pith. sign in

arxiv: 2606.02228 · v1 · pith:7K73NCUMnew · submitted 2026-06-01 · 📊 stat.ML · cs.CV· cs.LG

Bayesian meta-learning for modeling Alzheimer's disease progression

Clara Hoffmann , Nadja Klein This is my paper

Pith reviewed 2026-06-28 12:41 UTC · model grok-4.3

classification 📊 stat.ML cs.CVcs.LG
keywords Alzheimer's diseasemeta-learningBayesian inferencedisease progressionlongitudinal modelingMRIADNIpredictive distributions
0
0 comments X

The pith

A Bayesian meta-learner tailors Alzheimer's disease score predictions to each individual's history without retraining.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a Bayesian meta-learning model that learns from data across many Alzheimer's patients to predict the distribution of future disease scores for new individuals, using their current MRI volume and limited past observations. This approach avoids the need to fit separate models per patient or ignore individual differences, which standard regression and neural networks struggle with due to sparse data. It achieves this by dynamically adapting predictions, scaling linearly with more observations, and providing a guarantee of lower overconfidence in long-term forecasts than deterministic meta-learners. On ADNI data, it matches the performance of other models while excelling in long-term prediction scenarios, enabling better personalized treatment planning.

Core claim

The proposed Bayesian meta-learner is trained on multiple individuals but tailors the predictive disease score distribution to each individual's historical data. It predicts on unseen individuals without retraining, scales linearly with the number of historical observations, and is guaranteed to be less overconfident when predicting long-term disease scores compared to its deterministic counterpart, while achieving competitive performance on real-world ADNI data.

What carries the argument

Bayesian meta-learner that adapts the predictive distribution of disease scores to an individual's historical trajectory and current MRI.

If this is right

  • Predicts disease score distributions for unseen patients without any retraining.
  • Scales linearly with the number of historical observations per patient.
  • Guarantees reduced overconfidence in long-term disease score predictions relative to deterministic meta-learners.
  • Achieves performance competitive with single-task models and deterministic meta-learners on ADNI data, with particular gains in long-term progression prediction.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could apply to modeling progression in other chronic diseases with sparse longitudinal measurements.
  • Uncertainty estimates from the Bayesian approach might help clinicians prioritize interventions based on prediction reliability.
  • Combining this with additional biomarkers beyond MRI could further refine individual predictions.
  • Linear scaling suggests feasibility for large-scale clinical datasets with many patients.

Load-bearing premise

The meta-learning setup can extract useful individual-level correlation structure from the limited observations per patient across the training population.

What would settle it

Demonstrating that the Bayesian model is more overconfident or performs worse than the deterministic version on long-term disease score predictions in a held-out ADNI test set would falsify the central claims.

Figures

Figures reproduced from arXiv: 2606.02228 by Clara Hoffmann, Nadja Klein.

Figure 1
Figure 1. Figure 1: Averaged axial MRI slice (across 10 individuals [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the meta-learning architecture. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Discretized sample from a Gaussian process used to model discrete disease score [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: MAP prediction ( ) and predictive uncertainty (red) of a deterministic meta￾learner (left) and Bayesian meta-learner (right) for an observed disease trajectory ( ) close to the context points (•; gray shaded area) and extrapolation regime. The deterministic meta-learner is overconfident ( ) far from the context points, while our Bayesian meta￾learner mitigates overconfidence by assigning probability mass m… view at source ↗
read the original abstract

