speed forecasting. Expert Systems with Applications , 241:122645

    2024

  27. [35]

    Hu, H., Wang, X., Zhang, Y., Chen, Q., and Guan, Q. (2024). A comprehensive survey on contrastive learning. Neurocomputing , 610:128645

    2024

  28. [36]

    C., and Tang, X

    Huang, C., Li, Y., Loy, C. C., and Tang, X. (2020). Deep Imbalanced Learning for Face Recognition and Attribute Prediction . IEEE Transactions on Pattern Analysis and Machine Intelligence , 42(11):2781--2794

    2020

  29. [37]

    Hunter, J. D. (2007). Matplotlib: A 2D Graphics Environment . Computing in Science & Engineering , 9(3):90--95

    2007

  30. [38]

    Isomura, T., Shimizu, R., and Goto, M. (2025). LLMOverTab : Tabular data augmentation with language model-driven oversampling. Expert Systems with Applications , 264:125852

    2025

  31. [39]

    A., Shovon, M

    Jahin, M. A., Shovon, M. S. H., Islam, M. S., Shin, J., Mridha, M. F., and Okuyama, Y. (2023). QAmplifyNet : pushing the boundaries of supply chain backorder prediction using interpretable hybrid quantum-classical neural network. Scientific Reports , 13(1):18246. Publisher: Na...

    2023

  32. [40]

    V., and Gusev, G

    Kachan, O., Savchenko, A. V., and Gusev, G. (2025). Simplicial SMOTE: Oversampling Solution to the Imbalanced Learning Problem . In Sun, Y., Chierichetti, F., Lauw, H. W., Perlich, C., Tok, W. H., and Tomkins, A., editors, Proceedings of the 31st ACM SIGKDD Conference on Knowl...

    2025

  33. [41]

    Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y., Isola, P., Maschinot, A., Liu, C., and Krishnan, D. (2020). Supervised contrastive learning. In Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., and Lin, H., editors, Advances in Neural Information Processing System...

    2020

  34. [42]

    Kipf, T. N. and Welling, M. (2017). Semi-supervised classification with graph convolutional networks. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings . OpenReview.net

    2017

  35. [43]

    K., and Panigrahi, B

    Konar, D., Bhattacharyya, S., Gandhi, T. K., and Panigrahi, B. K. (2020). A Quantum - Inspired Self - Supervised Network model for automatic segmentation of brain MR images. Applied Soft Computing , 93:106348

    2020

  36. [44]

    and Khalil, I

    Kottahachchi Kankanamge Don, A. and Khalil, I. (2025). Q- SupCon : Quantum - Enhanced Supervised Contrastive Learning Architecture within the Representation Learning Framework . ACM Transactions on Quantum Computing , 6(1):6:1--6:24

    2025

  37. [45]

    Krawczyk, B. (2016). Learning from imbalanced data: open challenges and future directions. Progress in Artificial Intelligence , 5(4):221--232

    2016

  38. [46]

    Krawczyk, B., Woźniak, M., and Schaefer, G. (2014). Cost-sensitive decision tree ensembles for effective imbalanced classification. Applied Soft Computing , 14:554--562

    2014

  39. [47]

    B., and Kim, D

    Lee, S., Park, M., Ahn, Y., Jung, G. B., and Kim, D. (2024). Analysis of tabular data based on graph neural network using supervised contrastive loss. Neurocomputing , 570:127137

    2024

  40. [48]

    Li, Q., Gao, X., Lu, H., Li, B., Zhai, F., Wang, T., Meng, Z., and Hao, Y. (2025a). A meta-learning imbalanced classification framework via boundary enhancement strategy with Bayes imbalance impact index. Neural Networks , 185:107126

    2025

  41. [49]

    Li, S., Hua, H., and Chen, S. (2025b). Graph neural networks for single-cell omics data: a review of approaches and applications. Briefings in Bioinformatics , 26(2):bbaf109

    2025

  42. [50]

    Liu, J., Liu, M., Liu, J.-P., Ye, Z., Wang, Y., Alexeev, Y., Eisert, J., and Jiang, L. (2024). Towards provably efficient quantum algorithms for large-scale machine-learning models. Nature Communications , 15(1):434. Publisher: Nature Publishing Group

