Pith. sign in

REVIEW 4 major objections 6 minor 39 references

Quantum Ensembling Methods for Healthcare and Life Science

T0 review · 4 major / 6 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 Quantum ensembles of cosine classifiers can match or slightly improve on random forests for predicting immunotherapy response in renal-cell carcinoma, using as few as two to four training samples.

desk verdict An honest exploratory QML study whose only significant quantum-over-classical result evaporates once you notice the random forest baseline is a constant predictor in most splits. read the letter →

arxiv 2506.02213 v1 pith:6FWGZGVT submitted 2025-06-02 cs.LG q-bio.GNq-bio.QM

Kahn Rhrissorrakrai , Kathleen E. Hamilton , Prerana Bangalore Parthsarathy , Aldo Guzman-Saenz , Tyler Alban , Filippo Utro , Laxmi Parida This is my paper
classification cs.LGq-bio.GNq-bio.QM
keywords quantummachinelearningensemblesboostingsmalldatahealthcarelifesciencerenalcellcarcinomaimmunotherapyresponseprediction
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 asks whether quantum ensemble classifiers can learn reliably from very small healthcare datasets, where biological samples are scarce and classical models often overfit. The authors construct several quantum ensembles—cosine-similarity weak learners, ensembles with random unitaries, and variational circuits—and test them on synthetic Gaussian blobs and on gene-expression data from 150 renal-cell carcinoma patients to predict immunotherapy response. Their central empirical claim is that the quantum classifiers perform comparably to, and in one configuration (a bagged variational ensemble) significantly better than, a tuned random forest on the full feature space, while the non-variational quantum ensemble achieves this using only two to four training samples. They also report that a 56-qubit hardware implementation with error mitigation reaches a similar F1 score to simulation. If correct, this suggests quantum ensembles could be a practical tool for data-constrained problems in healthcare and life science.

What carries the argument

The load-bearing mechanism is the quantum ensemble cosine classifier (QEC): a circuit that uses a $d$-qubit control register to place $2^d$ transformations of the training set into superposition, then performs a swap-test interference measurement that evaluates the cosine distance between test and training samples. Averaging over these trajectories in a single measurement is what lets the QEC behave as an ensemble while drawing on only a handful of training samples. The paper also contrasts this with variational ensembles, whose weak learners are shallow parameterized circuits with amplitude embedding and single-qubit rotations, aggregated classically by soft voting, bagging, or boosting.

What would settle it

A concrete check: run the same renal-cell carcinoma comparison under a single fixed 80/20 split with all hyperparameters chosen on the training fold before seeing the test fold, and apply a multiple-testing correction across the seven classifier types; if the bagged ensemble's F1 advantage over the random forest (p = 0.018) does not survive this protocol, the paper's headline comparison loses its support.

Watch

Extended reading notes

Core claim

The paper's central discovery is that an ensemble of quantum cosine classifiers, each a swap-test weak learner that compares a test sample to a random training sample, can match or slightly improve on a classical random forest for predicting immunotherapy response in renal-cell carcinoma from gene-expression data. In simulation over ten 80/20 splits of 150 patient samples, the quantum ensemble cosine classifier (QEC) achieved F1 scores comparable to the random forest, and a bagged variational ensemble achieved a significantly higher F1 (p = 0.018 by t-test) on the full feature set. The QEC needed only two to four training samples to reach this performance, and a 56-qubit hardware run with Pauli twirling and dynamical decoupling reached a weighted F1 similar to simulation. The authors read this as evidence that quantum ensembles, by averaging over many quantum trajectories in superposition, can learn effectively from very few examples in data-constrained healthcare settings.

Load-bearing premise

The load-bearing premise is that the evaluation protocol—ten random 80/20 splits with per-split hyperparameter tuning and a t-test across splits—gives an unbiased estimate of generalization, since the random forest predicted a single class on nine of the ten full-feature splits and some model scores were reported only for splits where predictions were not pathological.

