A Novel XAI-Enhanced Quantum Adversarial Networks for Velocity Dispersion Modeling in MaNGA Galaxies

Aik Beng Ng; Aman Chadha; Aswath Babu H; Manish M; Manjunath K Vanahalli; N V Saran Kumar; Sathwik Narkedimilli; Vinija Jain

REVIEW 5 major objections 8 minor 22 references

A Novel XAI-Enhanced Quantum Adversarial Networks for Velocity Dispersion Modeling in MaNGA Galaxies

T0 review · 5 major / 8 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a hybrid four-qubit quantum neural network, jointly trained with a LIME-guided adversarial evaluator, predicts galaxy velocity dispersion with RMSE 0.27 and R² 0.59, and that removing the quantum layer degrades RMSE to

desk verdict Incremental QML application whose statistical claims don't survive contact with a 3-fold CV. read the letter →

arxiv 2510.24598 v2 pith:A7Q6EDW3 submitted 2025-10-28 cs.LG cs.CR

Sathwik Narkedimilli , N V Saran Kumar , Aswath Babu H , Manjunath K Vanahalli , Manish M , Aik Beng Ng , Vinija Jain , Aman Chadha This is my paper · ORCID

classification cs.LGcs.CR

keywords quantummachinelearninghybridneuralnetworkvelocitydispersionMaNGAgalaxiesexplainableAILIMEadversarialtraininggalaxykinematics

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 tries to establish that a hybrid quantum-classical neural network—classical layers feeding a four-qubit circuit, with an adversarial evaluator trained on LIME explanations—can predict galaxy velocity dispersion from MaNGA data with balanced accuracy (RMSE 0.27, R² 0.59). The authors argue the quantum layer is load-bearing: ablating it raises RMSE to 0.32, while removing the evaluator feedback or classical layers also hurts. They also test GAN and self-supervised variants, but the plain hybrid remains the most stable. The intended payoff is a lightweight, interpretable quantum-enhanced regression model that works on real survey data and could transfer to other accuracy-plus-interpretability domains. The statistical case for the vanilla model's superiority, however, rests on paired tests over only three cross-validation folds.

What carries the argument

The load-bearing mechanism is the hybrid QNN plus evaluator loop. The QNN has five dense layers that map eight input features to eight parameters; those parameters drive Rx and Ry rotations on four qubits, and the circuit's expectation value under a Pauli-Z Hamiltonian becomes the prediction after a sigmoid, with gradients from the parameter-shift rule. In parallel, LIME generates local feature-attribution explanations for each prediction; the original features, prediction, and explanations are concatenated and fed to a small evaluator network whose MSE loss is added, with weight α=0.5, to the QNN's prediction loss. This coupling is what the paper claims enforces consistency between accuracy

What would settle it

Run Vanilla and Q-GAN-1 with 10-fold cross-validation repeated over several random seeds, and also compare the full model against a classical network with the same parameter count. If the RMSE/R² differences between Vanilla and Q-GAN-1 shrink to noise, or if removing the quantum layer no longer worsens RMSE by ~0.05, the paper's central claims of statistical superiority and quantum criticality would be refuted.

Watch

Extended reading notes

Core claim

On the authors' own terms, the central discovery is that an adversarial evaluator using LIME explanations can guide a hybrid quantum neural network toward both accurate and interpretable regression. The vanilla version of this architecture reaches an RMSE of 0.27, MSE 0.071, MAE 0.21, and R² 0.59 on a roughly 2,000-galaxy MaNGA sample, outperforming its Q-GAN and quantum self-supervised variants in overall balance. The ablation study is the key evidence: removing the quantum layer raises RMSE to 0.32 and lowers R² to 0.53, which the authors read as proof that the quantum component contributes beyond the classical layers. They also position the model as lightweight (about 46,700 trainable par

Load-bearing premise

The claim that the Vanilla model is significantly better than its variants rests on paired t-tests and Wilcoxon tests computed from only three cross-validation fold means; with two degrees of freedom, tiny fold-to-fold shifts could flip the conclusions.