Predicting whether an individual with Alzheimer's disease will experience mild or severe disease progression is essential for personalized treatment. Typically, practitioners seek to predict the distribution of a discrete disease score, conditional on an individual's current MRI volume and their historical disease trajectory. Classical statistical regression models and single-task neural networks are not well-suited for this purpose because fitting separate models is infeasible (since each individual typically has few observations), while ignoring individual-level correlation leads to poor generalization. Meta-learning, in contrast, provides a natural avenue to dynamically predict distributions without retraining and model nonlinear relationships between the outcome and covariates. Motivated by this, we propose a Bayesian meta-learner that is trained on multiple individuals but tailors the predictive disease score distribution to each individual's historical data. Our model predicts on unseen individuals without retraining, scales linearly with the number of historical observations, and is guaranteed to be less overconfident when predicting long-term disease scores compared to its deterministic counterpart. On real-world data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, our model achieves performance competitive with both single-task models and deterministic meta-learners, while substantially improving performance when predicting long-term disease progression.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper proposes a Bayesian meta-learner trained across multiple Alzheimer's patients to predict distributions of discrete disease scores for unseen individuals, conditioning on current MRI volume and historical trajectories. It claims the approach requires no per-patient retraining, scales linearly with the number of observations, is guaranteed to produce less overconfident long-term predictions than its deterministic counterpart, and achieves competitive performance on ADNI data.

Significance. If the claimed guarantee of reduced overconfidence holds under the paper's modeling assumptions and the empirical results are robust, the work would offer a practical advance for individualized long-term prognosis in Alzheimer's by combining meta-learning's adaptability with Bayesian calibration, addressing the data scarcity per patient that defeats single-task models.

major comments (1)
  1. [Abstract] Abstract: the load-bearing claim that the model 'is guaranteed to be less overconfident when predicting long-term disease scores compared to its deterministic counterpart' is stated without any derivation, theorem, or set of sufficient conditions (e.g., correct model specification, conjugate structure, or bounds on misspecification). The weakest_assumption note confirms no additional assumptions on the disease trajectory distribution are invoked, so the dominance in calibration is not established.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for highlighting the unsubstantiated claim in the abstract. We address the point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the load-bearing claim that the model 'is guaranteed to be less overconfident when predicting long-term disease scores compared to its deterministic counterpart' is stated without any derivation, theorem, or set of sufficient conditions (e.g., correct model specification, conjugate structure, or bounds on misspecification). The weakest_assumption note confirms no additional assumptions on the disease trajectory distribution are invoked, so the dominance in calibration is not established.

    Authors: We agree that the abstract asserts a guarantee of reduced overconfidence without a supporting derivation, theorem, or explicit set of sufficient conditions. The manuscript contains no formal proof of dominance in calibration under the stated modeling assumptions, and the weakest_assumption note indeed invokes no further restrictions on the trajectory distribution. We will revise the abstract (and any corresponding claims in the introduction) to remove the word 'guaranteed' and instead describe the property as an expected consequence of Bayesian marginalization that is supported by the empirical results on ADNI data. revision: yes

Circularity Check

0 steps flagged

No circularity detected; claims rest on standard Bayesian/meta-learning properties without reduction to inputs

full rationale

The provided abstract and context describe model properties (no-retraining prediction, linear scaling, reduced overconfidence vs deterministic counterpart) but contain no equations, fitting procedures, or self-citations that reduce any prediction or guarantee to a fitted parameter or prior result by construction. No load-bearing self-citation chain, ansatz smuggling, or self-definitional step is visible. The derivation is treated as self-contained per the rules, with the 'guarantee' presented as a stated property rather than derived from the paper's own inputs in a circular manner. This is the expected honest non-finding for papers without exhibited reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated or derivable.

pith-pipeline@v0.9.1-grok · 5733 in / 1041 out tokens · 27922 ms · 2026-06-28T12:41:29.406671+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

73 extracted references · 6 canonical work pages

  1. [1]

    Advances in Neural Information Processing Systems , volume=

    Matching networks for one shot learning , author=. Advances in Neural Information Processing Systems , volume=

  2. [2]

    Advances in Neural Information Processing Systems , volume=

    Fast and flexible multi-task classification using conditional neural adaptive processes , author=. Advances in Neural Information Processing Systems , volume=

  3. [3]

    7th International Conference on Learning Representations (ICLR) , year =

    Meta-Learning Probabilistic Inference for Prediction , author =. 7th International Conference on Learning Representations (ICLR) , year =

  4. [4]

