Pith. sign in

REVIEW 3 major objections 6 minor 48 references

Safe Active Learning for Gaussian Differential Equations

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A new algorithm, SAL GPODE, sequentially chooses initial states whose trajectories teach the most about an unknown differential-equation right-hand side while keeping the system inside a safe region.

desk verdict SAL GPODE is a sensible first step for safe active learning in GP ODEs, but the safety guarantee is only as good as the discrete checkpoints and the posterior calibration, and the evaluation needs a stronger baseline. read the letter →

arxiv 2412.09053 v1 pith:ZLXUT6I2 submitted 2024-12-12 cs.LG stat.ML

Leon Glass , Katharina Ensinger , Christoph Zimmer This is my paper · ORCID
classification cs.LGstat.ML
keywords GaussianprocessdifferentialequationssafeactivelearningacquisitionfunctionmutualinformationsafetyconstraintdynamicsmodelingVanderPoloscillatorLotka-Volterramodel
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

The paper proposes an active-learning loop for Gaussian-process differential equations (GPODE), models where the right-hand side of an ODE is an unknown function drawn from a Gaussian process. The algorithm, SAL GPODE, sequentially chooses the next initial state to measure by maximizing the mutual information between the resulting trajectory and the GP dynamics, subject to a probabilistic constraint that the trajectory stays within a safe state region. The authors' claim is that this safe, information-driven data collection builds a good GPODE model much faster than measuring randomly chosen initial states. They demonstrate the speedup on the Van der Pol oscillator and the Lotka-Volterra model, reporting lower validation negative log-likelihood after the same number of measurements and faster identification of safe versus unsafe initial states. The practical stake is that data collection on physical systems is expensive and can be safety-critical, so choosing where to measure matters as much as how many measurements are taken.

What carries the argument

The load-bearing object is the constrained optimization problem $\vartheta^* = \arg\max_{\vartheta\in\Theta} \alpha(y_{1:N}(\vartheta); g)$ subject to $\xi(\vartheta) \geq \delta$, where $\vartheta$ is a candidate initial state. The acquisition function $\alpha$ is the mutual information $I(y_{1:N}(\vartheta), g)$ between the trajectory measurement and the unknown GP dynamics $g$, approximated by drawing $K$ dynamics functions from the GP posterior via decoupled sampling, integrating each to get trajectories, and estimating the entropy of the resulting measurement distribution. The safety function $\xi(\vartheta)$ is the probability that the sampled trajectories satisfy $x_{\min} \leq x(t_i) \leq x_{\max}$ at all measurement times $t_i$, estimated from the same decoupled samples. The GPODE posterior itself comes from a variational sparse GP with inducing points, trained by maximizing an evidence lower bound, and the dynamics are integrated with an adaptive-step ODE solver.

What would settle it

On a simulator with known dynamics, run a trajectory that dips below $x_{\min}$ between two adjacent measurement times and check whether the safety constraint $\xi(\vartheta) \geq \delta$ still accepts that initial state; if it does, the discrete-time check missed a real violation. Separately, compare the predicted safety probability with the empirical violation frequency over many repeated runs to test whether the GP posterior is calibrated enough for the guarantee to hold.

Watch

Extended reading notes

Core claim

The central claim is that a GPODE model can be trained from significantly fewer measurements if each new measurement is chosen by solving a constrained optimization: maximize the mutual information between a candidate trajectory and the unknown GP dynamics, subject to the probability that the trajectory stays inside a predefined safe region. The paper introduces an acquisition function based on Monte Carlo entropy estimation from decoupled GP posterior samples, and a safety function that estimates the probability of staying within state bounds at the measurement times. On the Van der Pol oscillator and the Lotka-Volterra model, the resulting SAL GPODE algorithm reduces validation negative log-likelihood faster than a random, non-active data collection baseline, and it identifies safe versus unsafe initial states with higher F1 score after the same number of measurements.

Load-bearing premise

The safety check evaluates trajectories only at the chosen measurement times $t_i$ and takes the GP's probability estimates at face value, so the guarantee can fail if a violation happens between two checkpoints or if the model is miscalibrated.

