pith. sign in

arxiv: 2606.30329 · v1 · pith:PIJKHSF3new · submitted 2026-06-29 · 🧬 q-bio.NC

Cohort-amortized personalization: navigating the privacy-utility frontier for virtual brain twins

Amirhossein Esmaeili , Marmaduke Woodman , Nina Baldy , Abolfazl Ziaeemehr , Julia Makhalova , Huifang Wang , Daniele Marinazzo , Svenja Caspers
show 3 more authors
Fabrice Bartolomei Meysam Hashemi Viktor Jirsa
This is my paper

Pith reviewed 2026-06-30 03:18 UTC · model grok-4.3

classification 🧬 q-bio.NC
keywords cohort-amortized personalizationvirtual brain twinsprivacy-utility tradeoffepileptogenic zone localizationage predictionneural density estimatorcross-atlas autoencodermechanistic brain modeling
0
0 comments X

The pith

Cohort-amortized personalization matches per-subject brain model accuracy while reducing fitting time from hours to seconds.

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

Personalized generative brain models require individual neuroimaging data, yet privacy rules and re-identification risks block data sharing while per-subject fitting costs hours of compute. The paper introduces cohort-amortized personalization, which trains a neural density estimator once on simulations from a mechanistic whole-brain model under a low-rank cohort prior and distributes only the compact estimator. New subjects then personalize in seconds using their own data alone. A cross-atlas autoencoder maps connectomes from twenty anatomical atlases into one shared latent space so the estimator works across sites with different parcellations. Validation on twenty-one epilepsy patients and eight hundred thirty-two aging subjects shows the method matches or exceeds per-subject inference accuracy.

Core claim

Cohort-amortized personalization replaces data sharing with model sharing: a neural density estimator is trained on simulations from a mechanistic whole-brain model under a low-rank cohort prior, and only the compact estimator is distributed, so new subjects are personalized in seconds on their own data alone. To make this prior both compact and atlas-independent, a cross-atlas autoencoder maps connectomes from twenty anatomical atlases into a shared latent space. The approach is validated on two cohorts where it matches or exceeds per-subject inference with a hours-to-seconds speed-up and can serve as a mechanistic surrogate supporting in-silico experimentation and synthetic-cohort generati

What carries the argument

Cohort-amortized personalization (CAP), the procedure that trains and shares a neural density estimator under a low-rank cohort prior together with the cross-atlas autoencoder that produces an atlas-independent latent space.

If this is right

  • Personalized brain models become feasible for clinical translation without per-subject refitting or raw data movement.
  • Multi-site studies can proceed by exchanging only the trained estimator rather than sensitive imaging files.
  • The shared estimator functions as a mechanistic surrogate that supports in-silico experiments and synthetic cohort generation under governance controls.
  • Deployment across sites remains possible even when each site uses a different anatomical atlas.
  • Wider adoption of virtual brain twins becomes possible in settings that previously lacked the compute or data-sharing infrastructure.

Where Pith is reading between the lines

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

  • The hours-to-seconds reduction could allow iterative personalization during a single clinical session rather than overnight batch jobs.
  • Standardizing across atlases may let legacy datasets with mismatched parcellations contribute to new cohort priors without reprocessing raw images.
  • The same amortization pattern could apply to other mechanistic simulators outside neuroimaging where privacy and compute costs currently limit personalization.
  • Synthetic access via the shared estimator might encourage more sites to participate in modeling consortia by lowering perceived re-identification risk.

Load-bearing premise

The low-rank cohort prior combined with the cross-atlas autoencoder preserves enough individual variability for the shared estimator to match full per-subject fitting accuracy.