Editorial extensions

If this is right

Quantum-circuit layers can be dropped into a standard regression pipeline and, at least on this dataset, appear to improve accuracy over the classical-only version (RMSE 0.27 vs 0.35 in the classical counterpart, 0.32 without the quantum layer).
The adversarial evaluator with LIME feedback yields calibration errors (ECE 0.015) lower than the GAN and self-supervised variants, suggesting interpretability feedback helps calibration, not just interpretability.
Because the vanilla model needs only ~46,700 parameters and 1.8 ms per sample at inference, hybrid quantum-classical models of this kind are cheap enough for survey-scale astronomy.
The GAN and self-supervised extensions do not beat the plain hybrid, implying that added adversarial complexity is not automatically beneficial.
Classical baselines such as decision trees (RMSE 0.18) and linear regression (RMSE 0.19) still outperform the quantum hybrid on raw error, so the model's value lies in the accuracy-interpretability trade-off rather than in beating classical regressors.

Reading between the lines

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

Replicating with more folds or repeated shuffled splits would be the fastest way to check whether the reported significance (p-values from paired tests over only three fold means) survives; the differences between Vanilla and Q-GAN-2 already sit near the threshold.
The ablation removes whole components, so it does not isolate whether the quantum layer's benefit comes from its parameter count or nonlinearity; an equally sized classical network with the same number of trainable parameters would be a sharper control.
If the LIME-guided evaluator is the active ingredient, the same feedback loop should improve classical regressors too; testing the evaluator on a purely classical network would separate quantum effects from feedback effects.
All reported quantum results use a noiseless GPU simulator, so the claimed quantum-layer contribution is untested under real device noise; a hardware experiment with the same 4-qubit circuit would reveal how much of the benefit survives.

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.

Desk Editor's Note

read the letter

Two things you should know about arXiv:2510.24598. First, it is a clearly written application of a simulated hybrid QNN combined with LIME and an adversarial evaluator to predict velocity dispersion from MaNGA data. Second, the main statistical support is internally inconsistent: Section 6.1 says paired Wilcoxon signed-rank tests were run on fold means from a 3-fold CV, n=3. For n=3, the smallest possible two-sided p-value is 0.25. Table 5 reports p-w values like 0.027, 0.028, 0.030, 0.034. Those numbers cannot come from the stated test. The paired t-test with df=2 could produce small p-values, but the Wilcoxon results cannot. This is a load-bearing flaw, not a cosmetic one, because the entire "Vanilla model significantly outperforms" conclusion rests on it.

What is actually new is narrow: the specific combination of a hybrid QNN (CUDA-Quantum, simulated), LIME-guided feedback, and a separate evaluator for the Li et al. MaNGA dataset. The architecture is specified in enough detail to reimplement (pseudocode, layer sizes, hyperparameters), and the resource profiling with 1–4 qubits is a nice addition. The feature importance result—enclosed mass dominant—is physically sensible. The paper also contains an honest limitation paragraph that admits the margins are narrow.

Soft spots, in order:

1. The Wilcoxon impossibility. If the test was actually run on per-sample residuals, the stated unit of analysis is wrong; if run on fold means, the reported p-values are impossible. Either way, the significance claims are unreliable. 2. No code or data release, and outlier removal is described as "identified and excluded" without criteria. Combined with 3-fold CV, the reported confidence intervals in Table 4 are also tighter than they should be (they effectively assume no between-fold variance in the point estimates). 3. The classical baseline comparison is misleading. Decision Tree and Linear Regression in Table 10 both achieve lower RMSE (0.18, 0.19) than the Vanilla model (0.27), with comparable R². The text claims the Vanilla is "outperforming or matching" but that's only true among the QML variants. 4. The ablation shows removing the quantum layer raises RMSE to 0.32, but that is still worse than a simple Decision Tree. So the "quantum layer is critical" claim is relative to their own stack, not to the field. 5. The "Classical Counterpart" section (6.5) is confusing: it describes a model with RMSE 0.35, R² 0.38, but that model doesn't appear in Table 10, and the sentence "we discuss the proposed Vanilla model, which does not include quantum components" reads like a copy-paste error.

