Pith. sign in

REVIEW 3 major objections 4 minor 16 references

Sequential Monte Carlo samplers can estimate intractable gradients accurately enough to replace expensive MCMC inner loops in stochastic optimisation, with linear convergence for the idealized recursion.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Sequential Monte Carlo samplers can approximate intractable gradients inside a first-order optimizer, yielding a general SOSMC framework that speeds up reward tuning of energy-based models in the reported settings.

T0 review reviewed 2026-08-03 challenge →

load-bearing objection SOSMC is a clean umbrella over existing SMC-for-optimization methods with honest experiments, but the theory only covers the idealized recursion and the gap to the closed-loop algorithm is admitted, not closed. the 3 major comments →

arxiv 2601.22003 v2 pith:Y6IW5HIE submitted 2026-01-29 stat.ML cs.LGstat.CO

Efficient Stochastic Optimisation via Sequential Monte Carlo

classification stat.ML cs.LGstat.CO MSC 62L2065C05
keywords sequential Monte Carlostochastic optimisationintractable gradientsenergy-based modelsreward tuningFeynman-Kac identityPolyak-Łojasiewicz conditionimportance sampling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Many machine-learning and statistics problems—maximum marginal likelihood estimation, training energy-based models, and reward tuning of generative models—require gradients that are expectations over an intractable, parameter-dependent distribution. Standard approaches run expensive Markov-chain Monte Carlo loops on the inside; this paper proposes running a sequential Monte Carlo (SMC) sampler instead, reusing weighted particles across optimisation steps. The paper shows that in an idealized setting where the SMC expectations are exact, the parameter update reduces to exact gradient descent and converges linearly under the Polyak-Łojasiewicz and smoothness assumptions. For the particle implementation, it argues that standard SMC bias and mean-squared-error rates of order 1/N should transfer, and demonstrates empirically, on reward-tuning of energy-based models and Langevin samplers, that the weighted-particle estimate tracks the true expectation better than unweighted persistent particles and requires shorter evaluation chains.

Core claim

The paper claims that the intractable gradient—an expectation under a parameter-dependent distribution—can be approximated by the SMC-weighted particle average g_k = Σ_i w_k^(i) H_{θ_k}(X_k^(i)), maintained across optimisation steps via a Feynman–Kac identity. With exact expectations, this weighted recursion recovers the true gradient and reduces the update to exact gradient descent, converging linearly at rate (1−γμ)^t under Polyak-Łojasiewicz and smoothness assumptions. The paper argues this motivates the SOSMC algorithm and notes existing particle-based optimisers as special cases. Empirically, on reward tuning of energy-based models, the weighted-particle method outperforms unweighted pe

What carries the argument

The central machinery is the Feynman–Kac identity for SMC samplers (Theorem 1), which expresses the intractable expectation π_{θ_k}(H_{θ_k}) as an importance-weighted average over a sequence of proposals, with incremental weights G_k(x_{k-1}, x_k) = Π_{θ_k}(x_k) L_{k-1}(x_k, x_{k-1}) / (Π_{θ_{k-1}}(x_{k-1}) K_k(x_{k-1}, x_k)). The algorithm (SOSMC, Algorithm 1) maintains N weighted particles across gradient-descent iterations: at each step, particles are propagated through a Markov kernel K_k, reweighted, possibly resampled when the effective sample size falls below a threshold, and the gradient estimate g_k is the normalized weighted average of H_{θ_k}. The idealized analysis (Lemma 1 and P

Load-bearing premise

The practical efficiency claim relies on the unproven assumption that standard SMC approximation errors stay controlled when the target distributions are themselves moved by the particle-weighted gradient estimates, so the optimisation loop does not amplify bias.