What would settle it

A direct comparison in which CAP accuracy on held-out subjects falls below the per-subject baseline when either the low-rank cohort prior is replaced by an uninformative prior or the cross-atlas mapping is disabled.

read the original abstract

Personalized generative brain models require individual neuroimaging data that privacy constraints and re-identification risk make difficult to share, while per-subject fitting procedures cost hours of compute -- limiting clinical translation and multi-site collaboration. We introduce cohort-amortized personalization (CAP), which replaces data sharing with model sharing: a neural density estimator is trained on simulations from a mechanistic whole-brain model under a low-rank cohort prior, and only the compact estimator is distributed, so new subjects are personalized in seconds on their own data alone. To make this prior both compact and atlas-independent, a cross-atlas autoencoder (CrossCoder) maps connectomes from 20 anatomical atlases into a shared latent space, enabling deployment across sites with heterogeneous atlases. We validate CAP on two cohorts: 21 patients with drug-resistant epilepsy (epileptogenic-zone localization F1=0.56) and 832 subjects from the 1000BRAINS aging cohort (predicted age r=0.44); in both, CAP matches or exceeds per-subject inference with hours-to-seconds speed-up. Because the shared artifact couples a cohort prior to a mechanistic simulator, it can serve as a mechanistic surrogate supporting in-silico experimentation and synthetic-cohort generation without raw-data access -- a governance-audited alternative we term synthetic access, allowing for wider adoption of personalized modeling in more diverse settings.

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

2 major / 1 minor

Summary. The paper introduces cohort-amortized personalization (CAP), which trains a neural density estimator on simulations from a mechanistic whole-brain model under a low-rank cohort prior; a CrossCoder autoencoder maps connectomes across 20 atlases into a shared latent space so that only the compact estimator is shared. New subjects are then personalized in seconds using their own data. Validation is reported on 21 epilepsy patients (epileptogenic-zone localization F1=0.56) and 832 subjects from 1000BRAINS (age prediction r=0.44), with the claim that CAP matches or exceeds per-subject inference while providing hours-to-seconds speed-up and enabling synthetic access without raw-data sharing.

Significance. If the performance claims hold after proper verification of the low-rank prior's fidelity, the method would address a genuine barrier to clinical translation of personalized generative brain models by decoupling model sharing from data sharing and reducing compute from hours to seconds. The combination of mechanistic simulation with amortized inference and atlas-agnostic encoding is a concrete technical contribution that could support multi-site studies and in-silico experimentation under governance constraints.

major comments (2)
  1. [Abstract] Abstract (and presumably Results section): the central claim that CAP 'matches or exceeds per-subject inference' is load-bearing for the paper yet is presented without reported baseline per-subject fitting metrics, confidence intervals, data-exclusion criteria, or ablation on the latent dimension of the CrossCoder. Without these, it is impossible to verify whether the low-rank cohort prior plus cross-atlas compression preserves the subject-specific variability required for the reported F1=0.56 and r=0.44.
  2. [Abstract] Abstract: no quantitative diagnostic (reconstruction error, retained posterior variance, or latent-dimension ablation) is supplied to test the weakest assumption that the low-rank prior and CrossCoder do not systematically attenuate the signals used for epileptogenic-zone localization or age regression. If the effective rank is too low relative to inter-subject heterogeneity in the 21- and 832-subject cohorts, the amortized estimator will underperform true per-subject fitting by construction.
minor comments (1)
  1. The term 'synthetic access' is introduced without a formal definition or governance-audit protocol; a short clarifying paragraph would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for explicit verification of our central performance claims. We address each major comment below and will revise the manuscript to include the requested quantitative details.

read point-by-point responses

  1. Referee: [Abstract] Abstract (and presumably Results section): the central claim that CAP 'matches or exceeds per-subject inference' is load-bearing for the paper yet is presented without reported baseline per-subject fitting metrics, confidence intervals, data-exclusion criteria, or ablation on the latent dimension of the CrossCoder. Without these, it is impossible to verify whether the low-rank cohort prior plus cross-atlas compression preserves the subject-specific variability required for the reported F1=0.56 and r=0.44.

    Authors: We agree that the comparison to per-subject inference must be supported by explicit baseline metrics. The current manuscript states the claim based on internal comparisons but does not report the full per-subject fitting results, confidence intervals, exclusion criteria, or CrossCoder latent-dimension ablation in the abstract or main text. In the revised version we will add a dedicated results subsection (and update the abstract) that reports these quantities side-by-side with the CAP results, including 95% confidence intervals and an ablation across latent dimensions 8–64. revision: yes

  2. Referee: [Abstract] Abstract: no quantitative diagnostic (reconstruction error, retained posterior variance, or latent-dimension ablation) is supplied to test the weakest assumption that the low-rank prior and CrossCoder do not systematically attenuate the signals used for epileptogenic-zone localization or age regression. If the effective rank is too low relative to inter-subject heterogeneity in the 21- and 832-subject cohorts, the amortized estimator will underperform true per-subject fitting by construction.

