pith. sign in

arxiv: 2606.07488 · v1 · pith:WZQWVGRSnew · submitted 2026-06-05 · 💻 cs.LG

CoMetaPNS: Continually Meta-learning Personalized Neural Surrogates for Cardiac Electrophysiology Simulations

Ryan Missel , Xiajun Jiang , Linwei Wang This is my paper

Pith reviewed 2026-06-27 22:21 UTC · model grok-4.3

classification 💻 cs.LG
keywords continual meta-learningpersonalized neural surrogatescardiac electrophysiologyBayesian Gaussian mixture modelcatastrophic forgettingfew-shot personalizationsimulation adaptation
0
0 comments X

The pith

A continual Bayesian Gaussian Mixture Model over a memory buffer lets meta-learned neural surrogates personalize cardiac simulations from sequential unlabeled data without catastrophic forgetting.

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

Personalized virtual heart simulations need neural surrogates that adapt to individual patients from limited context data. Existing meta-learning methods for this task assume a fixed training set with known task labels and must retrain on all prior data when new information arrives, which is often impossible in clinical settings. The paper proposes a framework that maintains a memory buffer and applies a continual Bayesian Gaussian Mixture Model to automatically cluster incoming data, infer source identities, and decide whether each new stream is known or novel. This supports ongoing meta-learning for set-conditioned surrogates while avoiding full retraining. If the approach holds, it would allow cardiac electrophysiology models to keep incorporating fresh patient measurements over time at feasible cost.

Core claim

By leveraging a continual Bayesian Gaussian Mixture Model over a memory buffer, the framework can infer the identifiers and relationships of data over time, enabling effective continual meta-learning for personalized neural surrogates in cardiac electrophysiology simulations that integrate new information without requiring explicit task identifiers or full prior data access.

What carries the argument

Continual Bayesian Gaussian Mixture Model over a memory buffer that clusters sequential data to infer task identifiers and relationships for meta-learning.

If this is right

  • Superior simulation forecasting accuracy on synthetic cardiac data relative to baselines that require full retraining.
  • Improved computational scalability by avoiding repeated training on all accumulated data.
  • Resilience to catastrophic forgetting when new unlabeled data sources arrive sequentially.
  • Effective operation on unlabeled sequential data without needing explicit task boundaries.

Where Pith is reading between the lines

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

  • The same memory-buffered clustering step could support continual meta-learning in other simulation domains that receive streaming personalized data.
  • Automatic source identification may lessen dependence on hand-labeled task boundaries across continual learning applications.
  • Empirical checks on real rather than synthetic cardiac recordings would reveal whether clustering accuracy survives natural physiological variability.

Load-bearing premise

The approach assumes that a continual Bayesian Gaussian Mixture Model can reliably cluster and identify data sources over time from a memory buffer without explicit task identifiers or full prior data access.

What would settle it

A test showing that the Bayesian mixture model assigns a large fraction of new data to incorrect clusters or fails to flag novel sources, producing degraded personalization accuracy on both recent and earlier cardiac simulation tasks.

Figures

Figures reproduced from arXiv: 2606.07488 by Linwei Wang, Ryan Missel, Xiajun Jiang.

Figure 1
Figure 1. Figure 1: Overview of CoMetaPNS, showing A) the framework of few-shot generative modeling via Bayesian meta-learning that B) continually aggregate a heterogeneous data stream of cardiac electrophysiology dynamics with a sample reservoir that identifies sub￾jects via Gaussian mixture models. any newly-presented data corresponds to a previously-known or novel subject. 3. We assume the presence of local stationarity wh… view at source ↗
Figure 2
Figure 2. Figure 2: Spatial correlation coefficient (SCC) performance comparison on the synthetic [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: CoMetaPNS with Task-Aware Meta-Learning (left) overcomes catastrophic for [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual examples of reconstructed electrical activity from the proposed continual [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Dice coefficient (DC) performance comparison on the synthetic data for the [PITH_FULL_IMAGE:figures/full_fig_p025_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visual examples of reconstructed electrical activity on both meta-learners con [PITH_FULL_IMAGE:figures/full_fig_p027_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance metrics evaluating the effectiveness of using a pre-trained meta [PITH_FULL_IMAGE:figures/full_fig_p029_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: GMM log-likelihoods of all tasks’ samples over the sequence, becoming known [PITH_FULL_IMAGE:figures/full_fig_p030_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: 2D t-SNE visualization of the task-relational reservoir over the task sequence. [PITH_FULL_IMAGE:figures/full_fig_p031_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Average cluster purity of each task’s samples over the task sequence. A purity of [PITH_FULL_IMAGE:figures/full_fig_p032_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Task-Relational vs. Task-Aware strategies on imbalanced tasks where the first [PITH_FULL_IMAGE:figures/full_fig_p033_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Visual examples of reconstructed electrical activity from the proposed [PITH_FULL_IMAGE:figures/full_fig_p035_12.png] view at source ↗
read the original abstract