Editorial extensions

If this is right

  • The QEC's ability to match a tuned random forest with two to four training samples suggests it could be applied to other small-data healthcare problems, such as predicting response to a newly discovered therapy where only a few patient samples exist.
  • The 56-qubit hardware result, reaching a weighted F1 similar to simulation with minimal error mitigation, indicates that quantum ensembles can be deployed on noisy near-term devices without elaborate error correction.
  • The circuit depth grows rapidly with control-register size, number of swaps, and training size, so improving qubit connectivity or using circuit-knitting techniques could make larger quantum ensembles practical.
  • The bagged variational ensemble's significantly higher F1 over the random forest, if it replicates, would make quantum bagging a candidate for high-dimensional small-sample classification tasks.
  • The paper's comparison suggests that quantum ensemble methods are a viable alternative to quantum deep learning for data-constrained life-science problems, replacing deep circuits with ensembles of shallow weak learners.

Reading between the lines

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

  • The significance test behind the bagging advantage (p = 0.018) is computed across ten splits without multiple-testing correction across the seven classifier types; a stronger test would be to preregister a single evaluation protocol and see whether the advantage survives.
  • The random forest's collapse to predicting a single class on nine of the ten full-feature splits complicates the comparison; testing the same quantum ensembles on a dataset where the classical baseline behaves well would clarify whether the quantum methods are genuinely more robust or just less prone to this specific pathology.
  • The QEC's success with very few training samples could be tested against a purely classical analogue—an ensemble of cosine-similarity classifiers averaging over random training subsets—to see whether the quantum superposition itself adds predictive power or simply reproduces classical bagging of a weak learner.
  • The 56-qubit hardware experiment used only eight training samples; scaling up the control register and training size would show whether the QEC's performance continues to improve or saturates under hardware noise.
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 / 6 minor

Summary. The paper studies quantum ensemble classifiers for small-data healthcare and life-science problems. It adapts quantum cosine classifiers (QEC, QECRU) and constructs variational quantum ensembles (soft voting, bagging, boosting), then benchmarks them on synthetic Gaussian blobs and a 150-sample renal cell carcinoma (RCC) immunotherapy-response dataset with gene expression features. The main claimed result is that quantum ensembles achieve similar or slightly better F1/accuracy than a tuned random forest, with the bagging variational classifier significantly higher than RF (p=0.018 by t-test) on the full RCC feature set; the paper further claims that QEC performs comparably using only 2-4 training samples, and that a 56-qubit hardware QEC with Pauli twirling and dynamical decoupling reaches simulation-level F1 with a lower Brier score.

Significance. If the empirical claims held, the paper would provide useful evidence for small-data quantum machine learning in HCLS and a concrete hardware demonstration. The paper is honest in reporting that the random forest predicted a single class for 9 of 10 splits and in giving non-pathologic-split results. The overhead and scaling analysis distinguishes serial variational ensembles from parallel cosine-based ensembles. However, the headline significance result is undermined by the degenerate random forest baseline, the few-sample claim lacks a matched classical baseline, and the multiple-comparison protocol is not controlled. The exploratory contribution is real but the evidence as presented does not support the conclusions as stated.