    Authors: We acknowledge that the manuscript currently lacks direct quantitative checks on signal preservation. We will add, in the revised Results and Methods, (i) CrossCoder reconstruction error (MSE and cosine similarity) on held-out connectomes, (ii) comparison of posterior variance before and after amortization, and (iii) the same latent-dimension ablation already planned for the performance comparison. These diagnostics will be computed separately on the epilepsy and 1000BRAINS cohorts to confirm that the low-rank prior and atlas mapping retain the variability relevant to the downstream tasks. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation or validation chain

full rationale

The paper introduces CAP as a new amortized inference procedure that trains a neural density estimator on simulations from a mechanistic model under a low-rank cohort prior, then deploys the estimator for fast per-subject personalization. Validation consists of empirical performance metrics (F1=0.56 on epilepsy cohort, r=0.44 on aging cohort) compared against per-subject baselines on independent held-out subjects. No equations are presented in which a reported prediction or performance quantity is defined in terms of itself, no fitted parameter is relabeled as a prediction, and no load-bearing premise reduces to a self-citation chain. The low-rank prior and CrossCoder are modeling choices whose adequacy is tested by external data rather than by construction; the derivation therefore remains self-contained against the reported benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Only the abstract is available, so specific free parameters, axioms, and invented entities cannot be exhaustively identified from the text. The approach appears to rest on the unstated assumption that the mechanistic simulator plus low-rank cohort prior is adequate for the target tasks.

invented entities (2)
  • CrossCoder no independent evidence
    purpose: Maps connectomes from 20 different anatomical atlases into a shared latent space
    Introduced to enable atlas-independent deployment of the cohort prior.
  • synthetic access no independent evidence
    purpose: Governance-audited alternative to raw data access using the shared mechanistic surrogate
    New term introduced for the governance benefit of the shared artifact.

pith-pipeline@v0.9.1-grok · 5826 in / 1281 out tokens · 58705 ms · 2026-06-30T03:18:55.604534+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

125 extracted references · 11 canonical work pages

  1. [1]

    Bajada, C. J. and Sant, M. M. , title =. Aperture Neuro , volume =. 2025 , doi =

    2025

  2. [2]

    , title =

    Yuste, R. , title =. Nature Protocols , volume =. 2023 , doi =

    2023

  3. [3]

    and Alsaigh, R

    Alshanqiti, A. and Alsaigh, R. and Mehmood, R. and Katib, I. and Liang, X. and Corchado, J. M. and See, S. , title =. Frontiers in Neuroinformatics , volume =. 2024 , doi =

    2024

  4. [4]

    and Blume, W

    Wiebe, S. and Blume, W. T. and Girvin, J. P. and Eliasziw, M. , title =. New England Journal of Medicine , volume =. 2001 , doi =

    2001

  5. [5]

    and Jirsa, V

    Hashemi, M. and Jirsa, V. K. and Sorrentino, P. and Petkoski, S. and Wang, H. and Woodman, M. and Breyton, M. and Athanasiadis, A. and Sip, V. and Ziaeemehr, A. and Fousek, J. and Rabuffo, G. and Triebkorn, P. and Saggio, M. and Depannemaecker, D. , title =. IEEE Reviews in Biomedical Engineering , volume=. 2025 , publisher=

    2025

  6. [6]

    Wang, H. E. and Dollomaja, B. and Triebkorn, P. and others , title =. Nature Computational Science , volume =. 2025 , doi =

    2025

  7. [7]

    Wang, H. E. and Woodman, M. and Triebkorn, P. and Lemarechal, J.-D. and Jha, J. and Dollomaja, B. and Vattikonda, A. N. and Sip, V. and Medina Villalon, S. and Hashemi, M. and Guye, M. and Makhalova, J. and Bartolomei, F. and Jirsa, V. , title =. Science Translational Medicine , volume =. 2023 , doi =

    2023

  8. [8]

    and Brehmer, J

    Cranmer, K. and Brehmer, J. and Louppe, G. , title =. Proceedings of the National Academy of Sciences , volume =. 2020 , doi =