Personalized virtual heart simulations face challenges in model personalization and computational cost. While neural surrogates offer state-of-the-art solutions, they typically address either efficient personalization or training generalizable models. Recent work reframes this by learning the process of personalizing a surrogate using limited subject-specific context data, through few-shot generative modeling with set-conditioned surrogates and meta-learned amortized inference. These methods, however, assume a static and diverse training distribution with known task identifiers. When new data becomes available, they require costly retraining with all prior data to avoid catastrophic forgetting - a phenomena where the model forgets earlier tasks when trained on new ones. This is a major limitation in clinical settings where often unlabeled data arrives sequentially and full retraining is infeasible. This paper presents a new continual meta-learning framework to achieve personalized neural surrogates able to not only continually integrate information but also identify whether incoming data stems from a known or unknown dynamics source. By leveraging a continual Bayesian Gaussian Mixture Model over a memory buffer, our framework can infer the identifiers and relationships of data over time - required for effective meta-learning. Empirical results on synthetic cardiac data demonstrate superior simulation forecasting, computational scalability, and resilience to catastrophic forgetting compared to existing baselines.

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 presents CoMetaPNS, a continual meta-learning framework for personalized neural surrogates in cardiac electrophysiology. It addresses limitations of prior few-shot generative modeling approaches by handling sequential unlabeled data without task identifiers or full retraining, using a continual Bayesian Gaussian Mixture Model over a memory buffer to infer data identifiers and relationships. This enables ongoing integration while avoiding catastrophic forgetting. The framework is evaluated on synthetic cardiac data, where it reportedly outperforms baselines in simulation forecasting accuracy, computational scalability, and forgetting resistance.

Significance. If the empirical claims hold after addressing validation gaps, the work would meaningfully extend meta-learning techniques to continual settings in a high-stakes domain, directly tackling the retraining infeasibility noted in clinical personalization workflows. The explicit handling of unknown dynamics sources via the memory-buffered GMM is a targeted contribution, and the focus on synthetic data allows controlled testing of the continual aspect. However, the absence of real patient data or detailed controls limits translational significance until further validation.

major comments (2)
  1. [Abstract] Abstract: the central claim that the continual Bayesian GMM over a memory buffer 'can infer the identifiers and relationships of data over time - required for effective meta-learning' is load-bearing, yet no clustering accuracy metrics, ablation on mixture components, or validation against ground-truth dynamics sources are reported; without this, it is impossible to confirm that gains in forecasting and forgetting resistance arise from successful identification rather than other components.
  2. [Abstract] Abstract (empirical results paragraph): the superiority claims for 'simulation forecasting, computational scalability, and resilience to catastrophic forgetting' rest on synthetic data only, but the abstract supplies no baseline methods, quantitative metrics (e.g., MSE, forgetting measure), number of sequential tasks, or controls for buffer size and mixture components; this prevents assessment of whether the data supports the framework's advantages.
minor comments (1)
  1. [Abstract] The abstract would be clearer if it briefly defined the memory buffer size and number of mixture components used in the reported experiments.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight opportunities to strengthen the abstract's clarity and supporting evidence. We respond to each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the continual Bayesian GMM over a memory buffer 'can infer the identifiers and relationships of data over time - required for effective meta-learning' is load-bearing, yet no clustering accuracy metrics, ablation on mixture components, or validation against ground-truth dynamics sources are reported; without this, it is impossible to confirm that gains in forecasting and forgetting resistance arise from successful identification rather than other components.

