acceptodds
Withdrawn from ICLR 2027

Formal Safety Verification of Stochastic Systems under Diffusion-Induced Uncertainty

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Formal safety verification of autonomous systems in open environments is challenged by interaction-induced stochasticity. Existing methods typically assume pre-specified stochastic models or handcrafted disturbance distributions, limiting their ability to capture complex disturbance patterns encountered in real-world operation. In this paper, we leverage expressive diffusion models to learn data-driven stochastic behaviors from observed trajectories and adapts it to the evolving system state. By exploiting their intrinsic score-based reverse stochastic dynamics, we analytically propagate the learned uncertainty into an explicit verification-oriented stochastic differential equation (SDE), enabling rigorous probabilistic safety verification via stochastic barrier certificates. Experimental results on representative benchmarks demonstrate the effectiveness of our proposed approach.

Primary area:
applications to robotics, autonomy, planning
Keywords:
Formal Verification, Probabilistic Safety Verification, Stochastic Differential Equation, Barrier Certificate, Diffusion Model

Reviews

Among:
Shape:
1234this paper, Bottom 10%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.67 over 3 reviews, above 10% of the others.
Reviewer_enyWOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

This paper proposes DiffSBC, a framework for probabilistic safety verification under unknown stochastic uncertainty. It uses conditional diffusion models and reverse-polynomial moment propagation to estimate disturbance covariance, then constructs a stochastic differential equation for formal verification using neural stochastic barrier certificates and SOS optimization. Experiments on a 7-dim system demonstrate improved verification efficiency.

Critical strengths

The paper addresses an important problem with a well-structured framework connecting uncertainty learning and formal verification. The theoretical results justify the proposed moment approximation under stated assumptions and provide safety guarantees for the constructed stochastic model. Experiments demonstrate promising efficiency compared with existing baselines.

Critical weaknesses

The motivation for using diffusion models is unclear, since the final verification model only requires disturbance covariance. Conceptual and experimental comparisons with simpler covariance estimation methods under the same SOS framework will help establish the benefits of the proposed approach.

The safety guarantees rely on strong modeling assumptions and approximations, including polynomial fitting, Gaussian moment closure, and constant rank-one covariance. These assumptions are not sufficiently justified, and the guarantees apply only to the approximated stochastic model rather than necessarily to the original system.

The experiments are limited to synthetic polynomial systems with additive Gaussian noise. Moreover, although the data generator perturbs all state coordinates, the verification model retains uncertainty only in the last coordinate. This simplification limits the relevance of the results to general state-dependent and multidimensional stochastic systems.

Questions for authors

Could the authors clarify the advantages of diffusion-based covariance estimation over simpler alternatives?

Can the authors justify the rank-one approximation and explain when the resulting safety guarantees extend to the original stochastic system?

Reviewer_XYGyOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

The paper proposes DiffSBC, a pipeline for probabilistic safety verification of systems whose nominal polynomial drift is known and whose stochastic perturbation is learned from data. A conditional score-based diffusion model is trained on one-step transition residuals, conditioned on the current state and an encoding of a short history window. The first two moments of the learned residual distribution are then obtained by reverse-polynomial moment propagation (RPM). A polynomial stochastic barrier certificate (SBC) for this SDE is learned by a neural network with quadratic activation functions and certified by sum-of-squares (SOS) programming inside a counterexample-guided (CEGIS) loop. Experiments on nine polynomial benchmarks compare against an additive-disturbance baseline and a discrete-time Gaussian c-SBC baseline.

Critical strengths

S1) The SOS solution step is standard and seems correctly applied.

S2) The appendix is detailed: hyperparameters, solver tolerances, barrier plots, and code is provided.

Critical weaknesses

W1) The certified guarantee concerns a hand-constructed proxy SDE, not the system that generated the data. This is my main concern, and it is only visible in the appendix.