Editorial extensions

If this is right

  • With SAL GPODE, a usable Van der Pol model is learned within about two to three chosen measurements, whereas the random baseline needs substantially more data to reach comparable validation negative log-likelihood.
  • The same pattern holds on Lotka-Volterra: validation error drops faster and the F1 score for identifying safe initial states rises faster than with non-active collection.
  • The acquisition and safety estimates are both sampling-based, so the algorithm needs only the ability to draw posterior dynamics trajectories and integrate them; it is not tied to a specific kernel or integrator.
  • Because safety is a constraint in the optimization, the method separates 'where is it informative to measure' from 'is it safe to measure there,' making the safety threshold $\delta$ a user-tunable trade-off between exploration and risk.

Reading between the lines

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

  • The discrete-time safety check means the guarantee is only as strong as the measurement grid; a trajectory that dips out of the safe box between two checkpoints would not be counted as unsafe. A natural extension is a Lipschitz- or solver-step-aware bound that covers all intermediate times.
  • If the GP posterior is miscalibrated, the safety probability $\xi(\vartheta)$ will be overconfident; the paper's experiments use a correct simulator, so the method's safety behavior on a real miscalibrated system remains an open test.
  • The mutual-information acquisition could be replaced or augmented by the covariance-based variant $\alpha_2$ that the paper defines; the paper notes the two differ exactly when trajectory distributions are non-Gaussian, which is typical for nonlinear ODEs, so comparing them on a strongly nonlinear benchmark would be a direct next experiment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes SAL GPODE, an active learning method for Gaussian process ordinary differential equation (GPODE) models. The method sequentially chooses new initial states whose resulting time-series measurements are maximally informative about the unknown GP dynamics, subject to a probabilistic safety constraint on the state trajectory. The acquisition function is based on mutual information between the candidate trajectory and the GP, approximated by Monte Carlo sampling of decoupled GP posterior trajectories; the safety function is the empirical probability that sampled trajectories stay within a state-space box. Experiments on Van der Pol and Lotka-Volterra systems compare the proposed method against random data collection, reporting faster reduction of validation negative log-likelihood and higher F1 scores for safe-set identification.

Significance. If the claims hold, the paper addresses a real gap in GPODE/hybrid modeling: designing informative and safe data-collection experiments for dynamical systems. The mutual-information derivation is sound: conditioned on the GP, the observation entropy is independent of the initial value, so maximizing the marginal entropy is equivalent to maximizing the mutual information. The use of decoupled sampling to propagate uncertainty through the ODE is appropriate and builds on a relevant body of work. However, the safety guarantee is currently only as strong as the discretization and the calibration of the approximate GP posterior, and the empirical evaluation is narrow, comparing only against a random baseline. The core idea is promising and likely useful, but the safety claim and the empirical support need strengthening before the paper can be accepted.