What would settle it

Run SOSMC on a small latent-variable model where exact gradients are available by numerical integration and check whether the bias of g_k decays like 1/N and the parameter path tracks exact gradient descent as N grows; if the bias accumulates with outer iterations or fails to vanish, the central claim is false.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If SMC-based gradient estimates are accurate enough, MMLE, EBM training, and reward tuning can replace expensive MCMC inner loops with a single SMC sweep per optimisation step, cutting compute substantially.
  • The idealized linear rate (1−γμ)^t means that whenever the loss is Polyak-Łojasiewicz and the gradient estimates are near-exact, the outer optimisation inherits fast convergence without inner-loop convergence requirements.
  • The weighted-particle estimator reduces non-equilibrium bias, so persistent particles lag less behind the evolving target; consequently, reliable gradient estimates can be obtained with far shorter (or no) fresh evaluation chains.
  • The framework subsumes several existing particle-based optimisers as special cases, giving a unified way to reason about their design (choice of forward/backward kernels, resampling threshold, step size).
  • The ESS-based adaptive step-size rule offers a practical, scale-free way to keep successive targets close enough to avoid weight degeneracy during optimisation.

Where Pith is reading between the lines

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

  • A direct test of the paper's logic would be to compare the closed-loop bias of g_k against high-accuracy MCMC gradients on a small problem; if the bias grows with the number of outer iterations rather than staying at the single-step SMC bias, the feedback loop is the weak point.
  • The ESS formula implies that step size should shrink in directions of large gradient norm or high Fisher information; a curvature-aware adaptive step rule could outperform the simple multiplicative rule in Remark 4.
  • The same construction—reusing weighted particles across parameter updates—should transfer to other optimisation problems of the form (2), such as policy-gradient estimation or stochastic variational inference, provided the kernels can be chosen tractably.
  • If the feedback bias is benign, particle counts could be much lower than for fixed-target SMC, because the optimisation dynamics themselves average out particle noise; this is testable by running SOSMC at N=100 versus N=10000 and comparing final objective values.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes SOSMC, a sequential Monte Carlo (SMC) framework for stochastic optimisation when the gradient of the loss is an intractable expectation under a parameter-dependent distribution. Algorithm 1 maintains a weighted particle system that is propagated with Markov kernels and periodically resampled; the weighted average (13) is used as a gradient estimate in a generic first-order optimiser. The theoretical section analyses an idealised version of the recursion in which Feynman–Kac expectations are computed exactly: Lemma 1 shows exact gradient recovery and Proposition 2 gives a (1−γμ)^t rate under a Polyak–Łojasiewicz condition. Section 4.2 discusses the effective sample size, with an exact Gaussian formula and local expansions of the χ² divergence. Experiments on reward tuning of Langevin processes, 2D EBMs, and MNIST compare SOSMC-ULA with ImpDiff and SOUL, reporting wall-clock and iteration-level convergence.

Significance. If the practical algorithm inherited the idealised convergence behaviour, the framework would be a useful general alternative to MCMC inner loops for MMLE, EBM training, and reward tuning. The paper is clearly written and the empirical section is carefully controlled: device-synchronised timing, equal numbers of kernel applications per outer step, and separate reporting of particle-based and fresh rewards are genuine strengths. The surrogate-loss derivations in Appendix F are also a useful practical contribution. However, the main theoretical support for Algorithm 1 itself is absent: Remark 3 explicitly defers the particle analysis, and the accompanying ESS analysis contains a substantively incorrect proposition. The central efficiency claim is therefore conditional on an unproved closed-loop bias control.