  • The benchmark data are generated with additive Gaussian noise on every coordinate (Eq. (125)). The SDE that is actually verified has noise in the last coordinate only (Eq. (127)), which explains the zeros in \overline{G}_{\mathrm{diff}} in Example 1. The certified system is therefore NOT the data-generating system, and the certificate is not shown to apply to it.
  • It remains unclear whether the authors' method actually accounts for diffusion-model estimation and the sampling errors.
  • The paper itself states that "the reported probability bounds certify the fixed stochastic proxy used by each verifier" and promises that the statistical relationship between this proxy and the unknown disturbance "is discussed separately at the end of this section" (p. 27). I could not find this discussion anywhere in the paper.

Consequently, the "certified infinite-horizon safety lower bound" is a formal statement about an SDE that the authors construct, not about the unknown system. Calling this "formal safety verification" in the title and abstract, without stating any of the above in the main text, substantially overclaims what the method delivers.

W2) It is unclear whether the authors' method actually provides continuous-time or discrete-time guarantees. The problem is initially posed for the Ito SDE (1), but Assumption 1 replaces it by a discrete-time model x_{t+1} = \Psi_{\Delta t}(x_t) + \delta_t with a residual whose conditional law depends on a trajectory history \mathcal{H}t. Everything that is learned is a one-step residual distribution. The continuous-time model is then recovered by the heuristic Q\theta = \Sigma_{\Delta x}/\Delta t (Eq. (105), Proposition 3). Proposition 3 only states that a positive semidefinite matrix has a square root; it does not establish any relation between the safety probability of the resulting SDE and that of the discrete-time process the data come from.

Several points remain unclear:

  • How is \Psi_{\Delta t} obtained from f?. For data from a genuine continuous-time system, the residual would also contain discretization error of order \Delta t^2, which this construction cannot separate from the noise.
  • Assumption 1 allows the residual to depend on the history (i.e., a non-Markovian perturbation), yet the verified SDE (17) is Markovian with a constant diffusion coefficient.
  • Assumption 1 calls \mathcal{H}_t an "observable trajectory history", but it consists of full states, i.e., full state observability is assumed.

The paper should state precisely which guarantee holds for which object (e.g., discrete-time process or SDE), and in continuous time or not.

W3) The experiments do not support the claimed advantage of the diffusion model.

  • The uncertainty in all benchmarks is additive, state-independent, and Gaussian (Eq. (125)). None of the "complex", "interaction-induced", or state-dependent disturbance patterns that motivate the paper (abstract and Section 1) is ever presented. The benchmarks are polynomial ODEs perturbed by additive Gaussian noise of constant amplitude, for which a diffusion model is unnecessary.
  • The key ablation is missing: the same continuous-time SOS pipeline with \lambda_G (or with the true rate) instead of \lambda_{\mathrm{diff}}. Given Table 7, I expect this to perform at least as well as DiffSBC, which would mean that the diffusion component contributes nothing to the results.
  • The state-dependent structure is discarded before verification. The covariance is collapsed to a single constant, and the residuals are symmetric as {r, -r} (Table 4), which forces a zero mean. The statement in Section 4.2 that "preserving the state-dependent stochastic structure learned by the diffusion model can lead to more informative certificates" is therefore not supported: no state-dependent structure is preserved in the verified model. If the only difference from the baselines is that the learned uncertainty is not reduced to an averaged additive disturbance, this should be stated plainly.
  • The baselines are weak. "Additive+Neural" (called "Gauss.+Neural" in Appendix A.8) verifies a deterministic system, i.e., it is not a stochastic verification method at all. Table 5 also lists "Baseline CEGIS rounds: 1"; please clarify whether the baselines were given the same refinement budget as DiffSBC. There is no comparison with methods that make weaker assumptions (e.g., unknown drift, non-additive noise, or partial observability), nor with abstraction-based methods.

W4) The assumptions are substantially stronger than the introduction and abstract suggest. The abstract and introduction present the method as learning data-driven stochastic behaviors "from observed trajectories" and adapting to "complex disturbance patterns encountered in real-world operation". In fact the method requires: (i) known polynomial nominal dynamics, stated only in a side sentence before Assumption 1 and in the "Require" line of Algorithm 1; (ii) full state observability; (iii) compact semi-algebraic state, initial, and unsafe sets; (iv) polynomial barrier functions of low degree (2 to 6); and (v) in all experiments, rank-one, constant, additive noise. Each of these is a substantial restriction, and together they should be stated up front. The introduction further claims that the prior data-driven methods it cites were developed "to relax these assumptions", but Mathiesen et al. (2026) assume additive noise and Salamati et al. (2024) assume Lipschitz continuity, so the contrast drawn with them is not accurate.

