- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper lifts partial physical relations in the observation space into constraints on the latent vector field through the decoder Jacobian, and enforces them at each vector-field evaluation using a weighted projection. The method combines continuous-time variational inference over neural network weights with inference over physical parameters, and analyzes the effect of projection on local uncertainty. Experiments cover six controlled dynamical systems under observation corruption, partial observability, and physical misspecification.
Critical strengths
-
The problem is clearly defined and practically relevant. The core method is clearly formulated, and the theoretical results provide useful interpretations of its behavior.
-
The proposed method achieves the lowest mean MSE on five of the six systems. The experiments further cover various types of data corruption and physical misspecification, providing reasonably broad initial evidence of its practical value.
Critical weaknesses
-
Limited novelty and insufficient validation of the robustness mechanism. The core projection formula (Eq. 7) is mathematically equivalent to the general weighted projection for linear equality constraints in ProbHardE2E (Utkarsh et al., End-to-End Probabilistic Framework for Learning with Hard Constraints, ICLR 2026, Appendix C.1, Eq. 15). Specifically, setting
z = f_theta,A = C J_phi,b = g, andQ = J_phi^T J_phi + lambda Iin the general projection formula recovers Eq. 7. Therefore, the overall method appears largely to integrate existing techniques. Furthermore, although the experiments cover six dynamical systems, five of them impose essentially the same position–velocity kinematic relation. Thus, the diversity of embedded physical constraints is rather limited, and it remains unclear whether the proposed method generalizes effectively to more complex forms of partial physical knowledge. -
Discrepancies between the paper and the supplementary code. The provided code appears inconsistent with the manuscript in several important respects. First, a nonlinear
tanhbounding operation is applied after projection, which can violate the claimed exact satisfaction of physical constraints. Second, the saved prediction trajectories are generated from a training trajectory rather than held-out test trajectories, although the script separately computes test losses. Third, the claimed Bayesian posterior over physical parameters does not appear to be implemented: physical parameters are represented as deterministic arrays, and the code does not appear to include the variational posteriorq_eta(Lambda)or its corresponding KL term in the ELBO. These discrepancies raise concerns about the reproducibility of the reported results. The authors should clarify whether they arise from a code version mismatch or from important implementation details not described in the paper.
Questions for authors
-
Could the authors clarify the discrepancies between the supplementary code and the manuscript? Was a different version of the code used to produce the reported results?
-
Could the authors provide experiments involving more diverse physical constraints beyond simple position–velocity relations to better demonstrate the generality of the proposed framework?
3 comments
Response to reviewer dk8k
We embed 12 separate constraint sets as combinations of the kinematics and dynamic rows, and misspecified variants of each. The dynamic rows go beyond the position–velocity relation. The table reports MSE and CRPS on the held out test set
| Embedded relations | Latent projection (ours) | State-space projection |
|---|---|---|
| 1 kinematic | 0.031 ± 0.0052 / 0.74 ± 0.012 | 1.1 ± 0.44 / 5.5 ± 1.7 |
| 2 kinematic | 0.03 ± 0.0053 / 0.73 ± 0.018 | 1 ± 0.39 / 3.8 ± 1.5 |
| 3 kinematic | 0.027 ± 0.0024 / 0.67 ± 0.035 | 0.16 ± 0.21 / 1.9 ± 0.9 |
| 1 dynamic | 0.017 ± 0.0015 / 0.57 ± 0.017 | 0.81 ± 0.32 / 4.2 ± 0.61 |
| 2 dynamic | 0.017 ± 0.0026 / 0.52 ± 0.049 | 0.62 ± 0.46 / 3.2 ± 1.3 |
| 3 dynamic | 0.015 ± 0.0011 / 0.46 ± 0.022 | 0.49 ± 0.0089 / 3 ± 0.13 |
| 3 kinematic + 1 dynamic | 0.014 ± 0.0017 / 0.53 ± 0.012 | 0.29 ± 0.34 / 2 ± 1.4 |
| 3 kinematic + 2 dynamic | 0.011 ± 0.0014 / 0.45 ± 0.013 | 0.032 ± 0.013 / 0.65 ± 0.12 |
| Misspecified kinematic (3 wrong) | 0.031 ± 0.0056 / 0.75 ± 0.016 | all seeds diverged |
| Misspecified kinematic (1 of 3 wrong) | 0.032 ± 0.0073 / 0.74 ± 0.013 | all seeds diverged |
| Misspecified kinematic (2 of 3 wrong) | 0.031 ± 0.0052 / 0.74 ± 0.014 | all seeds diverged |