major comments (3)
  1. [§4.1, Remark 3, Algorithm 1 (Eq. 13)] The convergence result in Proposition 2 applies to an idealised recursion in which expectations are computed exactly. In Algorithm 1, the gradient estimate g_k in Eq. (13) is computed from the current weighted particles, and the next parameter θ_{k+1} is a function of that same particle system. The target sequence (π_{θ_k}) is therefore not exogenous; Remark 4 concedes that it is “an optimisation-coupled random quantity.” Standard SMC O(1/N) bias/MSE results apply to Feynman–Kac flows with deterministic or prespecified target sequences, and the paper gives no bound on the accumulated feedback bias E[g_k] − ∇ℓ(θ_k) when θ_k itself depends on past particles. This is the load-bearing gap: without such a bound, the practical convergence claim for Algorithm 1 is unsupported. The deferral in Remark 3 does not resolve it, even though the abstract states that convergence results are established
  2. [Proposition 5 (Appendix B), with Proposition 3 and Eq. (19)] Proposition 5 is mathematically incorrect. It concludes χ²(π_{θ_k}, π_{θ_{k-1}}) = sqrt(γ² ∇ℓ^T I ∇ℓ) + O(γ^{3/2}), whose leading term is O(γ). This contradicts the exact Gaussian formula (19), where χ² = exp(γ² ∥∇ℓ∥²_{Σ^{-1}}) − 1 = O(γ²), and also contradicts Proposition 3 and Proposition 4, which give χ² = γ² ∇ℓ^T I ∇ℓ + o(γ²). The proof applies Cauchy–Schwarz to obtain an L¹-type upper bound; an upper bound of order γ cannot be an asymptotic expansion for a quantity that is O(γ²). The displayed square root is also dimensionally inconsistent as an equality. This proposition should be corrected or removed; as written, it undermines the claimed ESS analysis.
  3. [§4.2, Eq. (19), Remark 4] The ESS analysis in Section 4.2 assumes the parameter update is θ_k = θ_{k−1} − γ∇ℓ(θ_{k−1}) with a fixed step size and exact gradients. Algorithm 1, however, uses an arbitrary optimiser (including Adam and SGD) and a stochastic, biased gradient estimate g_k. Remark 4 then introduces an adaptive γ_k driven by the ESS itself. Consequently, Eqs. (18)–(19) and Proposition 3 do not quantify the ESS of the actual algorithm; they describe a much simpler idealised recursion. This is acceptable if the section is explicitly framed as intuition for the idealised flow, but the current presentation invites the reader to apply these conclusions to SOSMC directly. Please state clearly that these results are heuristic for the closed-loop algorithm.
minor comments (4)
  1. [Proof of Proposition 2] The proof refers to “Proposition 1” but the statement it uses is Lemma 1. Please correct the cross-reference.
  2. [Appendix B, Proposition 5 proof] There is a typo: “Cauchy-Swartz” should be “Cauchy–Schwarz.” More importantly, the proof alternates between t and γ inconsistently; this should be cleaned up regardless of the corrected statement.
  3. [Algorithm 2] The line “logW_k = logW_k − max_i logW_i” is ambiguous; the subtraction should be applied elementwise per particle. Please clarify the notation.
  4. [Abstract / Section 1] The abstract says “we establish convergence results for the basic recursions defined by our methodology which SMC samplers approximate.” This is accurate only for the idealised recursions, not the particle algorithm. Rephrase to avoid overclaiming.

Circularity Check

0 steps flagged

No significant circularity: the idealized Feynman-Kac identity is a standard exact identity, and the closed-loop gap is explicitly acknowledged rather than disguised.

full rationale

The central derivation chain is not circular. Theorem 1 is an external standard Feynman-Kac identity (Del Moral 2004), and Lemma 1 shows only that the importance-weight recursion (8)/(10) gives E[HW]/E[W] = pi(H); this is a consistency property of an importance-sampling construction, not a fitted prediction. Proposition 2 then invokes a standard gradient-descent convergence theorem under A1, and it is explicitly restricted to the idealized exact-expectation recursion (14)-(16). The paper does not claim the particle algorithm inherits these rates: Remark 3 defers a particle-level analysis to future work, and Remark 4 explicitly states that the target sequence is 'an optimisation-coupled random quantity' when theta is updated from the weighted particles. That is an honest, acknowledged gap (a correctness/rigor risk, not circularity). The self-citations to Cuin et al. (2025) and Carbone et al. (2024) are used to identify special cases and to match algebraic weight expressions in Appendix A; they are not load-bearing for the convergence claims, and Appendix A derives the weights directly. Separate issues such as the dimensional inconsistency in Proposition 5 and the missing 'Proposition 1' reference are mathematical/correctness concerns, not circularity. The score of 2 reflects only minor non-load-bearing self-citation and the acknowledged closed-loop gap, not derivation-equivalent-to-input circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

