Pith. sign in

REVIEW 4 major objections 6 minor 39 references

Adaptive Bayesian Single-Shot Quantum Sensing

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that an adaptive Bayesian policy that selects probe and measurement parameters by maximizing active information gain lets a single quantum probe track a time-varying phase, and that multi-agent fusion of such estimates…

desk verdict Good new idea, but the estimator as implemented never uses the current measurement, so the central single-shot sensing claim is untested. read the letter →

arxiv 2507.16477 v1 pith:GYOLMCUK submitted 2025-07-22 quant-ph cs.LGeess.SP

Ivana Nikoloska , Ruud Van Sloun , Osvaldo Simeone This is my paper
classification quant-phcs.LGeess.SP
keywords quantumsensingvariationalcircuitsBayesianexperimentaldesignactiveinformationgainsingle-shotestimationsensorfusionmagnetometryNISQ
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces a fully Bayesian, online protocol for variational quantum sensing in which a single probe is deployed at each time step. Instead of fixing probe and measurement parameters in advance, the sensor treats them as actions chosen to maximize the expected reduction in uncertainty about the target parameter, measured by the active information gain between the observation and the parameter. The paper claims that this information-seeking policy enables accurate estimation of a time-varying quantum phase, that it remains accurate when the probe parameters are corrupted by noise, and that fusing the estimates of several independently acting agents yields more robust estimates than a single agent spending the same number of probes. The practical payoff is a blueprint for using current NISQ devices as agile single-shot sensors without precomputed optimal protocols.

What carries the argument

The central object is the active information gain, defined as the mutual information $I_q(s_t; x_t \mid a_t, \mathcal{D}_{0:t-1})$ computed under the variational joint model $q(s_t, x_t | a_t, \mathcal{D}_{0:t-1}) = q(s_t | x_t, a_t) q(x_t | \mathcal{D}_{0:t-1})$. The first factor is an observation model built from a simulator of the quantum channel, and the second is the Gaussian world model whose mean is the output of a neural network. Each time step, the agent takes one gradient step on this objective—evaluated by Monte Carlo sampling and differentiated via parameter-shift rules—to update the probe parameters $\theta_t$ and measurement parameters $\mu_t$, then updates the world-model weights with one gradient step on the squared prediction error using feedback of the true parameter. The same machinery, with per-agent world models and inverse-variance fusion, underlies the multi-sensor extension.

What would settle it

Run the protocol on a target that changes randomly at each time step with no temporal correlation, so that past data carry no information about the current phase; if the estimator still ignores the current outcome, the estimates cannot track the jumps, and the claim that information-gain-maximizing probes enable estimation would fail. Alternatively, compare the published estimator $f_w(\mathcal{D}_{0:t-1})$ against one that also conditions on the current outcome $s_t$ on the sawtooth task; if the two perform identically, the current measurement is not actually being used for estimation.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the epistemic value of a sensing action—quantified as the mutual information $I_q(s_t; x_t \mid a_t, \mathcal{D}_{0:t-1})$ between the measurement outcome and the target parameter under the agent's variational world model—is a sufficient guide for online probe design. By selecting probe parameters $\theta_t$ and measurement parameters $\mu_t$ that maximize this quantity at every time step, the agent learns a world model $q(x_t | \mathcal{D}_{0:t-1}) = \mathcal{N}(f_w(\mathcal{D}_{0:t-1}), \sigma)$ whose posterior mean $f_w(\mathcal{D}_{0:t-1})$ tracks the true phase over time in a sawtooth magnetometry task. The same principle extends to $K$ agents, each maximizing its own information gain, whose posterior-mean estimates are combined by inverse-variance weighting. The paper interprets the performance gain over a static Bayesian scheme with random actions as evidence that the information-seeking policy supplies the agent with the data it needs to learn a strong world model, rather than the world model being informative on its own.

Load-bearing premise

The whole protocol rests on the assumption that predicting the current target from past true values and past observations is equivalent to estimating it from the current measurement outcome, since the estimator never conditions on $s_t$; the sawtooth experiments, which feed the true value back, do not test this equivalence.