    2020

  9. [9]

    and Boelts, J

    Tejero-Cantero, A. and Boelts, J. and Deistler, M. and Lueckmann, J.-M. and Durkan, C. and Gon. Journal of Open Source Software , volume =. 2020 , doi =

    2020

  10. [10]

    and Lueckmann, J.-M

    Boelts, J. and Lueckmann, J.-M. and Gao, R. and Macke, J. H. , title =. eLife , volume =. 2022 , doi =

    2022

  11. [11]

    Greenberg, D. S. and Nonnenmacher, M. and Macke, J. H. , title =. Proceedings of the 36th International Conference on Machine Learning (ICML 2019) , series =

    2019

  12. [12]

    and Deistler, M

    Gloeckler, M. and Deistler, M. and Weilbach, C. and Wood, F. and Macke, J. H. , title =. Proceedings of the 41st International Conference on Machine Learning (ICML 2024) , series =

    2024

  13. [13]

    and Martin, I

    Patow, G. and Martin, I. and Sanz Perl, Y. and Kringelbach, M. L. and Deco, G. , title =. Nature Reviews Methods Primers , volume =. 2024 , doi =

    2024

  14. [14]

    Bartolomei, F. and Tr. What is the concordance between the seizure onset zone and the irritative zone?. Clinical Neurophysiology , volume =. 2016 , doi =

    2016

  15. [15]

    and Knock, S

    Sanz Leon, P. and Knock, S. A. and Woodman, M. M. and Domide, L. and Mersmann, J. and McIntosh, A. R. and Jirsa, V. , title =. Frontiers in Neuroinformatics , volume =. 2013 , doi =

    2013

  16. [16]

    Woodman, M. M. and Pezard, L. and Domide, L. and Knock, S. A. and Sanz-Leon, P. and Mersmann, J. and McIntosh, A. R. and Jirsa, V. , title =. Frontiers in Neuroinformatics , volume =. 2014 , doi =

    2014

  17. [17]

    and Proix, T

    Jirsa, V. and Proix, T. and Perdikis, D. and Woodman, M. M. and Wang, H. and Gonzalez-Martinez, J. and Bernard, C. and B. The virtual epileptic patient: individualized whole-brain models of epilepsy spread , journal =

  18. [18]

    and Wang, H

    Dollomaja, B. and Wang, H. E. and Guye, M. and Makhalova, J. and Bartolomei, F. and Jirsa, V. K. , title =. PLOS Computational Biology , volume =. 2025 , doi =

    2025

  19. [19]

    and Ziaeemehr, A

    Hashemi, M. and Ziaeemehr, A. and Woodman, M. M. and Fousek, J. and Petkoski, S. and Jirsa, V. K. , title =. Machine Learning: Science and Technology , volume =. 2024 , doi =

    2024

  20. [20]

    and Woodman, M

    Ziaeemehr, A. and Woodman, M. and Domide, L. and Petkoski, S. and Jirsa, V. and Hashemi, M. , title =. eLife , volume =

  21. [21]

    Woodman, M. M. and others , title =. SpringerLink , pages =. 2025 , month = sep, issn =

    2025

  22. [22]

    and Domide, L

    Schirner, M. and Domide, L. and Perdikis, D. and Triebkorn, P. and Stefanovski, L. and Pai, R. and others , title =. NeuroImage , volume =. 2022 , doi =

    2022

  23. [23]

    Biological psychiatry , volume=

    Understanding heterogeneity in clinical cohorts using normative models: beyond case-control studies , author=. Biological psychiatry , volume=. 2016 , publisher=

    2016

  24. [24]

    Kia, S. M. and Huijsdens, H. and Rutherford, S. and de Boer, A. and Dinga, R. and Wolfers, T. and Berthet, P. and Mennes, M. and Andreassen, O. A. and Westlye, L. T. and Beckmann, C. F. and Marquand, A. F. , title =. PLOS ONE , volume =. 2022 , doi =

    2022

  25. [25]

    and Edde, M

    Jodoin, P.-M. and Edde, M. and Girard, G. and Dumais, F. and Theaud, G. and Dumont, M. and Houde, J.-C. and David, Y. and Descoteaux, M. , title =. Scientific Reports , volume =. 2025 , doi =