    2024

  43. [51]

    Liu, Z., Wu, F., Wang, Y., Yang, M., and Pan, X. (2023). FedCL : Federated contrastive learning for multi-center medical image classification. Pattern Recognition , 143:109739

    2023

  44. [52]

    López, V., Fernández, A., García, S., Palade, V., and Herrera, F. (2013). An insight into classification with imbalanced data: Empirical results and current trends on using data intrinsic characteristics. Information Sciences , 250:113--141

    2013

  45. [53]

    K., Nadim, M., Roy, C

    Mandal, A. K., Nadim, M., Roy, C. K., Roy, B., and Schneider, K. A. (2025). Quantum Software Engineering and Potential of Quantum Computing in Software Engineering Research : A Review . arXiv:2502.08925 [cs]

    2025 arXiv

  46. [54]

    A., Awal, M

    Orka, N. A., Awal, M. A., Liò, P., Pogrebna, G., Ross, A. G., and Moni, M. A. (2025). Quantum deep learning in neuroinformatics: a systematic review. Artificial Intelligence Review , 58(5):134

    2025

  47. [55]

    Paszke, A. (2019). Pytorch: An imperative style, high-performance deep learning library. arXiv preprint arXiv:1912.01703

    2019 arXiv

  48. [56]

    K., Ho, M

    Patnaik, R. K., Ho, M. C., and Yeh, J. A. (2023). Weighted SMOTE Algorithm : A Tool To Improve Disease Prediction With Imbalanced Data . In 2023 International Conference on Consumer Electronics - Taiwan ( ICCE - Taiwan ) , pages 259--260. ISSN: 2575-8284

    2023

  49. [57]

    Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., et al. (2011). Scikit-learn: Machine learning in Python . the Journal of machine Learning research , 12:2825--2830

    2011

  50. [58]

    J., Rivera, A

    Pulgar, F. J., Rivera, A. J., Charte, F., and del Jesus, M. J. (2017). On the Impact of Imbalanced Data in Convolutional Neural Networks Performance . In Hybrid Artificial Intelligent Systems , pages 220--232. Springer, Cham. ISSN: 1611-3349

    2017

  51. [59]

    B., Krishnan, S., and Niculescu, R

    Rao, R. B., Krishnan, S., and Niculescu, R. S. (2006). Data mining for improved cardiac care. SIGKDD Explor. Newsl. , 8(1):3--10

    2006

  52. [60]

    and Polak, W

    Rieffel, E. and Polak, W. (2000). An introduction to quantum computing for non-physicists. ACM Computing Surveys , 32(3):300--335

    2000

  53. [61]

    E., Hamann, A., Strömberg, T., Schiansky, P., Dunjko, V., Friis, N., Harris, N

    Saggio, V., Asenbeck, B. E., Hamann, A., Strömberg, T., Schiansky, P., Dunjko, V., Friis, N., Harris, N. C., Hochberg, M., Englund, D., Wölk, S., Briegel, H. J., and Walther, P. (2021). Experimental quantum speed-up in reinforcement learning agents. Nature , 591(7849):229--233...

    2021

  54. [62]

    C., Hagenbuchner, M., and Monfardini, G

    Scarselli, F., Gori, M., Tsoi, A. C., Hagenbuchner, M., and Monfardini, G. (2009). The graph neural network model. IEEE Transactions on Neural Networks , 20(1):61--80

    2009

  55. [63]

    Sharief, F., Ijaz, H., Shojafar, M., and Naeem, M. A. (2025). Multi-Class Imbalanced Data Handling with Concept Drift in Fog Computing: A Taxonomy, Review, and Future Directions . ACM Comput. Surv. , 57(1):16:1--16:48

    2025

  56. [64]

    Sharma, S., Gosain, A., and Jain, S. (2022). A Review of the Oversampling Techniques in Class Imbalance Problem . In Khanna, A., Gupta, D., Bhattacharyya, S., Hassanien, A. E., Anand, S., and Jaiswal, A., editors, International Conference on Innovative Computing and Communicat...