    Authors: We acknowledge that direct metrics on the clustering performance of the Bayesian GMM (such as accuracy against ground-truth task identifiers in the synthetic setup) are not reported in the abstract and would provide stronger evidence for the mechanism. The manuscript's primary evaluations focus on end-to-end performance in forecasting accuracy and forgetting resistance as indirect validation of the identification process. To address this concern, we will add clustering accuracy results, an ablation on the number of mixture components, and explicit validation against known dynamics sources in a revised version of the paper, including updates to the abstract. revision: yes

  2. Referee: [Abstract] Abstract (empirical results paragraph): the superiority claims for 'simulation forecasting, computational scalability, and resilience to catastrophic forgetting' rest on synthetic data only, but the abstract supplies no baseline methods, quantitative metrics (e.g., MSE, forgetting measure), number of sequential tasks, or controls for buffer size and mixture components; this prevents assessment of whether the data supports the framework's advantages.

    Authors: We agree that the abstract is concise and omits specific experimental details such as exact baseline names, quantitative values (e.g., MSE or forgetting measures), the number of sequential tasks, and controls for buffer size or mixture components. These details are provided in the experimental sections of the full manuscript, where we compare against relevant baselines on synthetic cardiac data with known ground-truth dynamics. We will revise the abstract to incorporate key quantitative metrics, the number of tasks, and mention of controls to better support the claims while maintaining conciseness. The use of synthetic data is intentional to enable controlled evaluation of continual adaptation with verifiable task relationships. revision: yes

Circularity Check

0 steps flagged

No load-bearing circularity; continual GMM inference is an independent component

full rationale

The paper introduces a continual Bayesian Gaussian Mixture Model over a memory buffer as the mechanism to infer identifiers and relationships from sequential unlabeled data, enabling the meta-learning framework. This is presented as a distinct architectural addition rather than a quantity derived by construction from prior fits, self-citations, or ansatzes. The abstract references recent work on set-conditioned surrogates and meta-learned inference only to contrast assumptions (static distribution, known task IDs), without making the new continual results dependent on those citations. No equations or claims reduce predictions to inputs by definition, and empirical results on synthetic data are offered as external validation. This yields at most a minor self-citation that is not load-bearing for the central claim.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

The central claim depends on a new continual inference module whose effectiveness is not independently verified outside the synthetic experiments described.

free parameters (2)
  • memory buffer size
    Hyperparameter controlling retention of past data for the BGMM; value not specified in abstract.
  • number of mixture components
    Hyperparameter for modeling distinct dynamics sources; value not specified in abstract.
axioms (1)
  • domain assumption Sequential data arrives from a mixture of unknown dynamics sources that can be modeled by an incrementally updated Bayesian Gaussian Mixture Model.
    Invoked to enable inference of task identifiers without known labels or full retraining.
invented entities (1)
  • Continual Bayesian Gaussian Mixture Model over memory buffer no independent evidence
    purpose: To infer identifiers and relationships among data sources for ongoing meta-learning.
    New component introduced to address the continual aspect; no independent evidence provided beyond the synthetic results.

pith-pipeline@v0.9.1-grok · 5752 in / 1405 out tokens · 33165 ms · 2026-06-27T22:21:01.129946+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

186 extracted references · 24 canonical work pages · 16 internal anchors

  1. [1]

    M. F. Kasim and D. Watson-Parris and L. Deaconu and S. Oliver and H. Hatfield and D. H. Froula and G. Gregori and M. Jarvis and S. Khatiwala and J. Korenaga and J. Topp-Mugglestone and E. Viezzer and S. M. Vinko , Journal =. Building high accuracy emulators for scientific simulations with deep neural architecture search , pages =

  2. [2]

    2020 , publisher=

    Electromechanical Wave Imaging in the clinic: localization of atrial and ventricular arrhythmias and quantification of cardiac resynchronization therapy response , author=. 2020 , publisher=

    2020

  3. [3]

    Nature biomedical engineering , volume=

    Personalized virtual-heart technology for guiding the ablation of infarct-related ventricular tachycardia , author=. Nature biomedical engineering , volume=. 2018 , publisher=

    2018

  4. [4]

    Nature communications , volume=

    Arrhythmia risk stratification of patients after myocardial infarction using personalized heart models , author=. Nature communications , volume=. 2016 , publisher=

    2016

  5. [5]