    2010 , publisher=

    Petersen, Ronald Carl and Aisen, Paul S and Beckett, Laurel A and Donohue, Michael C and Gamst, Anthony Collins and Harvey, Danielle J and Jack Jr, CR and Jagust, William J and Shaw, Leslie M and Toga, Arthur W and others , journal=. 2010 , publisher=

    2010

  5. [5]

    and Fitsiori, A

    Fischbach-Boulanger, C. and Fitsiori, A. and Noblet, V. and Baloglu, S. and Oesterle, H. and Draghici, S. and Philippi, N. and Duron, E. and Hanon, O. and Dietemann, J. L. and Blanc, F. and Kremer, S. , title =. European Journal of Neurology , year =. doi:10.1111/ene.13601 , pmid =

    work page doi:10.1111/ene.13601

  6. [6]

    Rationale for use of the

    Cedarbaum, Jesse M and Jaros, Mark and Hernandez, Chito and Coley, Nicola and Andrieu, Sandrine and Grundman, Michael and Vellas, Bruno and. Rationale for use of the. Alzheimer's & Dementia , volume=. 2013 , publisher=

    2013

  7. [7]

    Killiany, R. J. and Moss, M. B. and Albert, M. S. and Sandor, T. and Tieman, J. and Jolesz, F. , title =. Archives of Neurology , year =. doi:10.1001/archneur.1993.00540090052010 , pmid =

    work page doi:10.1001/archneur.1993.00540090052010 1993

  8. [8]

    Kesslak, J. P. and Nalcioglu, O. and Cotman, C. W. , title =. Neurology , year =. doi:10.1212/wnl.41.1.51 , pmid =

    work page doi:10.1212/wnl.41.1.51

  9. [9]

    Amygdalohippocampal

    Leh. Amygdalohippocampal. AJNR. American Journal of Neuroradiology , year =

  10. [10]

    2025 , publisher=

    Wang, Haoyu and Guo, Sizheng and Ye, Jin and Deng, Zhongying and Cheng, Junlong and Li, Tianbin and Chen, Jianpin and Su, Yanzhou and Huang, Ziyan and Shen, Yiqing and others , journal=. 2025 , publisher=

    2025

  11. [11]

    arXiv preprint arXiv:1609.09106 , year=

    Hypernetworks , author=. arXiv preprint arXiv:1609.09106 , year=

    Pith/arXiv arXiv

  12. [12]

    Artificial Intelligence Review , volume=

    A brief review of hypernetworks in deep learning , author=. Artificial Intelligence Review , volume=. 2024 , publisher=

    2024

  13. [13]

    Deep learning using rectified linear units (

    Agarap, Abien Fred , journal=. Deep learning using rectified linear units (

  14. [14]

    Obtaining well calibrated probabilities using

    Naeini, Mahdi Pakdaman and Cooper, Gregory and Hauskrecht, Milos , booktitle=. Obtaining well calibrated probabilities using

  15. [15]

    A scalable

    Ritter, Hippolyt and Botev, Aleksandar and Barber, David , booktitle=. A scalable. 2018 , organization=

    2018

  16. [16]

    1992 , publisher=

    MacKay, David JC , journal=. 1992 , publisher=

    1992

  17. [17]

    Kristiadi, Agustinus and Hein, Matthias and Hennig, Philipp , booktitle=. Being. 2020 , organization=

    2020

  18. [18]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Meta-learning in neural networks: A survey , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2021 , publisher=

    2021

  19. [19]

    IEEE Transactions on Emerging Topics in Computational Intelligence , year=

    Exploring the Horizons of Meta-Learning in Neural Networks: A Survey of the State-of-the-Art , author=. IEEE Transactions on Emerging Topics in Computational Intelligence , year=

  20. [20]

    Proceedings of the 35th International Conference on Machine Learning , pages =

    Conditional Neural Processes , author =. Proceedings of the 35th International Conference on Machine Learning , pages =. 2018 , editor =

    2018

  21. [21]

    arXiv preprint arXiv:1912.10985 , year=

    Backpack: Packing more into backprop , author=. arXiv preprint arXiv:1912.10985 , year=

    arXiv 1912

  22. [22]