    2022

  57. [65]

    Shi, J., Li, Z., Lai, W., Li, F., Shi, R., Feng, Y., and Zhang, S. (2023). Two End -to- End Quantum - Inspired Deep Neural Networks for Text Classification . IEEE Transactions on Knowledge and Data Engineering , 35(4):4335--4345

    2023

  58. [66]

    Expert system methodologies and applications—a decade review from 1995 to 2004

    Shu-Hsien Liao (2005). Expert system methodologies and applications—a decade review from 1995 to 2004. Expert Systems with Applications , 28(1):93--103

    2005

  59. [67]

    Sun, C., Shrivastava, A., Singh, S., and Gupta, A. (2017). Revisiting Unreasonable Effectiveness of Data in Deep Learning Era . In 2017 IEEE International Conference on Computer Vision ( ICCV ) , pages 843--852, Venice. IEEE

    2017

  60. [68]

    Y., Ong, H

    Tan, C. Y., Ong, H. F., Lim, C. H., Tan, M. S., Ooi, E. H., and Wong, K. (2025). Amogel: a multi-omics classification framework using associative graph neural networks with prior knowledge for biomarker identification. BMC Bioinformatics , 26(1):94

    2025

  61. [69]

    Tang, J., Li, Y., Hou, Z., Fu, S., and Tian, Y. (2024). Robust two-stage instance-level cost-sensitive learning method for class imbalance problem. Knowledge-Based Systems , 300:112143

    2024

  62. [70]

    Tao, S., Peng, P., Li, Y., Sun, H., Li, Q., and Wang, H. (2024). Supervised contrastive representation learning with tree-structured parzen estimator Bayesian optimization for imbalanced tabular data. Expert Systems with Applications , 237:121294

    2024

  63. [71]

    Thabtah, F., Hammoud, S., Kamalov, F., and Gonsalves, A. (2020). Data imbalance in classification: Experimental evaluation. Information Sciences , 513:429--441

    2020

  64. [72]

    N., Bischl, B., Torgo, L., Gao, B., Umaashankar, V., Fischer, S., Winter, P., Wiswedel, B., Berthold, M

    van Rijn, J. N., Bischl, B., Torgo, L., Gao, B., Umaashankar, V., Fischer, S., Winter, P., Wiswedel, B., Berthold, M. R., and Vanschoren, J. (2013). OpenML: A Collaborative Science Platform . In Machine Learning and Knowledge Discovery in Databases , pages 645--649, Berlin, He...

    2013

  65. [73]

    Velickovic, P., Cucurull, G., Casanova, A., Romero, A., Lio, P., Bengio, Y., et al. (2017). Graph attention networks. stat , 1050(20):10--48550

    2017

  66. [74]

    Villaizán-Vallelado, M., Salvatori, M., Carro, B., and Sanchez-Esguevillas, A. J. (2024). Graph Neural Network contextual embedding for Deep Learning on tabular data. Neural Networks , 173:106180

    2024

  67. [75]

    Wang, L.-H., Dai, Q., Du, T., and Chen, L.-f. (2024). Lightweight intrusion detection model based on CNN and knowledge distillation. Applied Soft Computing , 165:112118

    2024

  68. [76]

    Wang, L.-h., Dai, Q., Du, T., and Chen, L.-f. (2025). Novel resampling algorithms with maximal cliques for class-imbalance problems. Computers & Industrial Engineering , 199:110754

    2025

  69. [77]

    Wang, S., Leandro L., M., Nitesh, C., and Yao, X. (2019a). Learning from data streams and class imbalance. Connection Science , 31(2):103--104. Publisher: Taylor & Francis \_eprint: https://doi.org/10.1080/09540091.2019.1572975

    2019

  70. [78]

    Wang, X., Han, X., Huang, W., Dong, D., and Scott, M. R. (2019b). Multi-Similarity Loss With General Pair Weighting for Deep Metric Learning . In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 5017--5025, Los Alamitos, CA, USA. IEEE Computer Society

    2019

  71. [79]

    Wang, X., Xiang, J., Zhang, J., Yang, S., Yang, Z., Wang, M.-H., Zhang, J., Yang, W., Huang, J., and Han, X. (2022). SCL - WC : Cross - Slide Contrastive Learning for Weakly - Supervised Whole - Slide Image Classification . In Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D....