Who gets value: someone working on the mechanics of combining LIME-based explanations with a simulated QNN for regression, or someone who wants a case study of common statistical errors in QML papers. Not someone who needs a new method for velocity dispersion—the R² of 0.59 is modest and beaten by off-the-shelf classical methods.

My recommendation: this deserves peer review as an incremental application, not a desk reject, because the architecture is concrete and the flaws are fixable. But the referee should be told to focus on the statistics and the baseline framing. If the editor wants to desk reject, the impossible Wilcoxon p-values give grounds; I'd rather see the authors get the chance to fix it.

Referee Report

5 major / 8 minor

Summary. The manuscript proposes an XAI-enhanced quantum adversarial network for predicting galaxy velocity dispersion (log sigma_e) from MaNGA data. The core 'Vanilla' model combines a hybrid QNN (classical dense layers plus a 4-qubit parameterized circuit) with an adversarial evaluator that receives LIME explanations and feeds back into the QNN loss. Three variants are introduced: Q-GAN-1, Q-GAN-2, and Q-Self-Supervised. The paper reports Vanilla metrics of RMSE=0.27, MSE=0.071, MAE=0.21, R^2=0.59, and claims it is the 'most balanced' model based on 3-fold cross-validation, paired significance tests, robustness/calibration analyses, ablations, and baseline comparisons.

Significance. If the empirical claims are validated, the framework would offer a lightweight, interpretable approach to hybrid quantum-classical regression in astrophysics. The paper's strengths include a complete pseudocode (Algorithm 1), explicit mathematical formulation of the quantum layer and evaluator (Section 5.2), an ablation study isolating the quantum layer (Table 11), resource profiling (Section 6.6), and calibration/uncertainty analysis (Section 6.2). However, the central empirical claim is currently undermined by a mathematical impossibility in the statistical reporting (Wilcoxon p-values with n=3) and by baseline comparisons in which classical models achieve lower RMSE/MSE than the proposed Vanilla model. The 'Classical Counterpart' in Section 6.5 also conflicts with the ablation in Table 11. These issues are load-bearing for the stated conclusions.

major comments (5)

[Section 6.1, Table 5] The reported Wilcoxon signed-rank p-values (e.g., 0.028, 0.034, 0.030, 0.027, 0.017) are mathematically impossible for the stated design. A paired two-sided Wilcoxon test with n=3 fold means has a minimum p-value of 0.25 (achieved when all three differences share the same sign). The paper explicitly states 'the unit of analysis was defined at the level of fold means obtained from cross-validation.' Therefore, these p-values cannot arise from the described experiment. The statistical validation of the claim that Vanilla 'consistently outperforms' Q-GAN-1 and Q-Self-Supervised is invalid as reported. Please re-run the analysis with a valid number of units (e.g., repeated CV) or correct the p-values.
[Section 6.3, Table 10] The baseline comparison actually shows that simple classical regressors outperform the proposed Vanilla model on RMSE and MSE. For instance, Decision Tree achieves RMSE=0.18, MSE=0.032, R^2=0.56; Linear Regression RMSE=0.19, MSE=0.036, R^2=0.57; Random Forest RMSE=0.21, MSE=0.044, R^2=0.55. The proposed Vanilla reports RMSE=0.27, MSE=0.071, R^2=0.59. The surrounding prose claims 'MSE around 0.07–0.09' for classical methods, which does not match the table. This discrepancy must be resolved, and the paper should explicitly discuss how the proposed method advances the state of the art when several classical baselines achieve lower error.
[Section 6.5 vs Section 6.4, Table 11] The 'Classical Counterpart' (no quantum components) is reported with RMSE≈0.35, MSE≈0.07, MAE≈0.32, R^2≈0.38 (Fig. 15). However, the ablation 'w/o Quantum Layer' in Table 11 reports RMSE=0.32, MSE=0.102, MAE=0.24, R^2=0.53. If the classical counterpart is the same architecture minus the quantum layer, these numbers should be identical; if it is a different configuration, the difference must be described. This inconsistency prevents a clean assessment of the quantum layer's contribution and undermines the ablation conclusion.
[Section 4.3] Outlier handling is not specified. The text states 'outliers were identified and excluded' but gives no method (e.g., IQR, z-score), threshold, or number of removed samples. This is a reproducibility gap and could materially affect all reported metrics. Please specify the procedure and confirm that outlier removal was done on the training set only (or, if applied to the full dataset, discuss potential information leakage).
[Section 6.1] Even setting aside the Wilcoxon issue, using only three fold means for paired t-tests leaves df=2, yielding highly unstable p-values. The p-t values in Table 5 (e.g., 0.012, 0.019, 0.008) would change dramatically under a different 3-fold split. The manuscript should use repeated k-fold cross-validation (e.g., 5x5 or 10x10) or a bootstrap to obtain more reliable inference, and should report effect sizes and confidence intervals for the differences.