major comments (3)
  1. [Section 5.2, Eq. (10)] The safety constraint is evaluated only at the discrete measurement times t_i. The text acknowledges a discretization but gives no bound on the probability of a violation between these times. Because the title and abstract claim safe data collection, this gap is load-bearing: an initial state can pass the check and still drive the system outside [x_min, x_max] during integration. The authors should either provide a formal continuous-time safety guarantee (e.g., reachability or Lipschitz-based bounds) or explicitly restrict the claim to sampled time points and justify that restriction for the considered benchmarks.
  2. [Section 5.2] The safety probability xi(x) is estimated by counting decoupled GP posterior samples that stay in the box. If the variational GP approximation is overconfident or miscalibrated, the true violation probability can exceed 1 - delta even at the checkpoints. No calibration procedure or conservative correction is provided. Since safety is a central contribution, the authors should quantify calibration error on held-out trajectories or add a safeguard (e.g., uncertainty inflation or a safety margin) and report its effect.
  3. [Section 6] The empirical evidence consists of comparisons against a single random baseline, with no specification of the optimizer used for Eq. (11), the Monte Carlo sample count K, the number of inducing inputs L and random features S, the safety threshold delta, or the number of measurements M. Without these details, the claimed superiority is not reproducible, and the comparison does not establish competitiveness against other active learning strategies. The authors should provide these settings (and ideally code) and add at least one non-safe active learning baseline or a more extensive ablation to support the central claim.
minor comments (6)
  1. [Eq. (12b)] The Van der Pol dynamics are written as mu(1 - x1)^2 x2 - x1, but the standard form is mu(1 - x1^2)x2 - x1; please correct the equation and verify the experiments used the intended dynamics.
  2. [Eqs. (5b), (10), (14b), (15)] The symbol delta is used both for the safety threshold and for a parameter of the Lotka-Volterra model; this notational conflict should be resolved.
  3. [Section 2, Eq. (1)] Equation (1) uses f(x) for the right-hand side, while the text immediately afterward says g: Rd -> Rd represents an unknown component; please align the notation, e.g., write dx/dt = g(x) throughout.
  4. [Eq. (8)] The notation N(y_m 1:N | x_l 1:N, sigma^2) should specify the covariance structure of the multivariate normal, typically sigma^2 I, to avoid ambiguity.
  5. [Figures 2-5] Figure captions are incomplete: Figure 3 has no axis description, and Figures 4 and 5 lack details on what is plotted and which quantities are averaged over the five seeds.
  6. [Section 7] The limitations section does not mention the two main caveats of this work: the discrete-time safety check and the reliance on GP posterior calibration for the safety probability. These should be explicitly acknowledged as limitations.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the acquisition and safety functions are computed from the current GP posterior and evaluated against external simulator benchmarks, not against the fitted quantities.

full rationale

The paper's derivation chain is self-contained. The acquisition function in Eq. (6)-(9) is a Monte Carlo estimate of mutual information between a candidate trajectory and the GP dynamics, and the safety condition in Eq. (10) is a Monte Carlo estimate of the stated probability that a trajectory stays within the box at the discretized measurement times. Neither quantity is defined in terms of the outcome it is used to predict. The method is evaluated on held-out validation negative log-likelihood and F1 scores computed from an external simulator (Van der Pol and Lotka-Volterra), so no fitted parameter is renamed as a prediction. The only self-citation, 'As in (Zimmer et al., 2018) we resort to a discretization' in Section 5.2, is a disclosed approximation rather than a load-bearing uniqueness theorem or a hidden ansatz; the paper explicitly acknowledges the discretization and its stiff-system limitations in Sections 3 and 7. The safety guarantee being only pointwise-in-time is a correctness and robustness concern, not a circularity, since the claim is evaluated against an independent ground-truth simulator.

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

The central claim rests on the ability of the GP posterior, obtained via variational sparse GP and decoupled sampling, to represent the uncertainty of the true dynamics both for information estimation and safety assessment. The method also relies on a user-chosen safety threshold and on a discrete-time safety check.

free parameters (4)
  • K (number of Monte Carlo samples)
    K controls the accuracy of the entropy estimate (Eq. 8) and the safety probability estimate (Eq. 10); no value is given in the paper, and performance depends on it.
  • δ (safety threshold) = user-chosen
    The threshold in the safety constraint (Eq. 5b, 10) determines how conservative the algorithm is; the paper does not analyze its effect.
  • L (number of inducing inputs)
    The sparse GP approximation (Eq. 2) depends on L; no value is reported.
  • S (number of random Fourier features)
    The prior approximation in decoupled sampling uses S basis functions; no value is reported.
assumptions (4)
  • domain assumption The unknown dynamics g in dx/dt = g(x) are a sample from a zero-mean Gaussian process.
    This is the GPODE modeling assumption from Hedge et al. 2022, used throughout Section 4. If the true dynamics are not well described by the GP prior, the acquisition and safety estimates are biased.
  • domain assumption The variational posterior and decoupled sampling provide a faithful characterization of the posterior over dynamics functions.
    Section 4 uses decoupled sampling (Eq. 2) for both training and active learning; the algorithm's estimates of entropy and safety are only as good as this approximation.
  • ad hoc to paper Safety can be assessed by checking the state only at the discrete measurement times t_i.
    Eq. (10) defines safety as x_min ≤ x(t_i) ≤ x_max for all i. This ignores possible violations between time points; the paper acknowledges the discretization but does not bound the error.
  • domain assumption The constrained optimization problem (11) can be solved to a good approximation in practice.
    The paper does not describe the optimizer or candidate set; it implicitly assumes a solver exists.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Safe Active Learning for Gaussian Differential Equations." pith.science (2026). https://pith.science/paper/ZLXUT6I2