Editorial extensions

If this is right

  • Probe and measurement parameters can be re-optimized online with a single gradient step per time step, making the protocol compatible with NISQ hardware and with the single-probe-per-step constraint.
  • An information-seeking policy is sufficient for accurate tracking of a deterministic sawtooth phase even when the probe parameters are corrupted by Gaussian noise.
  • Fusing the estimates of K independently acting agents gives lower-variance estimates than one agent using K probes, at K times the probe-preparation overhead.
  • The approach replaces offline batch optimization of variational quantum sensors with an online Bayesian alternative, which the paper argues is missing in current VQS literature.

Reading between the lines

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

  • Inference: If the equivalence between predicting from past feedback and estimating from the current measurement were established, the same machinery could be applied to non-deterministic or abruptly changing targets, where the current outcome carries information unavailable in past data.
  • Inference: A natural testable extension would be to condition the estimator on the current outcome $s_t$ as well as on past data; the paper's results would then serve as a lower bound on what the adaptive policy can achieve.
  • Inference: The reliance on a known simulator (or a Stinespring dilation on a NISQ device) means the protocol's practical value depends on the fidelity of the surrogate channel; an interesting stress test would be to evaluate the policy when the simulator mismatches the true channel.
  • Inference: The framework is phrased for phase estimation, but the same active-information-gain objective is immediately portable to other parameters encoded in quantum channels, such as field strengths or temperatures in gradient sensing.
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

4 major / 6 minor

Summary. The manuscript proposes an adaptive Bayesian framework for variational quantum sensing. At each time step t, the agent chooses probe-preparation parameters θ_t and measurement parameters μ_t by maximizing the active information gain I_q(s_t; x_t | a_t, D_{0:t−1}) under a learned variational world model. The estimate xhat_t is claimed to be produced from the observed sample s_t. The framework is extended to K agents whose local estimates are fused by inverse-variance weighting. Numerical experiments on a sawtooth phase-encoding channel, with and without noise, are used to argue that information-maximizing actions enable accurate single-probe estimation and that multi-agent fusion improves robustness.

Significance. If the central claim were supported, this would be a useful contribution to adaptive variational quantum sensing: a Bayesian alternative to frequentist batch optimization, with a principled information-theoretic acquisition function and a multi-agent extension. The paper is clearly written and the probabilistic machinery in Sections 3.1–3.4 is standard. However, the validation suffers from a load-bearing mismatch: the implemented estimator does not condition on the current measurement, and the experiments therefore do not demonstrate single-shot quantum sensing. The comparison benchmark also confounds the action-selection policy with the estimator architecture.