minor comments (8)

[Section 4.2] The dataset size is inconsistent: 'approximately 2110 galaxies' in one paragraph and '2778 galaxies' in the next. Please clarify the exact sample size used.
[Section 4.6 vs Section 4.4] The GAN training is described as 100 epochs in the hyperparameters (Section 4.6) but as 10 epochs in the Q-GAN-1 training description (Section 4.4). These should be aligned.
[Figure 6] The caption uses 'Q-GAN-I' while the text uses 'Q-GAN-1'. Please standardize.
[Section 6.1, Table 4] The 95% confidence intervals appear too narrow for n=3. For Vanilla MSE, fold values 0.069, 0.074, 0.070 give a standard error of about 0.0015 and a t-based half-width of about 0.0065, not 0.004 as implied by [0.067, 0.075]. Please check the CI computation.
[Section 6.2] The Brier score is a classification metric; for regression, calibration is usually measured with ECE/ACE on discretized predictions. Please define the Brier score in this regression context or remove it.
[Section 6.3] The baseline methods (Attention-based Regression, VQR, Q-LR, Estimator-QNN, QML with JAX) are not described. Provide implementation details or references so the comparison is reproducible.
[Section 7] The conclusion's own limitation statement ('narrow performance margins ... require further optimization to outperform conventional methods') is in tension with the abstract's claim of 'high-performance'. The framing should be adjusted to match the evidence.
[References] References [13] and [22] are self-citations to previous work that is not directly used to ground the proposed method. Please recontextualize or remove them from the supporting citations.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; empirical evaluation with a statistical validity flaw unrelated to circularity.

full rationale

The paper is an empirical machine-learning evaluation, not a theoretical derivation, so the central claims rest on measured test-set metrics rather than on assumptions that already contain the conclusions. The proposed adversarial evaluator is a training mechanism (Algorithm 1 and Section 5.2.3), not a self-referential argument: it concatenates features, predictions, and LIME explanations and adds a feedback loss, but no equation reduces the final RMSE/MSE/MAE/R2 to that feedback by construction. Self-citations [13] and [22] are to prior work by the same authors, but they are cited only for CUDA-Quantum usage and an application example, not as the justification for the reported performance; there is no imported uniqueness theorem and no ansatz smuggled in through citation. The custom 'accuracy' metric is explicitly defined as a rescaling of the error metrics, so it is a disclosed transformation rather than a renamed result presented as independent evidence. The ablation study and baseline comparisons are empirical and could have come out differently. The one serious issue is in Section 6.1: Table 5 reports two-sided Wilcoxon signed-rank p-values as low as 0.014 with only three fold means, while the minimum possible two-sided p-value for n=3 is 0.25, making the reported significance internally inconsistent. That is a statistical-validity or reproducibility problem, not a circularity problem, because the significance claim does not reduce to its inputs by definition; it is a computational claim about the data. The paper's own conclusion even concedes 'narrow performance margins,' further indicating that the reported differences are not forced by construction. Accordingly, no circular step is present, and the score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities or forces. It is an empirical ML study resting on dataset assumptions and a collection of hand-picked hyperparameters. The central claim depends on the fidelity of the MaNGA-derived labels, the faithfulness of LIME explanations, and the validity of simulated quantum circuits as a proxy for quantum computing.