    2025

  26. [26]

    and Shinohara, R

    Gardner, M. and Shinohara, R. T. , title =. Human Brain Mapping , volume =. 2025 , doi =

    2025

  27. [27]

    and Moore, E

    McMahan, B. and Moore, E. and Ramage, D. and Hampson, S. and y Arcas, B. A. , title =. Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS) , series =

  28. [28]

    and Laton, J

    Denissen, S. and Laton, J. and Grothe, M. and Vaneckova, M. and Uher, T. and Kudrna, M. and Hor. Real-world federated learning for brain imaging scientists , journal =. 2026 , doi =

    2026

  29. [29]

    and others , title =

    Bajwa, S. and others , title =. npj Digital Medicine , volume =. 2025 , doi =

    2025

  30. [30]

    and Minssen, T

    Aboy, M. and Minssen, T. and Vayena, E. , title =. npj Digital Medicine , volume =. 2024 , doi =

    2024

  31. [31]

    Jirsa, V. K. and Stacey, W. C. and Quilichini, P. P. and Ivanov, A. I. and Bernard, C. , title =. Brain , volume =. 2014 , doi =

    2014

  32. [32]

    and Jirsa, V

    Proix, T. and Jirsa, V. K. and Bartolomei, F. and Guye, M. and Bernard, C. , title =. Journal of Neuroscience , volume =. 2014 , doi =

    2014

  33. [33]

    and Bartolomei, F

    Proix, T. and Bartolomei, F. and Guye, M. and Jirsa, V. K. , title =. Brain , volume =. 2017 , doi =

    2017

  34. [34]

    Macroscopic description for networks of spiking neurons , journal =

    Montbri. Macroscopic description for networks of spiking neurons , journal =. 2015 , doi =

    2015

  35. [35]

    NeuroImage , volume=

    Nonlinear responses in fMRI: the Balloon model, Volterra kernels, and other hemodynamics , author=. NeuroImage , volume=. 2000 , publisher=

    2000

  36. [36]

    Journal of Neuroscience , volume=

    Resting-state functional connectivity emerges from structurally and dynamically shaped slow linear fluctuations , author=. Journal of Neuroscience , volume=. 2013 , publisher=

    2013

  37. [37]

    NeuroImage , volume=

    The virtual aging brain: Causal inference supports interhemispheric dedifferentiation in healthy aging , author=. NeuroImage , volume=. 2023 , publisher=

    2023

  38. [38]

    Bishop, C. M. , title =

  39. [39]

    and Pavlakou, T

    Papamakarios, G. and Pavlakou, T. and Murray, I. , title =. Advances in Neural Information Processing Systems (NeurIPS 2017) , year =

    2017

  40. [40]

    and Bekasov, A

    Durkan, C. and Bekasov, A. and Murray, I. and Papamakarios, G. , title =. Advances in Neural Information Processing Systems (NeurIPS 2019) , year =

    2019

  41. [41]

    and Deistler, M

    Gloeckler, M. and Deistler, M. and Macke, J. H. , title =. Proceedings of the 40th International Conference on Machine Learning (ICML 2023) , series =

    2023

  42. [42]

    and Kucharsk

    Mishra, A. and Kucharsk. Unsupervised Continual Learning for Amortized. arXiv preprint arXiv:2602.22884 , year =

    work page arXiv

  43. [43]

    and Habermann, D

    Mishra, A. and Habermann, D. and others , title =. arXiv preprint arXiv:2501.13483 , year =

    work page arXiv

  44. [44]

    CrossCoder: Learning shared structure in brain connectivity , year =

  45. [45]

    Van Essen, D. C. and Smith, S. M. and Barch, D. M. and Behrens, T. E. J. and Yacoub, E. and Ugurbil, K. , title =. NeuroImage , volume =. 2013 , doi =

    2013

  46. [46]

    and Lepage, C

    Amunts, K. and Lepage, C. and Borgeat, L. and Mohlberg, H. and Dickscheid, T. and Rousseau, M.-E. and Bludau, S. and Bazin, P.-L. and Lewis, L. B. and Oros-Peusquens, A.-M. and Shah, N. J. and Lippert, T. and Zilles, K. and Evans, A. C. , title =. Science , volume =. 2013 , doi =