major comments (4)
  1. [§3.5, Eq. (5) and (10)] There is an internal inconsistency between the sensing task stated in Section 2 and the estimator implemented in Section 3.5. Section 2 states that 'using the observed data sample s_t, an estimate xhat_t of the target parameter x_t is finally produced,' but Eq. (5) defines the posterior as q(x_t|D_{0:t−1}) = N(f_w(D_{0:t−1}), σ), and Section 3.5 sets xhat_t = f_w(D_{0:t−1}). Since D_{0:t−1} contains no information from the current probe, the estimate is the prior mean before s_t is observed. The MAP argument in Section 3.5 is also incomplete: the posterior after observing s_t is proportional to q(s_t|x_t,a_t) q(x_t|D_{0:t−1}), whose mode is not generally f_w(D_{0:t−1}). As written, the method performs one-step-ahead prediction from past data, not estimation from the current measurement, so the central claim in Section 5.5 that information-maximizing probe parameters 'effectively enable accurate estimation of the target phase over time' is not supported.
  2. [§5.3 and §5.5] The experimental protocol cannot distinguish sensing from supervised prediction. The world-model input is only the previous-step triplet {s_{t−1}, x_{t−1}, a_{t−1}}, and the online update (10) feeds the true x_t to the network at every step. Since the target process is a deterministic sawtooth, a neural sequence predictor trained with true labels could track this process without using the quantum measurement at all. To support the single-shot sensing claim, the authors should either (i) change the estimator so that it conditions on s_t, or (ii) provide a control experiment in which s_t is withheld and show that performance degrades, and compare against a non-adaptive or random-action policy using the same estimator architecture rather than a different LSTM.
  3. [§3.4, Eq. (8)] The active information gain is evaluated under the agent's own variational observation model q(s_t|x_t,a_t). This is a standard Bayesian experimental design objective, but it does not by itself establish that maximizing I_q reduces estimation error with respect to the true likelihood. Since the central claim is empirical, the paper should report results that separate the effect of the acquisition function from the effect of the estimator update, for example by comparing I_q-maximizing actions with random actions under the identical world model and estimator. The current Static Bayesian benchmark uses a different LSTM estimator, so the comparison in Figure 4 confounds architecture and policy.
  4. [§4.2, Eq. (13)] The multi-agent extension inherits the same flaw: each local estimate xhat^k_t = f_{w_k}(D^k_{0:t−1}) from Eq. (11) is a prediction from past data, and the fusion rule (13) averages these predictions rather than fusing likelihoods based on the current observations s^k_t. Consequently, Figure 6 demonstrates fusion of supervised predictors, not fusion of quantum measurements. The authors should reformulate the local estimators to condition on the current measurement, or clearly restate the problem as one of prediction rather than single-shot estimation.
minor comments (6)
  1. [§5.3] The sentence 'The model takes as input the past' in Section 5.3 is incomplete and should be finished or removed.
  2. [§4.2, Eq. (13)] The summation in Eq. (13) runs over k = 0,...,K, but the sensors are indexed k = 1,...,K; it should presumably be k = 1,...,K.
  3. [§5.2] 'meteorological probes' should read 'metrological probes'.
  4. [§6] In the first future-research sentence, 'sensing precession' should be 'sensing precision'.
  5. [§1.2, Ref. [15]] Reference [15] (Jaakkola and Jordan) is cited as a Bayesian framework for VQS, but that paper is a general variational inference paper and does not appear to address quantum sensing; the citation should be checked and corrected.
  6. [§5.5, Fig. 7] The bit-flip noise model used for Figure 7 is not fully specified; please state whether the Pauli X errors are applied per qubit per layer and how the noise probability p relates to the circuit depth.

Circularity Check

2 steps flagged · score 6.0 of 10

The 'single-shot estimate' is defined as the prior mean before the shot, so the sensing result reduces to supervised sequence prediction from true-feedback data.

  1. fitted input called prediction [Section 3.5 (Estimator), Eqs. (5) and (10); Section 5.3 (World and Observation Models)]
    "Given the variational posterior (5), this yields the posterior mean estimator xhat_t = f_w(D_{0:t−1}) which is the output of the world model. ... wt+1←wt−η∇w(xt−fw(Dt))2. ... The input is given by the concatenated triplet {s_{t−1}, x_{t−1}, a_{t−1}} from the previous step only."

    By construction, xhat_t is the mean of q(x_t|D_{0:t−1}), the belief before the current measurement s_t is observed; s_t does not appear in the estimator. The current outcome enters only the action-selection objective (9) and the online update (10), which additionally requires the true value x_t. Because f_w is trained with the supervised loss (x_t − f_w(D_t))^2 on the deterministic sawtooth sequence (14), the estimate is a supervised sequence predictor fed with past true values, not a function of the current probe outcome. The Section 2 statement that 'using the observed data sample s_t, an estimate xhat_t is finally produced' is therefore not implemented by the paper's own equations.

  2. self definitional [Section 5.5 (Results), with Section 3.5 and Eq. (14)]
    "the probe parameters that maximize information gain Iq(ˆst, ˆxt|at,D0:t−1), as selected by the proposed adaptive Bayesian scheme, can effectively enable accurate estimation of the target phase over time."