@misc{pith2026241209053,
  author       = {Pith},
  title        = {Pith review of: Safe Active Learning for Gaussian Differential Equations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZLXUT6I2}},
  note         = {Machine review of arXiv:2412.09053}
}
read the original abstract

Gaussian Process differential equations (GPODE) have recently gained momentum due to their ability to capture dynamics behavior of systems and also represent uncertainty in predictions. Prior work has described the process of training the hyperparameters and, thereby, calibrating GPODE to data. How to design efficient algorithms to collect data for training GPODE models is still an open field of research. Nevertheless high-quality training data is key for model performance. Furthermore, data collection leads to time-cost and financial-cost and might in some areas even be safety critical to the system under test. Therefore, algorithms for safe and efficient data collection are central for building high quality GPODE models. Our novel Safe Active Learning (SAL) for GPODE algorithm addresses this challenge by suggesting a mechanism to propose efficient and non-safety-critical data to collect. SAL GPODE does so by sequentially suggesting new data, measuring it and updating the GPODE model with the new data. In this way, subsequent data points are iteratively suggested. The core of our SAL GPODE algorithm is a constrained optimization problem maximizing information of new data for GPODE model training constrained by the safety of the underlying system. We demonstrate our novel SAL GPODE's superiority compared to a standard, non-active way of measuring new data on two relevant examples.

Figures

Figures reproduced from arXiv: 2412.09053 by the authors.

Figure 1
Figure 1. The visualization shows the Safe Active Learning loop, from upper right: i) train/update a GPODE model, ii) plan new measurement by solving the constrained optimization problem that maximizes information constraint on safety, iii) conduct new measurement on the test system. • a new acquisition function measuring information of new candidate measurements • a new safety function allowing us to evaluate the safety of a… view at source ↗
Figure 2
Figure 2. shows that our novel SAL GPODE algorithm learns the systems dynamic much quicker than the random base￾line [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. shows that our novel SAL GPODE algorithm is also able to learn the safe area more quickly. We check how many initial states are correctly identified as safe or unsafe with our SAL GPODE in Equation (10) and how many are incorrectly identified and display the F1 score [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Validation negative log likelihood on the Lotka-Volterra task [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: F1 score on the Lotka-Volterra task [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 40 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    and Stueckler, J

    Achterhold, J. and Stueckler, J. Explore the context: Optimal data collection for context-conditional dynamics models. In International Conference on Artificial Intelligence and Statistics, pp.\ 3529--3537. PMLR, 2021

    work page 2021

  3. [3]

    Improving the efficiency of training physics-informed neural networks using active learning

    Aikawa, Y., Ueda, N., and Tanaka, T. Improving the efficiency of training physics-informed neural networks using active learning. New Generation Computing, pp.\ 1--22, 2024

    work page 2024

  4. [4]

    Arthurs, C. J. and King, A. P. Active training of physics-informed neural networks to aggregate and interpolate parametric solutions to the navier-stokes equations. Journal of Computational Physics, 438: 0 110364, 2021

    work page 2021

  5. [5]

    Atkinson, A. C. and Fedorov, V. V. Optimal design: Experiments for discriminating between several models. Biometrika, 62 0 (2): 0 289--303, 1975

    work page 1975

  6. [6]

    Berkenkamp, F., Krause, A., and Schoellig, A. P. Bayesian Optimization with Safety Constraints: Safe and Automatic Parameter Tuning in Robotics . Technical report, arXiv, February 2016 a

    work page 2016

  7. [7]