    2013

  47. [47]

    and Popovych, Oleksandr V

    Jung, Kyesam and Eickhoff, Simon B. and Popovych, Oleksandr V. , keywords =. Parcellation-based structural and resting-state functional whole-brain connectomes of 1000BRAINS cohort (v1.1) , publisher =. 2022 , copyright =. doi:10.25493/8XY5-BH7 , url =

    work page doi:10.25493/8xy5-bh7 2022

  48. [48]

    Domhof, Justin W. M. and Jung, Kyesam and Eickhoff, Simon B. and Popovych, Oleksandr V. , keywords =. None (v1.2) , publisher =. 2025 , copyright =. doi:10.25493/3PMM-FPW , url =

    work page doi:10.25493/3pmm-fpw 2025

  49. [49]

    Neuroimage , volume=

    The WU-Minn human connectome project: an overview , author=. Neuroimage , volume=. 2013 , publisher=

    2013

  50. [50]

    Journal of neuroscience methods , volume=

    VEP atlas: An anatomic and functional human brain atlas dedicated to epilepsy patients , author=. Journal of neuroscience methods , volume=. 2021 , publisher=

    2021

  51. [51]

    Proceedings of the National Academy of Sciences , volume=

    Sequential monte carlo without likelihoods , author=. Proceedings of the National Academy of Sciences , volume=. 2007 , publisher=

    2007

  52. [52]

    NeuroImage , volume=

    The Bayesian Virtual Epileptic Patient: A probabilistic framework designed to infer the spatial map of epileptogenicity in a personalized large-scale brain model of epilepsy spread , author=. NeuroImage , volume=. 2020 , publisher=

    2020

  53. [53]

    Neural Networks , volume=

    Amortized Bayesian inference on generative dynamical network models of epilepsy using deep neural density estimators , author=. Neural Networks , volume=. 2023 , publisher=

    2023

  54. [54]

    The Lancet Neurology , volume=

    Personalised virtual brain models in epilepsy , author=. The Lancet Neurology , volume=. 2023 , publisher=

    2023

  55. [55]

    Neural Computation , volume=

    Inference on the macroscopic dynamics of spiking neurons , author=. Neural Computation , volume=. 2024 , publisher=

    2024

  56. [56]

    International Conference on Artificial Neural Networks , pages=

    Amortizing personalization in virtual brain twins , author=. International Conference on Artificial Neural Networks , pages=. 2025 , organization=

    2025

  57. [57]

    Network Neuroscience , volume=

    High-resolution Bayesian Virtual Epileptic Patient using neural field models , author=. Network Neuroscience , volume=. 2026 , publisher=

    2026

  58. [58]

    and Gelman, A

    Carpenter, B. and Gelman, A. and Hoffman, M. and Lee, D. and Goodrich, B. and Betancourt, M. and Brubaker, M. and Guo, J. and Li, P. and Riddell, A. , year =. Stan : A Probabilistic Programming Language , volume =. Journal of Statistical Software , doi =

  59. [59]

    Proceedings of the National Academy of Sciences , volume=

    Adaptive Monte Carlo augmented with normalizing flows , author=. Proceedings of the National Academy of Sciences , volume=. 2022 , publisher=

    2022

  60. [60]

    Machine Learning: Science and Technology , volume=

    Fully Bayesian estimation of virtual brain parameters with self-tuning Hamiltonian Monte Carlo , author=. Machine Learning: Science and Technology , volume=. 2022 , publisher=

    2022

  61. [61]

    Journal of Neuroscience , volume=

    A recurrent network mechanism of time integration in perceptual decisions , author=. Journal of Neuroscience , volume=

  62. [62]

    Advances in neural information processing systems , volume=

    Learning structured output representation using deep conditional generative models , author=. Advances in neural information processing systems , volume=

  63. [63]

    PyMC: A Modern, and Comprehensive Probabilistic Programming Framework in Python

    Oriol Abril-Pla and Virgile Andreani and Colin Carroll and Larry Dong and Christopher J. Fonnesbeck and Maxim Kochurov and Ravin Kumar and Junpeng Lao and Christian C. Luhmann and Osvaldo A. Martin and Michael Osthege and Ricardo Vieira and Thomas Wiecki and Robert Zinkov , journal =. doi:10.7717/peerj-cs.1516 , year =

    work page doi:10.7717/peerj-cs.1516

  64. [64]