free parameters (5)

Feedback weight alpha = 0.5
Section 4.6: weighting coefficient for evaluator feedback into QNN loss. Hand-chosen, not optimized or justified.
Number of PCA components = 4
Section 4.3: PCA reduces 8 features to 4 components to fit a 4-qubit system. This choice determines the quantum circuit width and is coupled to the model design.
Qubit count = 4 (varied 1-4)
Section 6.6 profiles 1-4 qubits; the main results use 4. The choice is tied to PCA dimension and is a free architectural parameter.
Learning rates = 0.001 (QNN/Evaluator), 0.0002 (GAN)
Section 4.6: Adam optimizer settings. Standard values but not derived; affect convergence and results.
Training epochs = 10 for QNN/Evaluator, 100 for GAN
Section 4.6: number of epochs is small; loss curves show marginal convergence. A free choice that influences reported metrics.

assumptions (4)

domain assumption The MaNGA-derived galaxy properties from Li et al. [4] are accurate and the target log sigma_e is a reliable ground truth.
Section 4.2 relies on the dataset as ground truth; no independent validation of the JAM-derived velocity dispersions is performed.
domain assumption LIME provides faithful local explanations for the QNN's predictions.
Section 4.4 uses LIME explanations as part of the evaluator input and claims interpretability; faithfulness of LIME for the QNN is assumed, not tested.
domain assumption GPU simulation of quantum circuits (QPP backend) is an adequate proxy for QML behavior.
Section 4.1 explicitly states simulations are used instead of real quantum hardware; the authors acknowledge the gap but still frame results as quantum advantage.
domain assumption The train/test split is leak-free and the data preprocessing does not introduce bias.
Section 4.3 describes preprocessing including outlier removal and stratified sampling, but details of outlier exclusion and stratification are unspecified, so this assumption is unverified.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Novel XAI-Enhanced Quantum Adversarial Networks for Velocity Dispersion Modeling in MaNGA Galaxies." pith.science (2026). https://pith.science/paper/A7Q6EDW3

@misc{pith2026251024598,
  author       = {Pith},
  title        = {Pith review of: A Novel XAI-Enhanced Quantum Adversarial Networks for Velocity Dispersion Modeling in MaNGA Galaxies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A7Q6EDW3}},
  note         = {Machine review of arXiv:2510.24598}
}

read the original abstract

Current quantum machine learning approaches often face challenges balancing predictive accuracy, robustness, and interpretability. To address this, we propose a novel quantum adversarial framework that integrates a hybrid quantum neural network (QNN) with classical deep learning layers, guided by an evaluator model with LIME-based interpretability, and extended through quantum GAN and self-supervised variants. In the proposed model, an adversarial evaluator concurrently guides the QNN by computing feedback loss, thereby optimizing both prediction accuracy and model explainability. Empirical evaluations show that the Vanilla model achieves RMSE = 0.27, MSE = 0.071, MAE = 0.21, and R^2 = 0.59, delivering the most consistent performance across regression metrics compared to adversarial counterparts. These results demonstrate the potential of combining quantum-inspired methods with classical architectures to develop lightweight, high-performance, and interpretable predictive models, advancing the applicability of QML beyond current limitations.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 4 linked inside Pith

[1]

& Wadadekar, Y

Biswas, P. & Wadadekar, Y. Structure and Kinematics of Star-forming Elliptical Galaxies in SDSS-MaNGA.The Astrophysical Journal.970, 83 (2024) 27

2024
[2]

& Bianco, S

Alchieri, L., Badalotti, D., Bonardi, P. & Bianco, S. An introduction to quan- tum machine learning: from quantum logic to quantum deep learning.Quantum Machine Intelligence.3, 28 (2021)