major comments (4)
  1. [Section III.A, Table I, Fig. 3B] The only statistically significant quantum-over-classical result (bagging variational F1 > RF, p=0.018) is computed on the full RCC dataset where, as the text states, the random forest 'predicted only a single class for 9 of the 10 splits.' A classifier that emits a single class for 9 of 10 test splits has an F1 determined almost entirely by the test-set class composition, so the p-value measures improvement against a broken baseline rather than against a working classifier. The authors' own restricted analysis on non-pathologic splits reports similar mean F1 values (RF 0.52, QEC 0.58, QECRU 0.60) without significance testing. I ask the authors to re-run the comparison with a non-degenerate baseline (e.g., a class-balanced random forest, a constant-predictor baseline, or exclusion of splits where RF is degenerate), report whether the bagging advantage persists on non-pathologic splits, and state the class imbalance in the RCC dataset.
  2. [Section III.A and Section II] The p=0.018 value is selected from a large configuration grid: QEC/QECRU grids with d = [1,2,3], n_train = [2,4], n_swap = [1,2,4], and n_feature = [2,4,8]; a variational hyperparameter grid of 90 configurations; PCA component count f; and per-split RF hyperparameter search. For each classifier type the paper 'identifies the configuration with the maximal mean performance over all 10 splits' (Section III.A), and Table I reports the best-performing individual split. No multiple-testing correction is applied and no pre-specified comparison protocol is given. The authors should report the distribution of performance across all configurations and splits, or apply a suitable correction (e.g., Bonferroni or FDR), or pre-register the evaluation protocol.
  3. [Section III.A, statistical inference] The paired t-test is performed over 10 random 80/20 splits (and only 5 splits for QECRU), but these splits are not independent: the same 150 patients appear in multiple training and test sets, and per-split hyperparameter selection introduces additional dependence. Treating each split as an independent observation overstates the effective sample size. The authors should use repeated cross-validation with an appropriate estimator, such as bootstrap over patients or a corrected repeated-CV t-test, and report confidence intervals for the performance differences.
  4. [Section II.C and Section IV.A] The conclusion that QEC is applicable to data-constrained problems rests on the claim that QEC achieves its performance 'only using anywhere from 2-4 training samples.' However, the random forest baseline is trained on all training samples (120 of the 150 RCC samples per split, with per-split hyperparameter optimization), so the comparison is not matched on training-data size. To support the few-shot claim, the authors should include a classical baseline trained on the same 2-4 samples (e.g., random forest with 2-4 training samples, k-nearest neighbors, or logistic regression) and compare QEC against that baseline.
minor comments (6)
  1. [Section II.A] The text 'sampling in superposition from2 d transformations' should read 'from 2^d transformations'; the control register of size d produces 2^d superposed transformations, and the current typo obscures the encoding.
  2. [Figure 1 caption] The caption is malformed: 'A C RCC (subset feat.)RCC (all feat.)Blob B' does not correctly describe panels A-C. Please revise the caption to label each panel clearly.
  3. [Section II.B, soft-voting paragraph] The sentence 'however learning rate α=1×10^{-3}.' is incomplete; it should state that this learning rate did not reach the 70% median validation accuracy criterion.
  4. [Section V] The phrase 'with relatively review qubits' appears to be a typo; it should read 'with relatively few qubits.'
  5. [Table I] The caption says 'F1 score' while the text defines weighted F1; please make the metric name consistent and state the number of test samples for the selected split.
  6. [Figure 4 and Section III.A] The claim that the 56-qubit hardware QEC with PT and DD reaches a 'significantly lower Brier score than QEC and QECRU' is not supported by a statistical test; please report the test or reword to 'numerically lower.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all reported performances are genuine held-out evaluations against independent benchmarks, and the cited prior work is external, not self-derived.

full rationale

This paper is an empirical benchmark study; its central claims are supported by held-out evaluations on independent data, not by a derivation chain that reduces to its own inputs. The QEC/QECRU classifiers make predictions on held-out test samples from 10 random 80/20 splits of the McDermott RCC data and on 18 synthetic blob configurations; no fitted parameter is renamed as a prediction. Hyperparameter selection (4-fold CV for variational ensembles, RandomizedSearchCV on the training split for random forests) is standard model selection performed inside each training split, and the RCC test metrics are not inputs to any fit. The QEC construction is taken from an external prior work ([25], Macaluso et al.) and the RCC data from an external clinical source ([13]); neither derives from the present authors' prior claims. The self-citations ([4], [5], [7]) appear only as introductory background on quantum computing in healthcare and are not load-bearing for any reported result. No uniqueness theorem is invoked to force a modeling choice, and no known empirical pattern is renamed as unification. The potential weaknesses flagged by a skeptical reader — the random forest predicting a single class in 9 of 10 splits, the t-test on a degenerate baseline, and the absence of multiple-testing correction when the best configuration per classifier is reported — concern the statistical validity of the comparison, not circularity: the quantum classifiers' predictions are still genuine evaluations on test data. Therefore no circular step is identified under the defined criteria.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a small number of model-selection choices and domain assumptions from the prior literature, but not on new postulated physical entities. The list below captures the configuration parameters, background results, and dataset assumptions that the benchmark depends on.