    Chaos, Solitons & Fractals , volume=

    A simple two-variable model of cardiac excitation , author=. Chaos, Solitons & Fractals , volume=. 1996 , publisher=

    1996

  6. [6]

    Progress in biophysics and molecular biology , volume=

    Electromechanical model of excitable tissue to study reentrant cardiac arrhythmias , author=. Progress in biophysics and molecular biology , volume=. 2004 , publisher=

    2004

  7. [7]

    Philosophical Transactions of the Royal Society A , volume=

    Creation and application of virtual patient cohorts of heart models , author=. Philosophical Transactions of the Royal Society A , volume=. 2020 , publisher=

    2020

  8. [8]

    Computers in biology and medicine , volume=

    Rethinking multiscale cardiac electrophysiology with machine learning and predictive modelling , author=. Computers in biology and medicine , volume=. 2019 , publisher=

    2019

  9. [9]

    International Conference on Functional Imaging and Modeling of the Heart , pages=

    Ep-net 2.0: out-of-domain generalisation for deep learning models of cardiac electrophysiology , author=. International Conference on Functional Imaging and Modeling of the Heart , pages=. 2021 , organization=

    2021

  10. [10]

    Deep Learning for Model Correction in Cardiac Electrophysiological Imaging , author=

  11. [11]

    Frontiers in physiology , volume=

    An Implementation of Patient-Specific Biventricular Mechanics Simulations With a Deep Learning and Computational Pipeline , author=. Frontiers in physiology , volume=. 2021 , publisher=

    2021

  12. [12]

    Frontiers in physiology , volume=

    POD-enhanced deep learning-based reduced order models for the real-time simulation of cardiac electrophysiology in the left atrium , author=. Frontiers in physiology , volume=. 2021 , publisher=

    2021

  13. [13]

    Medical image analysis , volume=

    Quantifying the uncertainty in model parameters using Gaussian process-based Markov chain Monte Carlo in cardiac electrophysiology , author=. Medical image analysis , volume=. 2018 , publisher=

    2018

  14. [14]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Strain-based regional nonlinear cardiac material properties estimation from medical images , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2012 , organization=

    2012

  15. [15]

    IEEE transactions on biomedical engineering , volume=

    Personalization of cardiac motion and contractility from images using variational data assimilation , author=. IEEE transactions on biomedical engineering , volume=. 2011 , publisher=

    2011

  16. [16]

    Journal of the mechanical behavior of biomedical materials , volume=

    Velocity-based cardiac contractility personalization from images using derivative-free optimization , author=. Journal of the mechanical behavior of biomedical materials , volume=. 2015 , publisher=

    2015

  17. [17]

    Medical image analysis , volume=

    Patient-specific electromechanical models of the heart for the prediction of pacing acute effects in CRT: a preliminary clinical validation , author=. Medical image analysis , volume=. 2012 , publisher=

    2012

  18. [18]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Fast data-driven calibration of a cardiac electrophysiology model from images and ECG , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2013 , organization=

    2013

  19. [19]

    Medical image analysis , volume=

    Embedding high-dimensional bayesian optimization via generative modeling: parameter personalization of cardiac electrophysiological models , author=. Medical image analysis , volume=. 2020 , publisher=

    2020

  20. [20]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Spatially-adaptive multi-scale optimization for local parameter estimation: application in cardiac electrophysiological models , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2016 , organization=

    2016

  21. [21]

    IEEE transactions on medical imaging , volume=

    Spatially adaptive multi-scale optimization for local parameter estimation in cardiac electrophysiology , author=. IEEE transactions on medical imaging , volume=. 2017 , publisher=

    2017

  22. [22]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    High-dimensional bayesian optimization of personalized cardiac model parameters via an embedded generative model , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2018 , organization=

    2018

  23. [23]

    Frontiers in Physiology , volume=

    Bayesian Calibration of Electrophysiology Models Using Restitution Curve Emulators , author=. Frontiers in Physiology , volume=. 2021 , publisher=

    2021

  24. [24]

    Medical image analysis , volume=

    A self-taught artificial agent for multi-physics computational model personalization , author=. Medical image analysis , volume=. 2016 , publisher=

    2016

  25. [25]

    Artificial Intelligence for Computational Modeling of the Heart , pages=

    Machine learning methods for robust parameter estimation , author=. Artificial Intelligence for Computational Modeling of the Heart , pages=. 2020 , publisher=