    P., and Krause, A

    Berkenkamp, F., Schoellig, A. P., and Krause, A. Safe controller optimization for quadrotors with G aussian processes. In 2016 IEEE international conference on robotics and automation (ICRA), pp.\ 491--496. IEEE, 2016 b

    work page 2016

  8. [8]

    Structural kernel search via bayesian optimization and symbolical optimal transport

    Bitzer, M., Meister, M., and Zimmer, C. Structural kernel search via bayesian optimization and symbolical optimal transport. Advances in Neural Information Processing Systems, 35: 0 39047--39058, 2022

    work page 2022

Show all 48 references
  1. [9]

    Amortized inference for G aussian process hyperparameters of structured kernels

    Bitzer, M., Meister, M., and Zimmer, C. Amortized inference for G aussian process hyperparameters of structured kernels. In Uncertainty in Artificial Intelligence, pp.\ 184--194. PMLR, 2023

    2023

  2. [10]

    Mathematical models in epidemiology, volume 32

    Brauer, F., Castillo-Chavez, C., Feng, Z., et al. Mathematical models in epidemiology, volume 32. Springer, 2019

    2019

  3. [11]

    Actively learning G aussian process dynamics

    Buisson-Fenet, M., Solowjow, F., and Trimpe, S. Actively learning G aussian process dynamics. In Learning for dynamics and control, pp.\ 5--15. PMLR, 2020

    2020

  4. [12]

    V., and Rackauckas, C

    Dandekar, R., Chung, K., Dixit, V., Tarek, M., Garcia-Valadez, A., Vemula, K. V., and Rackauckas, C. Bayesian neural ordinary differential equations. arXiv preprint arXiv:2012.07244, 2020

    2012 arXiv

  5. [13]

    Probabilistic recurrent state-space models

    Doerr, A., Daniel, C., Schiegg, M., Duy, N.-T., Schaal, S., Toussaint, M., and Sebastian, T. Probabilistic recurrent state-space models. In International conference on machine learning, pp.\ 1280--1289. PMLR, 2018

    2018

  6. [14]

    Exact inference for continuous-time G aussian process dynamics

    Ensinger, K., Tagliapietra, N., Ziesche, S., and Trimpe, S. Exact inference for continuous-time G aussian process dynamics. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pp.\ 11883--11891, 2024

    2024

  7. [15]

    A., and Hennig, P

    Garnett, R., Osborne, M. A., and Hennig, P. Active learning of linear embeddings for G aussian processes. arXiv preprint arXiv:1310.6740, 2013

    2013 arXiv

  8. [16]

    a hdesm \

    Hedge, P., Y ld z, C ., L \"a hdesm \"a ki, H., Kaski, S., and Heinonen, M. Variational multiple shooting for bayesian odes with G aussian processes. In Proceedings of the 38th Conference on Uncertainty in Artificial Intelligence (UAI 2022), PMLR, Proceedings of Machine Learni...

    2022

  9. [17]

    o m, H., Intosalmi, J., and L \

    Heinonen, M., Y ld z, C ., Mannerstr \"o m, H., Intosalmi, J., and L \"a hdesm \"a ki, H. Learning unknown ode models with G aussian processes. In International conference on machine learning, pp.\ 1959--1968. PMLR, 2018

    1959

  10. [18]

    Hensman, J., Matthews, A. G. d. G., Filippone, M., and Ghahramani, Z. Mcmc for variationally sparse G aussian processes. In Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 1, pp.\ 1648–1656, 2015

    2015

  11. [19]

    Hewing, L., Arcari, E., Fröhlich, L., and Zeilinger, M. N. On simulation and trajectory prediction with G aussian process dynamics. In Proceedings of the 2nd Conference on Learning for Dynamics and Control, volume 120, pp.\ 424 -- 434. PMLR, 2020

    2020

  12. [20]

    Bayesian active learning for classification and preference learning

    Houlsby, N., Husz \'a r, F., Ghahramani, Z., and Lengyel, M. Bayesian active learning for classification and preference learning. arXiv preprint arXiv:1112.5745, 2011

    2011 arXiv

  13. [21]