free parameters (4)
  • QEC/QECRU configuration (d, n_train, n_swap, n_feature) = d in [1,2,3], n_train in [2,4] (8 on hardware), n_swap in [1,2,4], n_feature in [2,4,8] (32 on hardware)
    These configuration choices determine circuit width, depth, and performance; they are searched over rather than forced by theory.
  • Variational ensemble hyperparameters (learning rate, batch size, ensemble size) = alpha in [1e-3,1e-1], batch b in [1,2,4,8,16], ensemble size n_lell in [1,2,3,4,5,6,7]; selected via 4-fold CV…
    Hyperparameters are selected on validation performance on Gaussian blobs and reused on RCC; this is data-dependent model selection.
  • Number of PCA components f for RCC features = experimental parameter, with f=2, 8, and 32 appearing in reported runs
    The first f principal components are used as input features; f is chosen per experiment and changes the problem dimensionality.
  • Random forest hyperparameters = n_estimators 100-1000, max_depth 5-20, min_samples_split 2-10, min_samples_leaf 1-5, max_features sqrt/log2, tuned by…
    The classical baseline is itself tuned on the training portion of each split, so the comparison is between a tuned classical model and selected quantum configurations.
assumptions (6)
  • domain assumption Amplitude embedding maps n classical features into a 2^n-dimensional Hilbert space, and the resulting state encodes feature information in a way suitable for classification.
    Invoked in Section II.B to construct variational learners; the paper notes this can place overlapping-blob features near the equator of the Bloch sphere, a modeling choice not derived from data.
  • standard math The swap test yields a valid cosine-distance comparison between test and training samples in the quantum cosine classifier.
    Borrowed from [27] and used in Section II.A; no proof given, treated as background.
  • standard math Parameter-shift rules provide unbiased analytic gradients for the variational circuits.
    Invoked in Section II.B for training; standard result from [30].
  • domain assumption The 8 hand-selected genes (CD8A, CXCL9, CXCL13, IFNG, CD274, PDCD1, VHL, GZMK) are known to associate with immunotherapy response.
    Section II.D; relies on external literature [37,38], not verified in this dataset.
  • domain assumption 150 RCC samples with DESeq2 normalization and VST are sufficient to estimate generalization via 10 random 80/20 splits.
    Section II.D; the dataset is small, and split-level variability is high; the paper itself notes pathologic single-class predictions for RF.
  • domain assumption Finite-shot measurements (8192 shots) are sufficient to estimate class probabilities for the Brier score and F1.
    Used throughout; shot noise is not quantified per prediction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Ensembling Methods for Healthcare and Life Science." pith.science (2026). https://pith.science/paper/6FWGZGVT

@misc{pith2026250602213,
  author       = {Pith},
  title        = {Pith review of: Quantum Ensembling Methods for Healthcare and Life Science},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6FWGZGVT}},
  note         = {Machine review of arXiv:2506.02213}
}
read the original abstract

Learning on small data is a challenge frequently encountered in many real-world applications. In this work we study how effective quantum ensemble models are when trained on small data problems in healthcare and life sciences. We constructed multiple types of quantum ensembles for binary classification using up to 26 qubits in simulation and 56 qubits on quantum hardware. Our ensemble designs use minimal trainable parameters but require long-range connections between qubits. We tested these quantum ensembles on synthetic datasets and gene expression data from renal cell carcinoma patients with the task of predicting patient response to immunotherapy. From the performance observed in simulation and initial hardware experiments, we demonstrate how quantum embedding structure affects performance and discuss how to extract informative features and build models that can learn and generalize effectively. We present these exploratory results in order to assist other researchers in the design of effective learning on small data using ensembles. Incorporating quantum computing in these data constrained problems offers hope for a wide range of studies in healthcare and life sciences where biological samples are relatively scarce given the feature space to be explored.

Figures

Figures reproduced from arXiv: 2506.02213 by the authors.