    This central claim attributes accurate estimation to information-maximizing probe parameters, but the displayed estimate is defined in Section 3.5 as xhat_t = f_w(D_{0:t−1}), which is independent of the current measurement s_t. In the experiments the target x_t is a deterministic sawtooth and the sensor receives true x_t as feedback for training via (10). Hence the validation demonstrates that a neural network can track a known deterministic sequence from past true values; it does not demonstrate that the measurement outcome enables the estimate. The claim is self-definitional with respect to the estimator: the estimated trajectory is the world-model output by definition, not a posterior conditioned on the current shot.

full rationale

The main circularity is internal to the estimator. Eq. (5) defines the variational posterior as N(f_w(D_{0:t−1}), σ), and Section 3.5 sets xhat_t = f_w(D_{0:t−1}); thus the quantity called the single-shot estimate is the prior mean before observing s_t. The active-information-gain objective (8)-(9) is a standard Bayesian experimental-design criterion, and the self-citations to [12], [22], [24], and [32] are contextual rather than load-bearing, so the score is not driven by self-citation. What raises the score is that the central empirical claim reduces by construction to a supervised sequence predictor: the online update (10) feeds true x_t, the input is the previous triplet {s_{t−1}, x_{t−1}, a_{t−1}}, and the target is a deterministic sawtooth, so accurate 'estimation' does not depend on the current measurement outcome. This is a definitional reduction of the claimed sensing prediction to the world-model output, giving partial circularity (score 6) rather than a fully self-citation-forced result (score 8-10).

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

The protocol rests on a small set of explicit assumptions: access to a simulator, access to ground-truth feedback at each step, and a Gaussian world model. It introduces no new physical entities. The experimental claims depend on the learned weights and on manually chosen task and noise parameters, none of which are released.

free parameters (4)
  • world model weights w and per-agent variance network weights w_k = not released
    The estimator and the fusion weights are outputs of neural networks trained online by Eq. (10); no trained checkpoints or random seeds are provided.
  • Gaussian variance sigma (single agent) and sigma_k (multi-agent) = network output
    The posterior width is learned, not measured; it directly enters the fusion weights in Eq. (13).
  • task and training hyperparameters = P=15 s, T=100, n=6, 3x256 ReLU, lr=0.001
    Chosen by hand; no sensitivity analysis is given, so the demonstration may not generalize.
  • noise magnitudes = Gaussian sigma=0.1, bit-flip probabilities 0.1, 0.2, 0.4
    These values drive the robustness claims without sensitivity or calibration details.
assumptions (5)
  • domain assumption Ground-truth feedback x_t is available after every estimate.
    Online training of the world model via Eq. (10) requires the true parameter at each step; in real sensing this may not be available. Stated in Section 2, final paragraph, and used in Section 3.5.
  • domain assumption A simulator E_hat(x_t) of the target quantum channel is available to score hypothetical observations.
    The mutual information estimate assumes the simulator matches the true channel; mismatch would bias the acquisition function. Introduced in Section 3.3.
  • domain assumption The posterior belief is constrained to a parametric Gaussian q(x_t|D) = N(f_w(D), sigma).
    All estimation claims depend on this model being adequate; no calibration checks or posterior checks are reported. Used in Eq. (5).
  • domain assumption Measurements are local complete POVMs in a rotated basis, and the action space is restricted to the chosen ansatz.
    Optimality of the policy is only with respect to this restricted class of circuits and measurements, not over all quantum strategies. Defined in Section 2 and Fig. 3.
  • standard math Gradients of the mutual information are estimated by Monte Carlo sampling and parameter-shift rules.
    The validity of the optimization relies on standard unbiased gradient estimators; implementation details are not given, but the principles are standard. Mentioned in Section 3.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Bayesian Single-Shot Quantum Sensing." pith.science (2026). https://pith.science/paper/GYOLMCUK

@misc{pith2026250716477,
  author       = {Pith},
  title        = {Pith review of: Adaptive Bayesian Single-Shot Quantum Sensing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GYOLMCUK}},
  note         = {Machine review of arXiv:2507.16477}
}
read the original abstract

Quantum sensing harnesses the unique properties of quantum systems to enable precision measurements of physical quantities such as time, magnetic and electric fields, acceleration, and gravitational gradients well beyond the limits of classical sensors. However, identifying suitable sensing probes and measurement schemes can be a classically intractable task, as it requires optimizing over Hilbert spaces of high dimension. In variational quantum sensing, a probe quantum system is generated via a parameterized quantum circuit (PQC), exposed to an unknown physical parameter through a quantum channel, and measured to collect classical data. PQCs and measurements are typically optimized using offline strategies based on frequentist learning criteria. This paper introduces an adaptive protocol that uses Bayesian inference to optimize the sensing policy via the maximization of the active information gain. The proposed variational methodology is tailored for non-asymptotic regimes where a single probe can be deployed in each time step, and is extended to support the fusion of estimates from multiple quantum sensing agents.