    2020

  26. [26]

    IEEE Transactions on Biomedical Engineering , volume=

    Noninvasive personalization of a cardiac electrophysiology model from body surface potential mapping , author=. IEEE Transactions on Biomedical Engineering , volume=. 2016 , publisher=

    2016

  27. [27]

    arXiv preprint arXiv:2004.05439 , year=

    Meta-learning in neural networks: A survey , author=. arXiv preprint arXiv:2004.05439 , year=

    work page arXiv 2004

  28. [28]

    International conference on machine learning , pages=

    Model-agnostic meta-learning for fast adaptation of deep networks , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  29. [29]

    International conference on learning representations , year=

    Optimization as a model for few-shot learning , author=. International conference on learning representations , year=

  30. [30]

    Advances in neural information processing systems , volume=

    Prototypical networks for few-shot learning , author=. Advances in neural information processing systems , volume=

  31. [31]

    arXiv preprint arXiv:1805.09921 , year=

    Meta-learning probabilistic inference for prediction , author=. arXiv preprint arXiv:1805.09921 , year=

    work page arXiv

  32. [32]

    The Variational Homoencoder: Learning to learn high capacity generative models from few examples

    The variational homoencoder: Learning to learn high capacity generative models from few examples , author=. arXiv preprint arXiv:1807.08919 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  33. [33]

    Neural Processes

    Neural processes , author=. arXiv preprint arXiv:1807.01622 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  34. [34]

    Auto-Encoding Variational Bayes

    Auto-encoding variational bayes , author=. arXiv preprint arXiv:1312.6114 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  35. [35]

    2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro , pages=

    Tetrahedral mesh generation from volumetric binary and grayscale images , author=. 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro , pages=. 2009 , organization=

    2009

  36. [36]

    Triangulated Surface Mesh Simplification , YEAR =

    Fernando Cacciola , BOOKTITLE =. Triangulated Surface Mesh Simplification , YEAR =

  37. [37]

    Proceedings Visualization'98 (Cat

    Fast and memory efficient polygonal simplification , author=. Proceedings Visualization'98 (Cat. No. 98CB36276) , pages=. 1998 , organization=

    1998

  38. [38]

    The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    SplineCNN: Fast geometric deep learning with continuous B-spline kernels , author=. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  39. [39]

    Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling

    Empirical evaluation of gated recurrent neural networks on sequence modeling , author=. arXiv preprint arXiv:1412.3555 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  40. [40]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Label-Free Physics-Informed Image Sequence Reconstruction with Disentangled Spatial-Temporal Modeling , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2021 , organization=

    2021

  41. [41]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Learning geometry-dependent and physics-based inverse image reconstruction , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2020 , organization=

    2020

  42. [42]

    IEEE transactions on systems, man, and cybernetics , volume=

    A threshold selection method from gray-level histograms , author=. IEEE transactions on systems, man, and cybernetics , volume=. 1979 , publisher=

    1979

  43. [43]

    Adam: A Method for Stochastic Optimization

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    Bergquist and W.W

    J.A. Bergquist and W.W. Good and B. Zenger and J.D. Tate and L.C. Rupp and R.S. MacLeod. The electrocardiographic forward problem: A benchmark study. Comput Biol Med. 2021

    2021

  45. [45]

    Multiplicative interactions and where to find them , author =

  46. [46]

    Advances in Neural Information Processing Systems , publisher =

    Meta-Learning Dynamics Forecasting Using Task Inference , author =. Advances in Neural Information Processing Systems , publisher =

  47. [47]

    International Conference on Machine Learning , pages =

    Generalizing to new physical systems via context-informed dynamics model , author =. International Conference on Machine Learning , pages =

  48. [48]

    Sequential Latent Variable Models for Few-Shot High-Dimensional Time-Series Forecasting , author =

  49. [49]

    N-BEATS: Neural basis expansion analysis for interpretable time series forecasting , author =

  50. [50]

    Advances in Neural Information Processing Systems , volume = 34, pages =

    LEADS: Learning dynamical systems that generalize across environments , author =. Advances in Neural Information Processing Systems , volume = 34, pages =

  51. [51]

    Elements of physical biology , author =

  52. [52]