2021
[3]

& Guestrin, C

Ribeiro, M., Singh, S. & Guestrin, C. ” Why should i trust you?” Explaining the predictions of any classifier.Proceedings Of The 22nd ACM SIGKDD Inter- national Conference On Knowledge Discovery And Data Mining. pp. 1135-1144 (2016)

2016
[4]

& Drory, N

Li, H., Mao, S., Cappellari, M., Ge, J., Long, R., Li, R., Mo, H., Li, C., Zheng, Z., Bundy, K., Thomas, D., Brownstein, J., Roman Lopes, A., Law, D. & Drory, N. SDSS-IV MaNGA: global stellar population and gra- dients for about 2000 early-type and spiral galaxies on the mass–size plane. Monthly Notices Of The Royal Astronomical Society.476, 1765-1775 (20...

doi:10.1093/mnras/sty334
[5]

& Tremaine, S

Binney, J. & Tremaine, S. Galactic Dynamics. (Princeton University Press,2008)

2008
[6]

& Others Overview of the SDSS-IV MaNGA Survey: Mapping Nearby Galaxies at Apache Point Observatory.The Astrophysical Journal.798, 7 (2015)

Bundy, K. & Others Overview of the SDSS-IV MaNGA Survey: Mapping Nearby Galaxies at Apache Point Observatory.The Astrophysical Journal.798, 7 (2015)

2015
[7]

& Emsellem, E

Cappellari, M. & Emsellem, E. Parametric Recovery of Line-of-Sight Velocity Distributions from Galaxy Spectra.Publications Of The Astronomical Society Of The Pacific.116, 138 (2004)

2004
[8]

& Others The SAURON Project – XVI

Cappellari, M. & Others The SAURON Project – XVI. Dynamical Models of Early-Type Galaxies with Dark Matter.Monthly Notices Of The Royal Astronomical Society.390, 71-96 (2008)

2008

Show all 22 references

[9]

A Numerical Model for a Triaxial Stellar System in Dynamical Equilibrium.The Astrophysical Journal.232pp

Schwarzschild, M. A Numerical Model for a Triaxial Stellar System in Dynamical Equilibrium.The Astrophysical Journal.232pp. 236-247 (1979)

1979
[10]

& Others Bayesian Neural Networks for Stellar Velocity Dispersion Esti- mation.Monthly Notices Of The Royal Astronomical Society.514, 1234-1250 (2022)

D ´ ıaz, C. & Others Bayesian Neural Networks for Stellar Velocity Dispersion Esti- mation.Monthly Notices Of The Royal Astronomical Society.514, 1234-1250 (2022)

2022
[11]

& Weedbrook, C

Lloyd, S. & Weedbrook, C. Quantum Generative Adversarial Learning.Nature Physics.14pp. 277-282 (2018)

2018
[12]

& Costa, T

Kim, J., McCaskey, A., Heim, B., Modani, M., Stanwyck, S. & Costa, T. CUDA Quantum: The Platform for Integrated Quantum-Classical Computing.2023 60th ACM/IEEE Design Automation Conference (DAC). pp. 1-4 (2023)

2023
[13]

Narkedimilli, S., Amballa, V., Kumar, N., Kumar, R., Reddy, R., Raghav, S., M, M. & H, A. Comparative Analysis of Black Hole Mass Estimation in Type-2 AGNs: Classical vs. Quantum Machine Learning and Deep Learning Approaches. (2025), https://arxiv.org/abs/2502.15297 28

2025 arXiv
[14]

& Lee, J

Kim, H., Park, S. & Lee, J. Galaxy velocity dispersion estimation with convolu- tional neural networks.Astrophysical Journal Letters.891, L35 (2020)

2020
[15]

Havl ´ ıˇ cek, V. & Al. Supervised learning with quantum-enhanced feature spaces. Nature.567pp. 209-212 (2019)

2019
[16]

& Killoran, N

Dallaire-Demers, P. & Killoran, N. Quantum generative adversarial networks. Physical Review A.98, 012324 (2018)