    The Annals of Applied Statistics , volume=

    Marginally calibrated response distributions for end-to-end learning in autonomous driving , author=. The Annals of Applied Statistics , volume=. 2023 , publisher=

    2023

  23. [23]

    Scalable

    Snoek, Jasper and Rippel, Oren and Swersky, Kevin and Kiros, Ryan and Satish, Nadathur and Sundaram, Narayanan and Patwary, Mostofa and Prabhat, Mr and Adams, Ryan , booktitle =. Scalable. 2015 , editor =

    2015

  24. [24]

    Carlos Riquelme and George Tucker and Jasper Snoek , year=. Deep

  25. [25]

    Variational

    Harrison, James and Willes, John and Snoek, Jasper , booktitle=. Variational

  26. [26]

    Neural Computation , volume=

    Long short-term memory , author=. Neural Computation , volume=

  27. [27]

    Cognitive science , volume=

    Finding structure in time , author=. Cognitive science , volume=. 1990 , publisher=

    1990

  28. [28]

    Patient subtyping via time-aware

    Baytas, Inci M and Xiao, Cao and Zhang, Xi and Wang, Fei and Jain, Anil K and Zhou, Jiayu , booktitle=. Patient subtyping via time-aware

  29. [29]

    arXiv preprint arXiv:1611.01491 , year=

    Understanding deep neural networks with rectified linear units , author=. arXiv preprint arXiv:1611.01491 , year=

    Pith/arXiv arXiv

  30. [30]

    Partially-shared Imaging Regression on Integrating Heterogeneous Brain-Cognition Associations across

    Sui, Yang and Xu, Qi and Li, Ting and Bai, Yang and Qu, Annie , journal=. Partially-shared Imaging Regression on Integrating Heterogeneous Brain-Cognition Associations across

  31. [31]

    A foundation model for generalized brain

    Tak, Divyanshu and Garomsa, Biniam A and Chaunzwa, Tafadzwa L and Zapaishchykova, Anna and Pardo, Juan Carlos Climent and Ye, Zezhong and Zielke, John and Ravipati, Yashwanth and Vajapeyam, Sri and Mahootiha, Maryam and others , journal=. A foundation model for generalized brain. doi:10.1038/s41593-026-02202-6 , year=

    work page doi:10.1038/s41593-026-02202-6

  32. [32]

    Building a General

    Kaczmarek, Emily and Szeto, Justin and Nichyporuk, Brennan and Arbel, Tal , journal=. Building a General

  33. [33]

    Anatomical foundation models for brain

    Barbano, Carlo Alberto and Brunello, Matteo and Dufumier, Benoit and Grangetto, Marco , journal=. Anatomical foundation models for brain

  34. [34]

    Paszke, Adam and Gross, Sam and Massa, Francisco and Lerer, Adam and Bradbury, James and Chanan, Gregory and Killeen, Trevor and Lin, Zeming and Gimelshein, Natalia and Antiga, Luca and Desmaison, Alban and Kopf, Andreas and Yang, Edward and DeVito, Zachary and Raison, Martin and Tejani, Alykhan and Chilamkurthy, Sasank and Steiner, Benoit and Fang, Lu an...

  35. [35]

    Neural pre-processing: A learning framework for end-to-end brain

    He, Xinzi and Wang, Alan Q and Sabuncu, Mert R , booktitle=. Neural pre-processing: A learning framework for end-to-end brain. 2023 , organization=

    2023

  36. [36]

    How well do the

    Benge, Jared F and Balsis, Steve and Geraci, Lisa and Massman, Paul J and Doody, Rachelle S , journal=. How well do the. 2009 , publisher=

    2009

  37. [37]

    Assessing clinical progression measures in

    McLaughlin, Jonathan and Scotton, William J and Ryan, Natalie S and Hardy, John A and Shoai, Maryam , journal=. Assessing clinical progression measures in. 2024 , publisher=

    2024

  38. [38]

    Annals of biomedical engineering , volume=

    The shrinking brain: cerebral atrophy following traumatic brain injury , author=. Annals of biomedical engineering , volume=. 2019 , publisher=