The method itself introduces no fitted constants: the theoretical results are parameter-free given the assumptions. The free parameters listed are experimental/tuning constants (resampling thresholds, step sizes, learning rates) that the practical recipe depends on; the adaptive step-size multiplier c is never specified. The key unproven axiom is the transfer of standard SMC error bounds to the closed-loop target sequence.

free parameters (5)
  • ESS resampling threshold τ = 0.9 (chosen)
    Algorithm 1 and experiments (App. E.2.2); controls resampling frequency and is a tuning constant, not fitted to data.
  • Adaptive threshold τ_adapt = 0.95 (chosen)
    Remark 4 and App. E.2.2; triggers step-size adjustment in the adaptive scheme.
  • Adaptive step-size multiplier c = unspecified
    Remark 4 defines γ_k = c γ_{k-1} or γ_k = γ_{k-1}/c but never gives c; a free tuning constant that affects the adaptive behavior and is needed for replication.
  • ULA step-size γ = 0.1 (Langevin exp.) or 5e-3 (EBM exp.), adapted
    Kernel step-size in Algorithm 2; chosen per experiment and central to the weight computation.
  • Optimizer learning rate η = 0.1 / 2e-4 / 1e-4
    Adam/SGD learning rate in experiments; standard hyperparameter, not fitted to data.
axioms (6)
  • domain assumption Gradient can be written as ∇ℓ(θ) = E_{X∼πθ}[Hθ(X)] with πθ(x) ∝ e^{-Uθ(x)} and Zθ < ∞.
    Eqs. (1)-(2) define the entire problem class.
  • domain assumption A1: ℓ is μ-PL and L-smooth.
    Assumption A1 in Section 4.1; required for Prop. 2's rate.
  • standard math Forward/backward kernels K_n, L_{n-1} have densities so the incremental weights (8) are well-defined.
    SMC samplers methodology (Del Moral et al. 2006); assumed throughout Section 3.
  • ad hoc to paper Standard SMC error bounds (bias/MSE O(1/N)) extend to the closed-loop, algorithm-dependent target sequence (πθk).
    Remark 3 calls this 'natural to expect' but does not prove it; the target sequence depends on previous particle estimates (Remark 4), so textbook SMC bounds do not apply directly. This is the load-bearing premise for the practical claim.
  • domain assumption Local bounded score/Hessian and exponential moment conditions for the χ² expansions.
    Propositions 4-5, Appendix B; needed for the ESS analysis.
  • domain assumption ULA kernels mix sufficiently fast and importance weights have finite variance in the experiments.
    Implicit in all experiments; not verified for non-log-concave EBMs.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Stochastic Optimisation via Sequential Monte Carlo." pith.science (2026). https://pith.science/paper/Y6IW5HIE

@misc{pith2026260122003,
  author       = {Pith},
  title        = {Pith review of: Efficient Stochastic Optimisation via Sequential Monte Carlo},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y6IW5HIE}},
  note         = {Machine review of arXiv:2601.22003}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The problem of optimising functions with intractable gradients frequently arises in machine learning and statistics, ranging from maximum marginal likelihood estimation procedures to fine-tuning of generative models. Stochastic approximation methods for this class of problems typically require inner sampling loops to obtain (biased) stochastic gradient estimates, which rapidly becomes computationally expensive. In this work, we develop sequential Monte Carlo (SMC) samplers for optimisation of functions with intractable gradients. Our approach replaces expensive inner sampling methods with efficient SMC approximations, which can result in significant computational gains. We establish convergence results for the basic recursions defined by our methodology which SMC samplers approximate. We demonstrate the effectiveness of our approach on the reward-tuning of energy-based models within various settings.

Figures

Figures reproduced from arXiv: 2601.22003 by Davide Carbone, James Cuin, O. Deniz Akyildiz, Yanbo Tang.