W5) Novelty and relevance for ICLR are low. Theorem 1 restates the standard stochastic barrier certificate result of Prajna et al. (2007) (going all the way back to Kushner), Theorem 2 is Putinar's Positivstellensatz, and Theorem 4 seems to be a classical result for diffusions. Proposition 3 is the existence of a matrix square root. The remaining technical contribution is RPM, a quadratic local fit of a noise predictor plus a Gaussian moment closure, used to estimate the variance of a one-dimensional residual. For a one-dimensional variance, sampling from the diffusion model and computing the empirical variance would be simpler; the claim (Section A.2.4) that Monte Carlo estimation is impractical is not substantiated. It is also not clear whether the first contribution (a conditional diffusion model of residuals given known nominal dynamics) is novel.

More generally, many recent papers have already combined machine-learning tools with barrier certificates. Like much of that research line, this paper does not demonstrate relevance to real-world problems: all benchmarks are low-dimensional academic polynomial systems. SOS-based barrier certificates are inherently limited to such systems (the paper itself reports resource-limit rejections already at degree 4 for n = 7), and barrier certificates are known to be conservative in general, i.e., they cannot in general provide tight safety bounds (Laurenti & Lahijanian, 2025). The authors cite this work but do not discuss its implications for their method. Without a demonstration on a system of practical relevance, I do not see how the presented results are suitable for ICLR. The paper may be better suited to a formal-methods or control venue after the issues above are addressed, in particular after the gap between the verified proxy and the actual system is closed and a benchmark with genuinely non-additive, state-dependent noise is added.

Questions for authors
  • What type of guarantees (ct/dt; target system) does the method actually establish? Provide a full list of assumptions under which these guarantees hold.
  • Which of the presented theoretical results are novel?
  • How is \Psi_{\Delta t} obtained from f?
  • Why do the authors not compare against statistical baseline methods that apply to broader classes of systems?
Reviewer_wgBmOpenReview ↗
Rating
2
Confidence
4
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

The paper considers the problem of data-driven verification of stochastic differential equations (SDEs). The approach proposed by the authors relies on stochastic barrier functions and combines them with Gaussian moment closure and sum-of-squares optimisation. The performance of the approach is empirically compared with two alternative works.

Critical strengths
  • The paper considers an important problem, and the solution is original

  • The experiments report improvements compared to the state-of-the-art

Critical weaknesses
  • The paper combines several established techniques, including stochastic barrier certificates, SDE discretization, and Gaussian moment closure. However, the distinction between original contributions and adaptations of existing methods is not always clear. For example, Theorem 1 restates and proves a standard stochastic barrier-certificate result and should be explicitly identified and attributed as such. The authors should clarify the main technical contributions beyond the integration of these existing components.

  • The paper does not establish end-to-end safety guarantees for the original system from which the data are collected. This weakness becomes important expecially because of the weaknesses in the experimental analysis discussed below.

  • The considered 7D benchmark has noise in only 1 of the 7 dimensions. Furthermore, the considered unsafe set seems to be quite far from the trajectories of the system. Because of the constraints on the dynamics, it is not even clear to me whether the unsafe set can be reached with nonzero probability. I believe more challenging benchmarks and settings are needed to showcase the effectiveness of the approach, also considering its lack of formal guarantees.

  • The literature review in the paper omits to explicitly discuss works using Gaussian process regression to compute safety guarantees for unknown systems, e.g., [Paul Griffioen et al. "Probabilistic invariance for gaussian process state space models." Learning for Dynamics and Control Conference, 2023] [John Skovbekk et al. "Formal verification of unknown dynamical systems via gaussian process regression." IEEE Transactions on Automatic Control 70.8 (2025): 4960-4975.]. Pros and Cons with respect to these approaches should be discussed.

Questions for authors
  • Can you clarify what the main original contributions are?

  • Can you clarify my doubts about the experiments and consider a more challenging setting?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.