    2019

  39. [39]

    Brain shape changes associated with cerebral atrophy in healthy aging and

    Blinkouskaya, Yana and Weickenmeier, Johannes , journal=. Brain shape changes associated with cerebral atrophy in healthy aging and. 2021 , publisher=

    2021

  40. [40]

    Jack Jr, Clifford R and Lowe, Val J and Weigand, Stephen D and Wiste, Heather J and Senjem, Matthew L and Knopman, David S and Shiung, Maria M and Gunter, Jeffrey L and Boeve, Bradley F and Kemp, Bradley J and others , journal=. Serial. 2009 , publisher=

    2009

  41. [41]

    Neurology , volume=

    -Amyloid burden in healthy aging: regional distribution and cognitive consequences , author=. Neurology , volume=. 2012 , publisher=

    2012

  42. [42]

    Science translational medicine , volume=

    Testing the right target and right drug at the right stage , author=. Science translational medicine , volume=. 2011 , publisher=

    2011

  43. [43]

    Update on hypothetical model of

    Jack Jr, Clifford R and Knopman, David S and Jagust, William J and Petersen, Ronald C and Weiner, Michael W and Aisen, Paul S and Shaw, Leslie M and Vemuri, Prashanthi and Wiste, Heather J and Weigand, Stephen D and others , journal=. Update on hypothetical model of

  44. [44]

    Cerebrospinal fluid levels of -amyloid 1-42, but not of tau, are fully changed already 5 to 10 years before the onset of

    Buchhave, Peder and Minthon, Lennart and Zetterberg, Henrik and Wallin, Asa K and Blennow, Kaj and Hansson, Oskar , journal=. Cerebrospinal fluid levels of -amyloid 1-42, but not of tau, are fully changed already 5 to 10 years before the onset of. 2012 , publisher=

    2012

  45. [45]

    Duration of preclinical, prodromal, and dementia stages of

    Vermunt, Lisa and Sikkes, Sietske AM and Van Den Hout, Ardo and Handels, Ron and Bos, Isabelle and Van Der Flier, Wiesje M and Kern, Silke and Ousset, Pierre-Jean and Maruff, Paul and Skoog, Ingmar and others , journal=. Duration of preclinical, prodromal, and dementia stages of. 2019 , publisher=

    2019

  46. [46]

    Suitability of the Clinical Dementia Rating-Sum of Boxes as a single primary endpoint for

    Coley, Nicola and Andrieu, Sandrine and Jaros, Mark and Weiner, Michael and Cedarbaum, Jesse and Vellas, Bruno , journal=. Suitability of the Clinical Dementia Rating-Sum of Boxes as a single primary endpoint for. 2011 , publisher=

    2011

  47. [47]

    The British journal of psychiatry , volume=

    A new clinical scale for the staging of dementia , author=. The British journal of psychiatry , volume=. 1982 , publisher=

    1982

  48. [48]

    The Clinical Dementia Rating (

    Morris, John C , journal=. The Clinical Dementia Rating (. 1993 , publisher=

    1993

  49. [49]

    Mild senile dementia of the

    Berg, Leonard and Miller, J Philip and Storandt, Martha and Duchek, Janet and Morris, John C and Rubin, Eugene H and Burke, William J and Coben, Lawrence A , journal=. Mild senile dementia of the. 1988 , publisher=

    1988

  50. [50]

    Journal of Machine Learning Research , volume=

    Integrating random effects in deep neural networks , author=. Journal of Machine Learning Research , volume=

  51. [51]

    Denoising diffusion probabilistic models for

    Khader, Firas and M. Denoising diffusion probabilistic models for. Scientific Reports , volume=. 2023 , publisher=

    2023

  52. [52]

    Journal of the American Statistical Association , volume=

    Tensor regression with applications in neuroimaging data analysis , author=. Journal of the American Statistical Association , volume=. 2013 , publisher=

    2013

  53. [53]

    Practical

    Botev, Aleksandar and Ritter, Hippolyt and Barber, David , year =. Practical. Proceedings of the 34th International Conference on Machine Learning - Volume 70 , pages =