    International conference on machine learning , pages =

    Model-agnostic meta-learning for fast adaptation of deep networks , author =. International conference on machine learning , pages =

  53. [53]

    Advances in Neural Information Processing Systems , volume = 33, pages =

    On the modularity of hypernetworks , author =. Advances in Neural Information Processing Systems , volume = 33, pages =

  54. [54]

    A disentangled recognition and nonlinear dynamics model for unsupervised learning , author =

  55. [55]

    Improved training of wasserstein gans , author =

  56. [56]

    ODE2VAE: Deep generative second order ODEs with Bayesian neural networks , author =

  57. [57]

    Advances in Neural Information Processing Systems , publisher =

    A Recurrent Latent Variable Model for Sequential Data , author =. Advances in Neural Information Processing Systems , publisher =

  58. [58]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume = 31, number = 11, doi =

    Structured Inference Networks for Nonlinear State Space Models , author =. Proceedings of the AAAI Conference on Artificial Intelligence , volume = 31, number = 11, doi =

  59. [59]

    Deep Kalman Filters , author =

  60. [60]

    doi:10.48550/arXiv.2111.05458 , note =

    Which priors matter? Benchmarking models for learning latent dynamics , author =. doi:10.48550/arXiv.2111.05458 , note =

    work page doi:10.48550/arxiv.2111.05458

  61. [61]

    Proceedings of the 20th International Conference on Artificial Intelligence and Statistics , publisher =

    Linderman, Scott and Johnson, Matthew and Miller, Andrew and Adams, Ryan and Blei, David and Paninski, Liam , year = 2017, month =. Proceedings of the 20th International Conference on Artificial Intelligence and Statistics , publisher =

    2017

  62. [62]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume = 35, number = 8, pages =

    Deep switching auto-regressive factorization: Application to time series forecasting , author =. Proceedings of the AAAI Conference on Artificial Intelligence , volume = 35, number = 8, pages =

  63. [63]

    International conference on machine learning , pages =

    Switching linear dynamics for variational bayes filtering , author =. International conference on machine learning , pages =

  64. [64]

    Sequential neural processes , author =

  65. [65]

    Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting , author =

  66. [66]

    International Conference on Machine Learning , pages =

    Collapsed amortized variational inference for switching nonlinear dynamical systems , author =. International Conference on Machine Learning , pages =

  67. [67]

    Latent Matters: Learning Deep-State-Space Models , author =

  68. [68]

    Learning Interpretable Deep State Space Model for Probabilistic Time Series Forecasting , author =

  69. [69]

    DEEP VARIATIONAL BAYES FILTERS: UNSUPERVISED LEARNING OF STATE SPACE MODELS FROM RAW DATA , author =

  70. [70]

    Deep State Space Models for Time Series Forecasting , author =

  71. [71]

    Filtering variational objectives , author =

  72. [72]

    Chen, Ricky T. Q. , title=. 2018 , url=

    2018

  73. [73]

    Advances in neural information processing systems , volume=

    Neural ordinary differential equations , author=. Advances in neural information processing systems , volume=

  74. [74]

    Searching for Activation Functions

    Searching for activation functions , author=. arXiv preprint arXiv:1710.05941 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  75. [75]

    arXiv preprint arXiv:2008.12595 , year=

    Dynamical variational autoencoders: A comprehensive review , author=. arXiv preprint arXiv:2008.12595 , year=

    work page arXiv 2008

  76. [76]

    1925 , publisher=

    Elements of physical biology , author=. 1925 , publisher=

    1925

  77. [77]

    Three scenarios for continual learning

    Three scenarios for continual learning , author=. arXiv preprint arXiv:1904.07734 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1904

  78. [78]

    arXiv preprint arXiv:2205.00865 , year=

    WeatherBench Probability: A benchmark dataset for probabilistic medium-range weather forecasting along with deep learning baseline models , author=. arXiv preprint arXiv:2205.00865 , year=

    work page arXiv

  79. [79]

    A Simple Neural Attentive Meta-Learner

    A simple neural attentive meta-learner , author=. arXiv preprint arXiv:1707.03141 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  80. [80]

    Learning to Learn without Forgetting by Maximizing Transfer and Minimizing Interference

    Learning to learn without forgetting by maximizing transfer and minimizing interference , author=. arXiv preprint arXiv:1810.11910 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

Showing first 80 references.