    2015 , publisher=

    TensorFlow: Large-scale machine learning on heterogeneous systems , author=. 2015 , publisher=

    2015

  65. [65]

    Advances in neural information processing systems , volume=

    Pytorch: An imperative style, high-performance deep learning library , author=. Advances in neural information processing systems , volume=

  66. [66]

    and Jankowiak, Martin and Obermeyer, Fritz and Pradhan, Neeraj and Karaletsos, Theofanis and Singh, Rohit and Szerlip, Paul and Horsfall, Paul and Goodman, Noah D

    Bingham, Eli and Chen, Jonathan P. and Jankowiak, Martin and Obermeyer, Fritz and Pradhan, Neeraj and Karaletsos, Theofanis and Singh, Rohit and Szerlip, Paul and Horsfall, Paul and Goodman, Noah D. , title =. Journal of Machine Learning Research , year =

  67. [67]

    James Bradbury and Roy Frostig and Peter Hawkins and Matthew James Johnson and Chris Leary and Dougal Maclaurin and George Necula and Adam Paszke and Jake Vander

  68. [68]

    2026 , eprint=

    A foundation model of vision, audition, and language for in-silico neuroscience , author=. 2026 , eprint=

    2026

  69. [69]

    2024 , eprint=

    Decoding individual words from non-invasive brain recordings across 723 participants , author=. 2024 , eprint=

    2024

  70. [70]

    2025 , publisher=

    Enshittification: Why everything suddenly got worse and what to do about it , author=. 2025 , publisher=

    2025

  71. [71]

    Physics in Medicine & Biology , volume=

    Basic mathematical and electromagnetic concepts of the biomagnetic inverse problem , author=. Physics in Medicine & Biology , volume=

  72. [72]

    and Rohlfing, Torsten and Pfefferbaum, Adolf , title =

    Sullivan, Edith V. and Rohlfing, Torsten and Pfefferbaum, Adolf , title =. Developmental Neuropsychology , volume =. 2010 , doi =

    2010

  73. [73]

    and Byrge, Lisa and He, Ye and Go

    Betzel, Richard F. and Byrge, Lisa and He, Ye and Go. Changes in structural and functional connectivity among resting-state networks across the human lifespan , journal =. 2014 , doi =

    2014

  74. [74]

    Psychology and Aging , volume =

    Cabeza, Roberto , title =. Psychology and Aging , volume =. 2002 , doi =

    2002

  75. [75]

    , title =

    Constantine, Paul G. , title =. 2015 , doi =

    2015

  76. [76]

    and Leskovec, Jure and Topol, Eric J

    Moor, Michael and Banerjee, Oishi and Abad, Zahra Shakeri Hossein and Krumholz, Harlan M. and Leskovec, Jure and Topol, Eric J. and Rajpurkar, Pranav , title =. Nature , volume =. 2023 , doi =

    2023

  77. [77]

    sbi: A toolkit for simulation-based inference , journal =

    Tejero-Cantero,. sbi: A toolkit for simulation-based inference , journal =. 2020 , doi =

    2020

  78. [78]

    Studying variability in human brain aging in a population-based

    Caspers, Svenja and Moebus, Susanne and Lux, Silke and Pundt, Noreen and Sch. Studying variability in human brain aging in a population-based. Frontiers in Aging Neuroscience , volume =. 2014 , doi =

    2014

  79. [79]

    Human Brain Mapping , volume =

    Stumme, Johanna and Krämer, Camilla and Miller, Tatiana and Schreiber, Jan and Caspers, Svenja and Jockwitz, Christiane , title =. Human Brain Mapping , volume =. doi:https://doi.org/10.1002/hbm.26030 , year =

    work page doi:10.1002/hbm.26030

  80. [80]

    and Laumann, Timothy O

    Schaefer, Alexander and Kong, Ru and Gordon, Evan M. and Laumann, Timothy O. and Zuo, Xi-Nian and Holmes, Avram J. and Eickhoff, Simon B. and Yeo, B. T. Thomas , title =. Cerebral Cortex , volume =. 2018 , doi =

    2018

Showing first 80 references.