Figure 1
Figure 1. UMAP projections of study datasets. A) Example of one tested [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Comparison of quantum ensemble constructions. Two variational ensembles (Soft Voting and Bagging), the cosine classifier, and the quantum ensemble [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Comparison of ensemble classifier performance. Mean and standard error of each metric is calculated over splits for blob (A) and RCC (B) datasets for [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparing simulation and quantum hardware QEC performance. Performance of the QEC executed on quantum hardware, [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 26 canonical work pages

  1. [1]

    Biology and medicine in the landscape of quantum advantages.Journal of the Royal Society Interface, 19(196):20220541, 2022

    Benjamin A Cordier, Nicolas PD Sawaya, Gian Giacomo Guerreschi, and Shannon K McWeeney. Biology and medicine in the landscape of quantum advantages.Journal of the Royal Society Interface, 19(196):20220541, 2022

    work page 2022

  2. [2]

    Mapping global dynamics of benchmark cre- ation and saturation in artificial intelligence.Nature Communications, 13(1):6793, 2022

    Simon Ott, Adriano Barbosa-Silva, Kathrin Blagec, Jan Brauner, and Matthias Samwald. Mapping global dynamics of benchmark cre- ation and saturation in artificial intelligence.Nature Communications, 13(1):6793, 2022

    work page 2022

  3. [3]

    J. G. Greener, S. M. Kandathil, L. Moffat, and D. T. Jones. A guide to machine learning for biologists.Nat Rev Mol Cell Biol, 23(1):40–55, 2022

    work page 2022

  4. [4]

    Flother, Gad Getz, Mark Goldsmith, Tanvi Gujarati, Aldo Guzman-Saenz, Dimitrios Iliopoulos, Gavin O

    Saugata Basu, Jannis Born, Aritra Bose, Sara Capponi, Dimitra Chalkia, Timothy A Chan, Hakan Doga, Frederik F. Flother, Gad Getz, Mark Goldsmith, Tanvi Gujarati, Aldo Guzman-Saenz, Dimitrios Iliopoulos, Gavin O. Jones, Stefan Knecht, Dhiraj Madan, Sabrina Maniscalco, Nicola Mariella, Joseph A. Morrone, Khadijeh Najafi, Pushpak Pati, Daniel Platt, Maria An...

    work page 2023

  5. [5]

    Fl ¨other, Daniel Blankenberg, Maria Demidik, Karl Jansen, Raga Krishnakumar, Rajiv Krishnakumar, Nouamane Laanait, Laxmi Parida, Carl Saab, and Filippo Utro

    Frederik F. Fl ¨other, Daniel Blankenberg, Maria Demidik, Karl Jansen, Raga Krishnakumar, Rajiv Krishnakumar, Nouamane Laanait, Laxmi Parida, Carl Saab, and Filippo Utro. How quantum computing can enhance biomarker discovery, 2025

    work page 2025

  6. [6]

    Quantum computing at the frontiers of biological sciences.Nature Methods, 18, 01 2021

    Prashant Emani, Jonathan Warrell, Alan Anticevic, Stefan Bekiranov, Michael Gandal, Michael Mcconnell, Guillermo Sapiro, Al ´an Aspuru- Guzik, Justin Baker, Matteo Bastiani, John Murray, Stamatios Sotiropou- los, Jacob Taylor, Geetha Senthil, Thomas Lehner, Mark Gerstein, and Aram Harrow. Quantum computing at the frontiers of biological sciences.Nature Me...

    work page 2021

  7. [7]

    How can quantum computing be applied in clinical trial design and optimization?Trends Pharmacol

    Hakan Doga, Aritra Bose, M Emre Sahin, Joao Bettencourt-Silva, Anh Pham, Eunyoung Kim, Alan Andress, Sudhir Saxena, Laxmi Parida, Jan Lukas Robertus, Hideaki Kawaguchi, Radwa Soliman, and Daniel Blankenberg. How can quantum computing be applied in clinical trial design and optimization?Trends Pharmacol. Sci., 45(10):880–891, October 2024

    work page 2024

  8. [8]

    Amin, Evgeny Andriyash, Jason Rolfe, Bohdan Kulchyt- skyy, and Roger Melko

    Mohammad H. Amin, Evgeny Andriyash, Jason Rolfe, Bohdan Kulchyt- skyy, and Roger Melko. Quantum boltzmann machine.Phys. Rev. X, 8:021050, May 2018

    work page 2018

Show all 39 references
  1. [9]

    Qboost: Large scale classifier training withadiabatic quantum optimization

    Hartmut Neven, Vasil S Denchev, Geordie Rose, and William G Macready. Qboost: Large scale classifier training withadiabatic quantum optimization. InAsian Conference on Machine Learning, pages 333–