Figures

Figures reproduced from arXiv: 2507.16477 by the authors.

Figure 1
Figure 1. (left) An illustration of variational quantum sensing. A probe quantum state is generated by a [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Target parameter xt as a function of time t. 5.3 World and Observation Models The mean for the world model fw(D0:t) is obtained as the output of a neural network parametrized by the parameters w. In a sim￾ilar way, in the multi-agent setting, the mean fwk (Dk 0:t ) and the log-variance gwk (Dk 0:t−1 ) are obtained using a neural network with parame￾ters w k and two output neurons. The neural network is comprised of … view at source ↗
Figure 3
Figure 3. Probe-preparation ansatz used in the experiments. We consider an S(n)-equivariant ansatz consisting of [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Phase estimates as a function of time. We [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 7
Figure 7. Figure 7: Standard deviation of the estimation errors [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 33 canonical work pages

  1. [1]

    Quantum sensing,

    C. L. Degen, F. Reinhard, and P. Cappel- laro, “Quantum sensing,”Reviews of modern physics, vol. 89, no. 3, p. 035002, 2017

    work page 2017

  2. [2]

    Exposing gravitational waves be- low the quantum sensing limit,

    H. Yu, D. Martynov, R. X. Adhikari, and Y. Chen, “Exposing gravitational waves be- low the quantum sensing limit,”Physical Re- view D, vol. 106, no. 6, p. 063017, 2022

    work page 2022

  3. [3]

    Quan- tum measurement theory in gravitational- wave detectors,

    S. L. Danilishin and F. Y. Khalili, “Quan- tum measurement theory in gravitational- wave detectors,” Living Reviews in Relativ- ity, vol. 15, pp. 1–147, 2012

    work page 2012

  4. [4]

    Blueprint for nanoscale nmr,

    I. Schwartz, J. Rosskopf, S. Schmitt, B. Tratzmiller, Q. Chen, L. P. McGuinness, F. Jelezko, and M. B. Plenio, “Blueprint for nanoscale nmr,” Scientific reports, vol. 9, no. 1, p. 6938, 2019

    work page 2019

  5. [5]

    Roadmap on nanoscale magnetic res- onance imaging,

    R. Budakian, A. Finkler, A. Eichler, M. Pog- gio, C. L. Degen, S. Tabatabaei, I. Lee, P. C. Hammel, S. P. Eugene, T. H. Taminiau et al., “Roadmap on nanoscale magnetic res- onance imaging,” Nanotechnology, vol. 35, no. 41, p. 412001, 2024

    work page 2024

  6. [6]

    Schuld and F

    M. Schuld and F. Petruccione, Machine learning with quantum computers. Springer, 2021

    work page 2021

  7. [7]

    An introduction to quantum machinelearningforengineers,

    O. Simeone, “An introduction to quantum machinelearningforengineers,” Foundations and Trends® in Signal Processing, 2022

    work page 2022

  8. [8]

    A variational toolbox for quantum multi- parameter estimation,

    J. J. Meyer, J. Borregaard, and J. Eisert, “A variational toolbox for quantum multi- parameter estimation,” npj Quantum Infor- mation, vol. 7, no. 1, p. 89, 2021

    work page 2021

Show all 39 references
  1. [9]

    End-to-end varia- tional quantum sensing,

    B. MacLellan, P. Roztocki, S. Czischek, and R. G. Melko, “End-to-end varia- tional quantum sensing,” arXiv preprint arXiv:2403.02394, 2024

    2024 arXiv

  2. [10]

    A ma- chine learning approach to bayesian param- eter estimation,

    S. Nolan, A. Smerzi, and L. Pezzè, “A ma- chine learning approach to bayesian param- eter estimation,” npj Quantum Information, vol. 7, no. 1, p. 169, 2021

    2021

  3. [11]