    Jordan, M. I. and Mitchell, T. M. Machine learning: Trends, perspectives, and prospects. Science, 349 0 (6245): 0 255--260, 2015

    2015

  14. [22]

    J., Porikli, F., and Papanikolopoulos, N

    Joshi, A. J., Porikli, F., and Papanikolopoulos, N. Multiclass active learning for image classification . In IEEE Conf. on Computer Vision and Pattern Recognition, pp.\ 2372--2379, 2009

    2009

  15. [23]

    On neural differential equations

    Kidger, P. On neural differential equations. PhD thesis, University of Oxford, 2021

    2021

  16. [24]

    Stiff neural ordinary differential equations

    Kim, S., Ji, W., Deng, S., Ma, Y., and Rackauckas, C. Stiff neural ordinary differential equations. Chaos: An Interdisciplinary Journal of Nonlinear Science, 31 0 (9), September 2021. ISSN 1089-7682. doi:10.1063/5.0060697

    2021 doi

  17. [25]

    Near-optimal sensor placements in G aussian processes: Theory, efficient algorithms and empirical studies

    Krause, A., Singh, A., and Guestrin, C. Near-optimal sensor placements in G aussian processes: Theory, efficient algorithms and empirical studies. Journal of Machine Learning Research, 9 0 (2), 2008

    2008

  18. [26]

    Hybrid modeling: towards the next level of scientific computing in engineering

    Kurz, S., De Gersem, H., Galetzka, A., Klaedtke, A., Liebsch, M., Loukrezis, D., Russenschuck, S., and Schmidt, M. Hybrid modeling: towards the next level of scientific computing in engineering. Journal of Mathematics in Industry, 12 0 (1): 0 8, 2022

    2022

  19. [27]

    Safe active learning for multi-output G aussian processes

    Li, C.-Y., Rakitsch, B., and Zimmer, C. Safe active learning for multi-output G aussian processes. In International Conference on Artificial Intelligence and Statistics, pp.\ 4512--4551. PMLR, 2022

    2022

  20. [28]

    J., and Adams, R

    Liu, S., Sun, X., Ramadge, P. J., and Adams, R. P. Task-agnostic amortized inference of G aussian process hyperparameters. Advances in Neural Information Processing Systems, 33: 0 21440--21452, 2020

    2020

  21. [29]

    Computational modeling of biochemical networks using copasi

    Mendes, P., Hoops, S., Sahle, S., Gauges, R., Dada, J., and Kummer, U. Computational modeling of biochemical networks using copasi. Systems Biology, pp.\ 17--59, 2009

    2009

  22. [30]

    Uncertainty and structure in neural ordinary differential equations

    Ott, K., Tiemann, M., and Hennig, P. Uncertainty and structure in neural ordinary differential equations. arXiv preprint arXiv:2305.13290, 2023

    2023 arXiv

  23. [31]

    J., and Sager, S

    Plate, C., Martensen, C. J., and Sager, S. Optimal experimental design for universal differential equations. arXiv preprint arXiv:2408.07143, 2024

    2024 arXiv

  24. [32]

    Mathematical models in science and engineering

    Quarteroni, A. Mathematical models in science and engineering. Notices of the AMS, 56 0 (1): 0 10--19, 2009

    2009

  25. [33]

    Universal differential equations for scientific machine learning

    Rackauckas, C., Ma, Y., Martensen, J., Warner, C., Zubov, K., Supekar, R., Skinner, D., Ramadhan, A., and Edelman, A. Universal differential equations for scientific machine learning. arXiv preprint arXiv:2001.04385, 2020

    2001 arXiv

  26. [34]

    and Recht, B

    Rahimi, A. and Recht, B. Random features for large-scale kernel machines. In Advances in Neural Information Processing Systems, volume 20, 2008

    2008

  27. [35]

    and Kevrekidis, I

    Rico-Martinez, R. and Kevrekidis, I. Continuous time modeling of nonlinear systems: a neural network-based approach. In IEEE International Conference on Neural Networks, pp.\ 1522--1525 vol.3, 1993. doi:10.1109/ICNN.1993.298782

    1993

  28. [36]

    a fer, A., Mara, H., Freudenreich, J., Breuckmann, B., D \

    Sch \"a fer, A., Mara, H., Freudenreich, J., Breuckmann, B., D \"u ffort, C., and Bock, G. Large scale angkor style reliefs: high definition 3d acquisition and improved visualization using local feature estimation. In Revive the Past. CAA 2011--Proceedings of the 39th Conferen...