2018
[17]

& Thang, T

Ngo, T., Nguyen, T. & Thang, T. A survey of recent advances in quantum generative adversarial networks.Electronics.12, 856 (2023)

2023
[18]

& Liu, C

Chen, K., Li, X., Xu, X., Wang, Y. & Liu, C. Multi-GPU-Enabled Hybrid Quantum-Classical Workflow in Quantum-HPC Middleware: Applications in Quantum Simulations.ArXiv Preprint ArXiv:2403.05828. (2024)

2024 arXiv
[19]

& Jacobs, C

Fluke, C. & Jacobs, C. Surveying the reach and maturity of machine learning and artificial intelligence in astronomy.Wiley Interdisciplinary Reviews: Data Mining And Knowledge Discovery.10, e1349 (2020)

2020
[20]

Machine learning in astronomy: A practical overview.ArXiv Preprint ArXiv:1904.07248

Baron, D. Machine learning in astronomy: A practical overview.ArXiv Preprint ArXiv:1904.07248. (2019)

1904 arXiv
[21]

& Pattnaik, R

Kembhavi, A. & Pattnaik, R. Machine learning in astronomy.Journal Of Astrophysics And Astronomy.43, 76 (2022)

2022
[22]

& Mallellu, S

Narkedimilli, S., Sriram, A., Makam, S., Sathvik, M. & Mallellu, S. F APL-DM-BC: A Secure and Scalable FL Framework with Adaptive Pri- vacy and Dynamic Masking, Blockchain, and XAI for the IoVs. (2025), https://arxiv.org/abs/2501.01063 29

2025 arXiv

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

& Wadadekar, Y

Biswas, P. & Wadadekar, Y. Structure and Kinematics of Star-forming Elliptical Galaxies in SDSS-MaNGA.The Astrophysical Journal.970, 83 (2024) 27

2024

[2] [2]

& Bianco, S

Alchieri, L., Badalotti, D., Bonardi, P. & Bianco, S. An introduction to quan- tum machine learning: from quantum logic to quantum deep learning.Quantum Machine Intelligence.3, 28 (2021)

2021

[3] [3]

& Guestrin, C

Ribeiro, M., Singh, S. & Guestrin, C. ” Why should i trust you?” Explaining the predictions of any classifier.Proceedings Of The 22nd ACM SIGKDD Inter- national Conference On Knowledge Discovery And Data Mining. pp. 1135-1144 (2016)

2016

[4] [4]

& Drory, N

Li, H., Mao, S., Cappellari, M., Ge, J., Long, R., Li, R., Mo, H., Li, C., Zheng, Z., Bundy, K., Thomas, D., Brownstein, J., Roman Lopes, A., Law, D. & Drory, N. SDSS-IV MaNGA: global stellar population and gra- dients for about 2000 early-type and spiral galaxies on the mass–size plane. Monthly Notices Of The Royal Astronomical Society.476, 1765-1775 (20...

doi:10.1093/mnras/sty334

[5] [5]

& Tremaine, S

Binney, J. & Tremaine, S. Galactic Dynamics. (Princeton University Press,2008)

2008

[6] [6]

& Others Overview of the SDSS-IV MaNGA Survey: Mapping Nearby Galaxies at Apache Point Observatory.The Astrophysical Journal.798, 7 (2015)

Bundy, K. & Others Overview of the SDSS-IV MaNGA Survey: Mapping Nearby Galaxies at Apache Point Observatory.The Astrophysical Journal.798, 7 (2015)

2015

[7] [7]

& Emsellem, E

Cappellari, M. & Emsellem, E. Parametric Recovery of Line-of-Sight Velocity Distributions from Galaxy Spectra.Publications Of The Astronomical Society Of The Pacific.116, 138 (2004)

2004

[8] [8]

& Others The SAURON Project – XVI

Cappellari, M. & Others The SAURON Project – XVI. Dynamical Models of Early-Type Galaxies with Dark Matter.Monthly Notices Of The Royal Astronomical Society.390, 71-96 (2008)

2008