Figure 1
Figure 1. Figure 1: Wall-clock convergence of mean reward (left) and NLL (right), across [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Terminal density snapshots of πθK , using identical initialisation and shared noise, across increasing βKL, for Rlower. 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 ̂ Rfresh 0.0 0.1 0.2 0.3 0.4 0.5 0.6 ̂ KL(π θ ‖ π0) 0.48 0.54 0.60 0.66 0.72 0.78 0.84 0.90 0.96 Best ̂ ℓ (ImpDiff)=0.829 ImpDiff SOSMC-ULA Best ̂ ℓ (ImpDiff) [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Tuning trajectories for ImpDiff and SOSMC￾ULA in the (Rbfresh, KL) c plane, for βKL = 0.25. Contours denote level sets of ℓ, with SOSMC-ULA to the right of (i.e. better than) the best ImpDiff solution achieved dur￾ing tuning (red). A key distinction in this setting is be￾tween (i) the particle reward Rbparticle(k), computed on the particles used for gradi￾ent estimation, and (ii) the fresh reward Rbfresh(k… view at source ↗
Figure 4
Figure 4. Figure 4: Density snapshots along the sampling evolution of [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example sampling trajectories, with step index, under the pre-training sampler kernel, for [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Wall-clock (top) and outer iteration (bottom) convergence of reward (left) and NLL (right), for [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Convergence of mean reward (left) and NLL (right), for [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Convergence of mean reward (left) and NLL (right), for [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Convergence of mean reward (left) and NLL (right), for [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: 2D Datasets and EBMs: The dataset samples (black), with corresponding contours, where darker lines indicating higher probability, and samples generated from the corresponding pretrained EBM (red), obtained after running 300 Langevin steps from random noise, for D2M, DC, and DB (from left to right). for i = 1, . . . , dx. With the choice of Cmin = −6 and Cmax = 6 the clamping restricts the particle evoluti… view at source ↗
Figure 11
Figure 11. Figure 11: Terminal density snapshots of πθK , using identical initialisation and shared noise, across increasing βKL, for Rlower. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Density snapshots along the sampling evolution of [PITH_FULL_IMAGE:figures/full_fig_p028_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Representative sampling trajectories for [PITH_FULL_IMAGE:figures/full_fig_p029_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Trajectories of Rbparticle and Rbfresh for the illustrated example. The (weighted) Rbparticle serves as a far better proxy for Rbfresh in the case of SOSMC-ULA, whilst Rbfresh also converges closer to π ∗ (H) for this method. 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 ̂ Rfresh 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 ̂ KL(π θ ‖ π 0) β = 0.1 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 0.60 0.65 0.70 0.7… view at source ↗
Figure 15
Figure 15. Figure 15: Tuning trajectories for ImpDiff and SOSMC-ULA in the (Rbfresh, KL) c plane, across values of βKL. Contours denote level sets of ℓ, with the best ImpDiff solution highlighted in red. Note this corresponds to a Rbfresh budget of 20, 000 Langevin steps, with 15, 000 of those being burn-in steps. 30 [PITH_FULL_IMAGE:figures/full_fig_p030_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Tuning trajectories for ImpDiff and SOSMC-ULA in the (Rbfresh, KL) c plane, across values of βKL. Contours denote level sets of ℓ, with the best ImpDiff solution highlighted in red. Note this corresponds to a Rbfresh budget of 5, 000 Langevin steps, with 500 of those being burn-in steps. 0 5000 10000 15000 20000 25000 30000 Step 0.5 0.6 0.7 0.8 0.9 1.0 Fraction of ̃ X entered H ImpDiff SOSMC-ULA [PITH_FU… view at source ↗
Figure 17
Figure 17. Figure 17: Proportion of sampling trajectory, for πθK , in the reward region as in the illustrated example, for both ImpDiff and SOSMC-ULA. Identical initialisation and shared noise are utilised for each trajec￾tory across the methods. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Samples obtained from the frozen πθ0 (top), and πθK across various values of βKL. In (a) Rbright favours brighter images, whereas in (b) Rdark favours darker images. All sampling trajectories are identically initialised and share noise. 33 [PITH_FULL_IMAGE:figures/full_fig_p033_18.png] view at source ↗