    2022

  72. [80]

    Waskom, M. L. (2021). Seaborn: statistical data visualization. Journal of Open Source Software , 6(60):3021

    2021

  73. [81]

    Wei, W., Li, J., Cao, L., Ou, Y., and Chen, J. (2013). Effective detection of sophisticated online banking fraud on extremely imbalanced data. World Wide Web , 16(4):449--475

    2013

  74. [82]

    Wu, Y., Zhu, D., and Wang, X. (2023). Contrastive learning enhanced deep neural network with serial regularization for high-dimensional tabular data. Expert Systems with Applications , 228:120243

    2023

  75. [83]

    Xu, K., Hu, W., Leskovec, J., and Jegelka, S. (2019). How powerful are graph neural networks? In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019 . OpenReview.net

    2019

  76. [84]

    Xu, K., Li, C., Tian, Y., Sonobe, T., Kawarabayashi, K.-i., and Jegelka, S. (2018). Representation learning on graphs with jumping knowledge networks. In International conference on machine learning , pages 5453--5462. PMLR

    2018

  77. [85]

    and Wong, S

    Xu, X. and Wong, S. T. C. (2025). Contrastive learning in brain imaging. Computerized Medical Imaging and Graphics , 121:102500

    2025

  78. [86]

    and Xu, Z

    Yang, Y. and Xu, Z. (2020). Rethinking the Value of Labels for Improving Class - Imbalanced Learning . In Advances in Neural Information Processing Systems , volume 33, pages 19290--19301. Curran Associates, Inc

    2020

  79. [87]

    Yang, Y., Zhang, G., Xu, Z., and Katabi, D. (2019). ME-Net: Towards Effective Adversarial Robustness with Matrix Estimation . In Chaudhuri, K. and Salakhutdinov, R., editors, Proceedings of the 36th International Conference on Machine Learning, ICML 2019, 9-15 June 2019, Long ...

    2019

  80. [88]

    Ye, M., Yu, Y., Shen, Z., Yu, W., and Zeng, Q. (2024). Cross- Feature Interactive Tabular Data Modeling With Multiplex Graph Neural Networks . IEEE Transactions on Knowledge and Data Engineering , 36(12):7851--7864

    2024

  81. [89]

    Yu, H., Du, Y., and Wu, J. (2025). Reviving undersampling for long-tailed learning. Pattern Recognition , 161:111200

    2025

  82. [90]

    Yuan, Y., Wei, J., Huang, H., Jiao, W., Wang, J., and Chen, H. (2023). Review of resampling techniques for the treatment of imbalanced industrial data classification in equipment condition monitoring. Engineering Applications of Artificial Intelligence , 126:106911

    2023

  83. [91]

    N., and Lopez - Paz, D

    Zhang, H., Ciss \' e , M., Dauphin, Y. N., and Lopez - Paz, D. (2018). mixup: Beyond Empirical Risk Minimization . In 6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings

    2018

  84. [92]

    Zhang, R., Zhang, Z., and Wang, D. (2021). RFCL : A new under-sampling method of reducing the degree of imbalance and overlap. Pattern Analysis and Applications , 24(2):641--654

    2021

  85. [93]

    Zhang, Y., Han, Q., Wang, L., Cheng, K., Wang, B., and Zhan, K. (2025). Graph-weighted contrastive learning for semi-supervised hyperspectral image classification. Journal of Electronic Imaging , 34(2):023044. Publisher: SPIE

    2025

  86. [94]

    Zhao, L., Zhang, Y., Duan, J., and Yu, J. (2025). Cross-supervised contrastive learning domain adaptation network for steel defect segmentation. Advanced Engineering Informatics , 64:102964

    2025

  87. [95]

    Zheng, M., Hu, X., Hu, Y., Zheng, X., and Luo, Y. (2025). Fed-ugi: Federated undersampling learning framework with gini impurity for imbalanced network intrusion detection. IEEE Trans. Inf. Forensics Secur. , 20:1262--1277

    2025

  88. [96]

    Zhou, W., Lin, K., Zheng, Z., Chen, D., Su, T., and Hu, H. (2025). DRTN : Dual Relation Transformer Network with feature erasure and contrastive learning for multi-label image classification. Neural Networks , 187:107309

    2025

  89. [97]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.