    Parameter es- timation in quantum sensing based on deep reinforcement learning,

    T. Xiao, J. Fan, and G. Zeng, “Parameter es- timation in quantum sensing based on deep reinforcement learning,”npj Quantum Infor- mation, vol. 8, no. 1, p. 2, 2022

    2022

  4. [12]

    Dynamic estimation loss con- trol in variational quantum sensing via on- line conformal inference,

    I. Nikoloska, H. Joudeh, R. van Sloun, and O. Simeone, “Dynamic estimation loss con- trol in variational quantum sensing via on- line conformal inference,” arXiv preprint arXiv:2505.23389, 2025

    2025 arXiv

  5. [13]

    Optimal and robust quantum metrology us- inginteraction-basedreadouts,

    S. P. Nolan, S. S. Szigeti, and S. A. Haine, “Optimal and robust quantum metrology us- inginteraction-basedreadouts,” Physical Re- view Letters, vol. 119, no. 19, p. 193601, 2017

    2017

  6. [14]

    Quantum graph neural networks,

    G. Verdon, T. McCourt, E. Luzhnica, V. Singh, S. Leichenauer, and J. Hidary, “Quantum graph neural networks,” arXiv preprint arXiv:1909.12264, 2019

    1909 arXiv

  7. [15]

    Bayesian parameter estimation via variational meth- ods,

    T. S. Jaakkola and M. I. Jordan, “Bayesian parameter estimation via variational meth- ods,” Statistics and Computing, vol. 10, pp. 25–37, 2000

    2000

  8. [16]

    Adaptive quantum state tomography with neural net- works,

    Y. Quek, S. Fort, and H. K. Ng, “Adaptive quantum state tomography with neural net- works,” npj Quantum Information, vol. 7, no. 1, p. 105, 2021

    2021

  9. [17]

    AdaptiveQuantumMetrol- ogy under General Markovian Noise,

    R. Demkowicz-Dobrzański, J. Czajkowski, andP.Sekatski, “AdaptiveQuantumMetrol- ogy under General Markovian Noise,”Phys. Rev. X, vol. 7, p. 041009, 2017

    2017

  10. [18]

    Von der Linden, V

    W. Von der Linden, V. Dose, and U. Von Toussaint,Bayesian probability the- ory: applications in the physical sciences . Cambridge University Press, 2014

    2014

  11. [19]

    Bayesian experimental design for implicit models by mutual information neural esti- mation,

    S. Kleinegesse and M. U. Gutmann, “Bayesian experimental design for implicit models by mutual information neural esti- mation,” inInternational conference on ma- chine learning. PMLR, 2020, pp. 5316– 5326

    2020

  12. [20]

    Bayesian optimization with known experimental and design con- straints for chemistry applications,

    R. J. Hickman, M. Aldeghi, F. Häse, and A. Aspuru-Guzik, “Bayesian optimization with known experimental and design con- straints for chemistry applications,”Digital Discovery, vol. 1, no. 5, pp. 732–744, 2022

    2022

  13. [21]

    Experimental multiparameter quan- tum metrology in adaptive regime,

    M.Valeri, V.Cimini, S.Piacentini, F.Cecca- relli, E. Polino, F. Hoch, G. Bizzarri, G. Cor- rielli, N. Spagnolo, R. Osellame, and F. Scia- 9 rrino, “Experimental multiparameter quan- tum metrology in adaptive regime,” Phys. Rev. Research, vol. 5, p. 013138, 2023

    2023

  14. [22]

    Mastering diverse control tasks through world models,

    D. Hafner, J. Pasukonis, J. Ba, and T. Lil- licrap, “Mastering diverse control tasks through world models,” Nature, pp. 1–7, 2025

    2025

  15. [23]

    Masked world mod- els for visual control,

    Y. Seo, D. Hafner, H. Liu, F. Liu, S. James, K. Lee, and P. Abbeel, “Masked world mod- els for visual control,” in Conference on Robot Learning. PMLR, 2023, pp. 1332– 1344

    2023

  16. [24]

    Active inference and deep generative modeling for cognitive ultra- sound,

    R. J. Van Sloun, “Active inference and deep generative modeling for cognitive ultra- sound,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 2024