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

16 extracted references · 4 linked inside Pith

  1. [1]

    (2024, Proposition 1)

    and Carbone et al. (2024, Proposition 1). B Proofs We give some additional results on the Chi-square distance betweenπθk andπ θk−1 as a function of the step sizeγin this section. First we present a result with a result with local boundedness assumptions on the gradient and the Hessian. Recall that χ2(πθk , πθk−1 ) = Z x∈Rdx πθk (x) πθk−1 (x) −1 2 πθk−1 (x...

  2. [12]

    , K−1do w(i) k−1 = 1/N Computeg k−1(wk−1)as in (24) for forward-KL, or as in (25) for reverse-KL

    fork= 1, . . . , K−1do w(i) k−1 = 1/N Computeg k−1(wk−1)as in (24) for forward-KL, or as in (25) for reverse-KL. θk = OPT(θk−1, gk−1(wk−1)) fori= 1, . . . , Ndo X (i) k =X (i) k−1 −γ∇ x Eθk−1 (X (i) k−1) + √2γ σ ε(i) k−1,whereε (i) k−1 ∼ N(0, I). end for end for Returntuned parametersθ K, and optionally history of other quantities. E Experimental Details ...

  3. [13]

    Regarding initialisation, weights are drawn fromN(0, σ2 N ), withσ N = 0.02, and biases are initialised to zero

    after each hidden layer, defined byσ(z) =z /(1 + exp(−z)), to help ensure∇ xEθ(x)is differentiable everywhere, for the Langevin dynamics. Regarding initialisation, weights are drawn fromN(0, σ2 N ), withσ N = 0.02, and biases are initialised to zero. T raining Objective.We train the energy model by minimising the negative log-likelihood (NLL) of the data,...

  4. [15]

    Reward F unctions.Here, we consider indicator rewards, for half-planes, of the form Rleft =1 {x1<0}, R right =1 {x1>0}, R lower =1 {x2<0}, R upper =1 {x2>0},(36) since not only do these rewards highlight thatR(x)may be non-differentiable, but these rewards also admit a closed form for the unique maximiser of (35). A standard variational argument shows tha...

  5. [16]

    Although the second term is indeed zero here, weleavethisinthecovarianceformtomatchMarionetal.(2025)andtherespectivecodeimplementation

    Now, we note that ∇θEπθ [R(X)] =E πθ [R(X)∇ θ logπ θ(X)] =−E πθ [R(X)∇ θV(X, θ)] +Eπθ [R(X)]E πθ [∇θV(X, θ)].(46) where, in the second equality, we have substituted in (45). Although the second term is indeed zero here, weleavethisinthecovarianceformtomatchMarionetal.(2025)andtherespectivecodeimplementation. Now, sinceπ ref does not depend onθ, we have th...

  6. [32]

    Pre-training details.As was the case for in the setting of 2D datasets, pre-training of the EBM proceeds viaPCD, albeit with a sampler that differs from our tuning kernel

    To emphasise the dimensionality of the problem, recall inputs to this network are single-channel images,x∈R 28×28×1. Pre-training details.As was the case for in the setting of 2D datasets, pre-training of the EBM proceeds viaPCD, albeit with a sampler that differs from our tuning kernel. Since many details are the same, we refer to Appendix E.2.1 for the ...