    2011

  29. [37]

    Safe Exploration for Active Learning with G aussian Processes

    Schreiter, J., Nguyen-Tuong, D., Eberts, M., Bischoff, B., Markert, H., and Toussaint, M. Safe Exploration for Active Learning with G aussian Processes . In ECML/PKDD, volume 9286, 2015

    2015

  30. [38]

    Active learning literature survey

    Settles, B. Active learning literature survey. 2009

    2009

  31. [39]

    Stagewise safe bayesian optimization with G aussian processes

    Sui, Y., Zhuang, V., Burdick, J., and Yue, Y. Stagewise safe bayesian optimization with G aussian processes. In International conference on machine learning, pp.\ 4781--4789. PMLR, 2018

    2018

  32. [40]

    Efficiently computable safety bounds for G aussian processes in active learning

    Tebbe, J., Zimmer, C., Steland, A., Lange-Hegermann, M., and Mies, F. Efficiently computable safety bounds for G aussian processes in active learning. In International Conference on Artificial Intelligence and Statistics, pp.\ 1333--1341. PMLR, 2024

    2024

  33. [41]

    Variational learning of inducing variables in sparse G aussian processes

    Titsias, M. Variational learning of inducing variables in sparse G aussian processes. In Proceedings of the Twelfth International Conference on Artificial Intelligence and Statistics, volume 5 of Proceedings of Machine Learning Research, pp.\ 567--574, 2009

    2009

  34. [42]

    Williams, C. K. and Rasmussen, C. E. G aussian processes for machine learning , volume 2. MIT press Cambridge, MA, 2006

    2006

  35. [43]

    Efficiently sampling functions from G aussian process posteriors

    Wilson, J., Borovitskiy, V., Terenin, A., Mostowsky, P., and Deisenroth, M. Efficiently sampling functions from G aussian process posteriors. In Proceedings of the 37th International Conference on Machine Learning, volume 119, pp.\ 10292--10302, 2020

    2020

  36. [44]

    L., Trayanova, N., Geman, D., and Miller, M

    Winslow, R. L., Trayanova, N., Geman, D., and Miller, M. I. Computational medicine: Translating models to clinical care. Science Translational Medicine, 4 0 (158): 0 158rv11--158rv11, 2012. doi:10.1126/scitranslmed.3003528. URL https://www.science.org/doi/abs/10.1126/scitransl...

    2012 doi

  37. [45]

    Learning interacting dynamical systems with latent G aussian process odes

    Y ld z, C ., Kandemir, M., and Rakitsch, B. Learning interacting dynamical systems with latent G aussian process odes. Advances in Neural Information Processing Systems, 35: 0 9188--9200, 2022

    2022

  38. [46]

    Yu, H. S. A., Yao, D., Zimmer, C., Toussaint, M., and Nguyen-Tuong, D. Active learning in G aussian process state space model. In Machine Learning and Knowledge Discovery in Databases. Research Track: European Conference, ECML PKDD 2021, Bilbao, Spain, September 13--17, 2021, ...

    2021

  39. [47]

    H., and Shi, J

    Yue, X., Wen, Y., Hunt, J. H., and Shi, J. Active learning for G aussian process considering uncertainties with application to shape control of composite fuselage. IEEE Transactions on Automation Science and Engineering, 18 0 (1): 0 36--46, 2020

    2020

  40. [48]

    Safe active learning for time-series modeling with G aussian processes

    Zimmer, C., Meister, M., and Nguyen-Tuong, D. Safe active learning for time-series modeling with G aussian processes. Advances in neural information processing systems, 31, 2018

    2018

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.