  2. [10]

    Classification with quantum neural networks on near term processors.arXiv preprint arXiv:1802.06002, 2018

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

    2018 arXiv

  3. [11]

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

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

    2019

  4. [12]

    Machine learning of high dimensional data on a noisy quantum processor.npj Quantum Information, 7(1):161, 2021

    Evan Peters, Jo ˜ao Caldeira, Alan Ho, Stefan Leichenauer, Masoud Mohseni, Hartmut Neven, Panagiotis Spentzouris, Doug Strain, and Gabriel N Perdue. Machine learning of high dimensional data on a noisy quantum processor.npj Quantum Information, 7(1):161, 2021

    2021

  5. [13]

    David F McDermott, Mahrukh A Huseni, Michael B Atkins, Robert J Motzer, Brian I Rini, Bernard Escudier, Lawrence Fong, Richard W Joseph, Sumanta K Pal, James A Reeves, et al. Clinical activity and molecular correlates of response to atezolizumab alone or in combination with be...

    2018

  6. [14]

    The effects of data aggregation in statistical analysis.Geographical Analysis, 8(4):428–438, 1976

    William A V Clark and Karen L Avery. The effects of data aggregation in statistical analysis.Geographical Analysis, 8(4):428–438, 1976

    1976

  7. [15]

    Bagging predictors.Machine learning, 24:123–140, 1996

    Leo Breiman. Bagging predictors.Machine learning, 24:123–140, 1996

    1996

  8. [16]

    Experiments with a new boosting algorithm

    Yoav Freund, Robert E Schapire, et al. Experiments with a new boosting algorithm. Inicml, volume 96, pages 148–156. Citeseer, 1996

    1996

  9. [17]

    Xgboost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. Xgboost: A scalable tree boosting system. InProceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, pages 785–794, 2016

    2016

  10. [18]

    Lightgbm: A highly efficient gradient boosting decision tree.Advances in neural information processing systems, 30, 2017

    Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. Lightgbm: A highly efficient gradient boosting decision tree.Advances in neural information processing systems, 30, 2017

    2017

  11. [19]

    Catboost: unbiased boosting with categorical features.Advances in neural information processing systems, 31, 2018

    Liudmila Prokhorenkova, Gleb Gusev, Aleksandr V orobev, Anna Veronika Dorogush, and Andrey Gulin. Catboost: unbiased boosting with categorical features.Advances in neural information processing systems, 31, 2018

    2018

  12. [20]

    Circuit-centric quantum classifiers.Physical Review A, 101(3):032308, 2020

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

    2020

  13. [21]

    Effect of data encoding on the expressive power of variational quantum-machine- learning models.Physical Review A, 103(3):032430, 2021

    Maria Schuld, Ryan Sweke, and Johannes Jakob Meyer. Effect of data encoding on the expressive power of variational quantum-machine- learning models.Physical Review A, 103(3):032430, 2021

    2021

  14. [22]

    Quantum ensembles of quantum classifiers.Scientific reports, 8(1):2772, 2018

    Maria Schuld and Francesco Petruccione. Quantum ensembles of quantum classifiers.Scientific reports, 8(1):2772, 2018

    2018

  15. [23]

    On quantum ensembles of quantum classifiers.Quantum Machine Intelligence, 2:1–8, 2020

    Amira Abbas, Maria Schuld, and Francesco Petruccione. On quantum ensembles of quantum classifiers.Quantum Machine Intelligence, 2:1–8, 2020

    2020

  16. [24]

    Daniel Silver, Tirthak Patel, and Devesh Tiwari. Quilt: Effective multi- class classification on quantum computers using an ensemble of diverse quantum classifiers.Proceedings of the AAAI Conference on Artificial Intelligence, 36(8):8324–8332, Jun. 2022

    2022

  17. [25]

    Quantum ensemble for classification.arXiv preprint arXiv:2007.01028, 2020

    Antonio Macaluso, Luca Clissa, Stefano Lodi, and Claudio Sartori. Quantum ensemble for classification.arXiv preprint arXiv:2007.01028, 2020