  7. [2003]

    Handbook of convergence theorems for (stochastic) gradient methods.arXiv preprint arXiv:2301.11235,

    Guillaume Garrigos and Robert M Gower. Handbook of convergence theorems for (stochastic) gradient methods.arXiv preprint arXiv:2301.11235,

  8. [2008]

    doi: 10.1561/ 2200000001

    ISSN 1935-8237. doi: 10.1561/ 2200000001. URLhttps://doi.org/10.1561/2200000001. Sinan Yildirim, Sumeetpal S Singh, Tom Dean, and Lan Jiang. A monte carlo expectation maximisation algorithm for multiple target tracking. In2012 15th International Conference on Information Fusion, pages 2094–2101. IEEE,

  9. [2010]

    Uniform-in-time convergence bounds for persistent contrastive divergence algorithms.arXiv preprint arXiv:2510.01944,

    13 Paul Felix Valsecchi Oliva, O Deniz Akyildiz, and Andrew Duncan. Uniform-in-time convergence bounds for persistent contrastive divergence algorithms.arXiv preprint arXiv:2510.01944,

  10. [2012]

    14 Equivalently, the log-incremental weight is logG k(xk−1, xk) =−U θk (xk) +U θk−1 (xk−1) − 1 4h ∥xk−1 −x k +h∇U θk (xk)∥2 + 1 4h xk −x k−1 +h∇U θk−1 (xk−1) 2

    A Derivation of the weights for the ULA kernel When the forward kernel is chosen as theULAkernel with step-sizeh >0: Kk(xk−1, xk) = (4πh)−d/2 exp − 1 4h ∥xk −x k−1 +h∇U θk−1 (xk−1)∥2 , and Lk−1(xk, xk−1) = (4πh)−d/2 exp − 1 4h ∥xk−1 −x k +h∇U θk (xk)∥2 , the incremental weight in (8) becomes Gk(xk−1, xk) = Πθk (xk)L k−1(xk, xk−1) Πθk−1 (xk−1)K k(xk−1, xk)...

  11. [2015]

    Non-asymptotic analysis of biased stochastic approximation scheme

    Belhal Karimi, Blazej Miasojedow, Eric Moulines, and Hoi-To Wai. Non-asymptotic analysis of biased stochastic approximation scheme. InConference on Learning Theory, pages 1944–1974. PMLR,

  12. [2017]

    Christoph Schönle, Davide Carbone, Marylou Gabrié, Tony Lelièvre, and Gabriel Stoltz

    URL https://arxiv.org/abs/1710.05941. Christoph Schönle, Davide Carbone, Marylou Gabrié, Tony Lelièvre, and Gabriel Stoltz. Efficient monte-carlo sampling of metastable systems using non-local collective variable updates.arXiv preprint arXiv:2512.16812, 2025a. Christoph Schönle, Marylou Gabrié, Tony Lelièvre, and Gabriel Stoltz. Sampling metastable system...

  13. [2020]

    Gersende Fort and Eric Moulines

    URL https://arxiv.org/abs/1903.08689. Gersende Fort and Eric Moulines. Convergence of the monte carlo expectation maximization for curved exponential families.The Annals of Statistics, 31(4):1220–1259,

  14. [2021]

    Training restricted boltzmann machines using approximations to the likelihood gra- dient

    Tijmen Tieleman. Training restricted boltzmann machines using approximations to the likelihood gra- dient. InProceedings of the 25th International Conference on Machine Learning, ICML ’08, page 1064–1071, New York, NY, USA, 2008a. Association for Computing Machinery. ISBN 9781605582054. doi: 10.1145/1390156.1390290. URLhttps://doi.org/10.1145/1390156.1390...

  15. [2023]

    Implicit diffusion: Efficient optimization through stochastic sampling

    PierreMarion, AnnaKorba, PeterBartlett, MathieuBlondel, ValentinDeBortoli, ArnaudDoucet, Felipe Llinares-López, Courtney Paquette, and Quentin Berthet. Implicit diffusion: Efficient optimization through stochastic sampling. InInternational Conference on Artificial Intelligence and Statistics, pages 1999–2007. PMLR,

  16. [2025]

    Learning latent energy-based models via interacting particle langevin dynamics.arXiv preprint arXiv:2510.12311,

    Joanna Marks, Tim YJ Wang, and O Deniz Akyildiz. Learning latent energy-based models via interacting particle langevin dynamics.arXiv preprint arXiv:2510.12311,

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.