    2024

  17. [25]

    The free energy principle for action and perception: A mathematical re- view,

    C. L. Buckley, C. S. Kim, S. McGregor, and A. K. Seth, “The free energy principle for action and perception: A mathematical re- view,” Journal of mathematical psychology, vol. 81, pp. 55–79, 2017

    2017

  18. [26]

    Cog- nitive radar: Step toward bridging the gap between neuroscience and engineering,

    S. Haykin, Y. Xue, and P. Setoodeh, “Cog- nitive radar: Step toward bridging the gap between neuroscience and engineering,”Pro- ceedings of the IEEE, vol. 100, no. 11, pp. 3102–3130, 2012

    2012

  19. [27]

    C. M. Bishop and N. M. Nasrabadi,Pattern recognition and machine learning. Springer, 2006, vol. 4, no. 4

    2006

  20. [28]

    Conditional density estimation with bayesian normalis- ing flows,

    B. L. Trippe and R. E. Turner, “Conditional density estimation with bayesian normalis- ing flows,”arXiv preprint arXiv:1802.04908, 2018

    2018 arXiv

  21. [29]

    Diffusion models: A comprehensive survey of methods and applications,

    L. Yang, Z. Zhang, Y. Song, S. Hong, R. Xu, Y. Zhao, W. Zhang, B. Cui, and M.-H. Yang, “Diffusion models: A comprehensive survey of methods and applications,” ACM Com- puting Surveys, vol. 56, no. 4, pp. 1–39, 2023

    2023

  22. [30]

    Learning and leveraging world models in visual representation learning,

    Q. Garrido, M. Assran, N. Ballas, A. Bardes, L. Najman, and Y. LeCun, “Learning and leveraging world models in visual representation learning,” arXiv preprint arXiv:2403.00504, 2024

    2024 arXiv

  23. [31]

    Dilation theory,

    P. Busch, P. Lahti, J.-P. Pellonpää, K. Yli- nen, P. Busch, P. Lahti, J.-P. Pellonpää, and K. Ylinen, “Dilation theory,”Quantum Mea- surement, pp. 137–162, 2016

    2016

  24. [32]

    Simeone, Classical and Quantum Infor- mation Theory: Uncertainty, Information, and Correlation

    O. Simeone, Classical and Quantum Infor- mation Theory: Uncertainty, Information, and Correlation. Cambridge University Press, 2025

    2025

  25. [33]

    Schuld and F

    M. Schuld and F. Petruccione, Supervised learning with quantum computers. Springer, 2018, vol. 17

    2018

  26. [34]

    SciPy 1.0: Funda- mental Algorithms for Scientific Computing in Python,

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. van der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Lar- son, C. J. Carey, İ. Polat, Y. Fen...

    2020

  27. [35]

    Fundamentals of statistical sig- nal processing: Estimation theory,

    S. M. Kay, “Fundamentals of statistical sig- nal processing: Estimation theory,” 1993

    1993

  28. [36]

    Fusion of probability density functions,

    G. Koliander, Y. El-Laham, P. M. Djurić, and F. Hlawatsch, “Fusion of probability density functions,”Proceedings of the IEEE, vol. 110, no. 4, pp. 404–453, 2022

    2022

  29. [37]

    Composite-pulse magnetometry with a solid-state quantum sensor,

    C. D. Aiello, M. Hirose, and P. Cappel- laro, “Composite-pulse magnetometry with a solid-state quantum sensor,”Nature com- munications, vol. 4, no. 1, p. 1419, 2013

    2013

  30. [38]

    Distributed quantum fiber mag- netometry,

    S. Maayani, C. Foy, D. Englund, and Y. Fink, “Distributed quantum fiber mag- netometry,” Laser & Photonics Reviews , vol. 13, no. 7, p. 1900075, 2019

    2019

  31. [39]

    Equivariant quantum graph cir- cuits,

    P. Mernyei, K. Meichanetzidis, and I. I. Ceylan, “Equivariant quantum graph cir- cuits,” in International Conference on Ma- chine Learning. PMLR, 2022, pp. 15401– 15420. 10

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.