| Misspecified dynamic | 0.014 ± 0.0044 / 0.45 ± 0.028 | 0.49 ± 0.054 / 3 ± 0.069 |
| Embedded relations (observation WIS, normalised) | Latent projection (ours) | State-space projection |
|---|---|---|
| 1 kinematic | 0.51 ± 0.0046 | 2.9 ± 0.91 |
| 2 kinematic | 0.51 ± 0.0097 | 2.4 ± 0.81 |
| 3 kinematic | 0.49 ± 0.022 | 1.8 ± 0.64 |
| 1 dynamic | 0.43 ± 0.013 | 2.3 ± 0.25 |
| 2 dynamic | 0.42 ± 0.047 | 2.1 ± 0.41 |
| 3 dynamic | 0.39 ± 0.02 | 1.9 ± 0.13 |
| 3 kinematic + 1 dynamic | 0.4 ± 0.012 | 1.7 ± 0.26 |
| 3 kinematic + 2 dynamic | 0.37 ± 0.031 | 1.5 ± 0.15 |
| Misspecified kinematic (3 wrong) | 0.51 ± 0.002 | all seeds diverged |
| Misspecified kinematic (1 of 3 wrong) | 0.52 ± 0.0037 | all seeds diverged |
| Misspecified kinematic (2 of 3 wrong) | 0.51 ± 0.0063 | all seeds diverged |
| Misspecified dynamic | 0.38 ± 0.0045 | 1.9 ± 0.16 |
2. Supplementary code versus the manuscript (Weakness 2, Question 1)
The supplementary code was a standalone extract of a larger codebase, which we reduced to a minimal working repository with an LLM coding agent. All results were generated from a larger respository which we will release subject to publication.
-
The tanh bound was applied as a numerical safeguard, and only acts when a drift increases too much in magnitude. We have verified that held-out trajectories the bound has a neglible impact on the constraint residuals.
-
The extract's plotting function use training trajectory 0. Both the plots in Figure 1 and the rest of the paper, and all metrics are computed on held-out test trajectories.
-
The posterior over is implemented in the full codebase. It is used when we make the parameters learnable (i.e. across the various misspecification regimes).
Thank you for the detailed response and additional experiments. The new experiments with more diverse physical constraints have addressed my concern about the generalisability. However, I still have concerns about the novelty and the reproducibility. Therefore, I will keep my original rating for now.
Rebuttal to reviewer dk8k
1. Novelty relative to ProbHardE2E, and diversity of the embedded constraints (Weakness 1, Question 2)
The reviewer has noted our projection is mathematically identical to one in ProbHardE2E - this is because both projections are derived as KKT solutions to a minimum distance optimization problem. Our related work cites projection based approaches as a popular method for embedding constraints, which encompasses ProbHardE2E (we note we did not cite this). However, the contexts used between them are completely different - the ProbHardE2E utilises this as an output conditioning layer on a probabilistic forecaster, that projects an output location scale distribution onto a constraint manifold, using a covariance weighting.
In our work, we utilize a projection inside a latent space, with our weighting instead being the pullback through a learned decoder. Our approach acts on a vector field at every solver evaluation, that allows us to constrain dynamics along a full trajectory, not outputs at one time. Both are results of projections derived from differentiable optimization layers, we are happy to reference ProbHardE2E as an alternative projection method in related work.
We appreciate the reviewers concerns regarding generalisability. We have added two experiments that address this directly: a new case study with physics that are not kinematic structures, and an ablation over the systems with multiple relations, number of embedded and type of physics.
Dynamic Chemical Reactor Firstly, we have added a non-isothermal CSTR with series, parallel and side reactions, comprising 6 states and 3 inputs with Arrhenius kinetics. The embedded knowledge is a set of species mole balances. We test it under full observation and under partial observation, where two species are unable to be accessed. Scoring the MSE / CRPS and WIS on the held out test set:
Secondly, on the systems that admit several relations (the reactor and the chained pendulum), we sweep the number and type of embedded relations. This shows how the method behaves as more physics is added, and when that physics is wrong.
For the reactor: we embed one or two species balances. Our latent approach under full observation improved as more balances are added and this remains ahead of state-space projection over the vast majority of settings.
This remains the case for the high pendulum, the latent projection is ahead. On the pendulum it also stays stable under misspecified relations, where state-space projection diverges on all 9 runs.