  54. [54]

    Advances in Neural Information Processing Systems , volume=

    Simple and principled uncertainty estimation with deterministic deep learning via distance awareness , author=. Advances in Neural Information Processing Systems , volume=

  55. [55]

    Advances in Neural Information Processing Systems , volume=

    Attentive state-space modeling of disease progression , author=. Advances in Neural Information Processing Systems , volume=

  56. [56]

    Zhang, Yuan , booktitle=

  57. [57]

    Efficient learning of continuous-time hidden

    Liu, Yu-Ying and Li, Shuang and Li, Fuxin and Song, Le and Rehg, James M , journal=. Efficient learning of continuous-time hidden

  58. [58]

    Advances in neural information processing systems , volume=

    Predictive uncertainty estimation via prior networks , author=. Advances in neural information processing systems , volume=

  59. [59]

    Matthias Hein and Maksym Andriushchenko and Julian Bitterwolf , booktitle =. Why

  60. [60]

    2021 , issn =

    Computer Methods and Programs in Biomedicine , pages =. 2021 , issn =. doi:10.1016/j.cmpb.2021.106236 , author =

    work page doi:10.1016/j.cmpb.2021.106236 2021

  61. [61]

    Standardization of analysis sets for reporting results from

    Wyman, Bradley T and Harvey, Danielle J and Crawford, Karen and Bernstein, Matt A and Carmichael, Owen and Cole, Patricia E and Crane, Paul K and DeCarli, Charles and Fox, Nick C and Gunter, Jeffrey L and others , journal=. Standardization of analysis sets for reporting results from. 2013 , publisher=

    2013

  62. [62]

    Data Leakage in Deep Learning for

    Young, Vanessa M and Gates, Samantha and Garcia, Layla Y and Salardini, Arash , journal=. Data Leakage in Deep Learning for. 2025 , publisher=

    2025

  63. [63]

    Advances in Neural Information Processing Systems , volume=

    Incorporating second-order functional knowledge for better option pricing , author=. Advances in Neural Information Processing Systems , volume=

  64. [64]

    International Conference on Learning Representations (ICLR) , year=

    Deep anomaly detection with outlier exposure , author=. International Conference on Learning Representations (ICLR) , year=

  65. [65]

    Mathematics of control, signals and systems , volume=

    Approximation by superpositions of a sigmoidal function , author=. Mathematics of control, signals and systems , volume=. 1989 , publisher=

    1989

  66. [66]

    Neural networks , volume=

    Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks , author=. Neural networks , volume=. 1990 , publisher=

    1990

  67. [67]

    A randomized gradient-free attack on

    Croce, Francesco and Hein, Matthias , booktitle=. A randomized gradient-free attack on. 2019 , organization=

    2019

  68. [68]

    2008 , publisher=

    Generalized, linear, and mixed models , author=. 2008 , publisher=

    2008

  69. [69]

    Neural networks , volume=

    Approximation capabilities of multilayer feedforward networks , author=. Neural networks , volume=. 1991 , publisher=

    1991

  70. [70]

    Archives of neurology , volume=

    Staging dementia using Clinical Dementia Rating Scale Sum of Boxes scores: a Texas Alzheimer's research consortium study , author=. Archives of neurology , volume=. 2008 , publisher=

    2008

  71. [71]

    7th International Conference on Learning Representations (ICLR) , year =

    Attentive neural processes , author=. 7th International Conference on Learning Representations (ICLR) , year =

  72. [72]

    Bayesian batch active learning as sparse subset approximation , volume =

    Pinsler, Robert and Gordon, Jonathan and Nalisnick, Eric and Hern\'. Bayesian batch active learning as sparse subset approximation , volume =. Advances in Neural Information Processing Systems , editor =

  73. [73]

    Local linear forests

    Marginally. Journal of Computational and Graphical Statistics , author =. 2020 , pages =. doi:10.1080/10618600.2020.1807996 , number =

    work page doi:10.1080/10618600.2020.1807996 2020