    2007 arXiv

  18. [26]

    Supervised- learning guarantee for quantum adaboost.Physical Review Applied, 22(5):054001, 2024

    Yabo Wang, Xin Wang, Bo Qi, and Daoyi Dong. Supervised- learning guarantee for quantum adaboost.Physical Review Applied, 22(5):054001, 2024

    2024

  19. [27]

    Quantum fingerprinting.Physical review letters, 87(16):167902, 2001

    Harry Buhrman, Richard Cleve, John Watrous, and Ronald De Wolf. Quantum fingerprinting.Physical review letters, 87(16):167902, 2001

    2001

  20. [28]

    How to generate random matrices from the classical compact groups, 2007

    Francesco Mezzadri. How to generate random matrices from the classical compact groups, 2007

    2007

  21. [29]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  22. [30]

    Evaluating analytic gradients on quantum hardware

    Maria Schuld, Ville Bergholm, Christian Gogolin, Josh Izaac, and Nathan Killoran. Evaluating analytic gradients on quantum hardware. Physical Review A, 99(3):032331, 2019

    2019

  23. [31]

    A desicion-theoretic generalization of on-line learning and an application to boosting

    Yoav Freund and Robert E Schapire. A desicion-theoretic generalization of on-line learning and an application to boosting. InEuropean conference on computational learning theory, pages 23–37. Springer, 1995

    1995

  24. [32]

    Boost- ing algorithms as gradient descent.Advances in neural information processing systems, 12, 1999

    Llew Mason, Jonathan Baxter, Peter Bartlett, and Marcus Frean. Boost- ing algorithms as gradient descent.Advances in neural information processing systems, 12, 1999

    1999

  25. [33]

    Greedy function approximation: a gradient boosting machine.Annals of statistics, pages 1189–1232, 2001

    Jerome H Friedman. Greedy function approximation: a gradient boosting machine.Annals of statistics, pages 1189–1232, 2001

    2001

  26. [34]

    Random forests.Machine learning, 45:5–32, 2001

    Leo Breiman. Random forests.Machine learning, 45:5–32, 2001

    2001

  27. [35]

    Why do tree- based models still outperform deep learning on typical tabular data? Advances in neural information processing systems, 35:507–520, 2022

    L ´eo Grinsztajn, Edouard Oyallon, and Ga ¨el Varoquaux. Why do tree- based models still outperform deep learning on typical tabular data? Advances in neural information processing systems, 35:507–520, 2022

    2022

  28. [36]

    Moderated estimation of fold change and dispersion for rna-seq data with deseq2

    Michael I Love, Wolfgang Huber, and Simon Anders. Moderated estimation of fold change and dispersion for rna-seq data with deseq2. Genome biology, 15:1–21, 2014

    2014

  29. [37]

    Meta-analysis of tumor-and t cell-intrinsic mechanisms of sensitization to checkpoint inhibition.Cell, 184(3):596–614, 2021

    Kevin Litchfield, James L Reading, Clare Puttick, Krupa Thakkar, Chris Abbosh, Robert Bentham, Thomas BK Watkins, Rachel Rosenthal, Dhruva Biswas, Andrew Rowan, et al. Meta-analysis of tumor-and t cell-intrinsic mechanisms of sensitization to checkpoint inhibition.Cell, 184(3)...

    2021

  30. [38]

    A minimal gene set characterizes til specific for diverse tumor antigens across different cancer types.Nature Communications, 16(1):1070, 2025

    Zhen Zeng, Tianbei Zhang, Jiajia Zhang, Shuai Li, Sydney Connor, Boyang Zhang, Yimin Zhao, Jordan Wilson, Dipika Singh, Rima Ku- likauskas, et al. A minimal gene set characterizes til specific for diverse tumor antigens across different cancer types.Nature Communications, 16(1...

    2025

  31. [39]

    Umap: Uniform manifold approximation and projection for dimension reduction.arXiv preprint arXiv:1802.03426, 2018

    Leland McInnes, John Healy, and James Melville. Umap: Uniform manifold approximation and projection for dimension reduction.arXiv preprint arXiv:1802.03426, 2018

    2018 arXiv

Pith tools

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