- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
The paper studies Schrödinger bridges with running and terminal costs that depend on the evolving marginal distribution. It expresses the optimality conditions through coupled Schrödinger potentials, related by a Cole–Hopf transformation to a Hamilton–Jacobi–Fokker–Planck system. Its main computational proposal is a two-type branching diffusion representation: offspring of the opposite type restart at the reflected time , allowing the coupled potentials to be evaluated by weighted Monte Carlo trees. Neural terminal potentials are trained jointly by matching the two boundary conditions with an MMD objective, and samples are generated using Monte Carlo estimates of the logarithmic gradient of a potential.
Validation includes classical bridge recovery, soft terminal constraints, a two-dimensional density-dependent penalty, obstacle avoidance, and latent transport on MNIST and FFHQ. The paper reports competitive classical-bridge results and a favorable FFHQ FID. However, the large-dimensional experiments shown in the main text use , and several central theoretical and objective-level claims require correction or additional justification.
Critical strengths
-
A meaningful computational approach to coupled nonlinear potentials. The combination of two particle types, reflected offspring times, and joint MMD fitting is an interesting way to approach a coupled forward–backward problem without a spatial mesh. The paper explicitly acknowledges its connection to existing mean-field formulations in Remark 2.3. The novelty should therefore be assessed primarily as a computational construction, rather than as the discovery of law-dependent Schrödinger bridges themselves. Prior work such as Deep Generalized Schrödinger Bridge already connects bridges with mean-field structure, but that does not eliminate the potential value of the different branching-based solver proposed here.
-
Substantial initial validation with relevant baselines and some appropriately bounded claims. Figure 3 compares against DeepGSB, LightSB, LightSBB-M, DSBM, and BranchSBM, rather than relying only on parameter sweeps. The reported mixture Wasserstein distance is competitive, and Figure 7 reports an FFHQ FID of 36.03, compared with 38.92 for DSBM and 39.54 for BranchSBM-M. Appendix K also provides a useful analytical sanity check for a time-reflected nonlinear equation. These experiments support the claim that the implementation can produce useful transports in the tested settings. They do not establish uniform superiority, which the discussion correctly refrains from claiming.
Critical weaknesses
- The central branching and sampling guarantees are not established as stated.
Proposition 3.1 assumes bounded terminal potentials and bounded polynomial coefficients, then claims an unbiased, square-integrable representation. Appendix L argues that its absolute-moment majorant is globally bounded because the product of two reflected potentials is constant. Constancy does not establish that this product can satisfy both endpoint conditions.
A concrete diagnostic is the spatially homogeneous case , , and . These functions meet the proposition's stated boundedness assumptions. In the canonical decomposition of Equation (10), and . Appendix L's absolute-coefficient majorant consequently has coefficient function . Solving its differential equations with the two terminal values forces its positive constant product to satisfy , which has no positive solution. In contrast, the original signed equations admit a positive homogeneous solution with . This example concerns the proposition's stated function class, not a normalized probability bridge; it directly invalidates the claimed automatic existence of the majorant. The familiar need for moment and non-explosion conditions in branching diffusion representations makes it consequential to specify and prove the actual admissible regime here.
Separately, Section 4.2 and Algorithm 2 identify as an unbiased estimate of , where . In general, , even if the potential and its gradient are separately unbiased. This is already visible in a bounded, strictly positive linear heat example: with , , , and , the true score is , whereas the expectation of the estimator is approximately 0.324611. Increasing can reduce bias; it does not establish finite-sample unbiasedness. Signed branching weights additionally require a justification that the empirical potential remains positive before its logarithm is taken. The explanation on page 8 attributing the absence of variance collapse to an unbiased score estimator is therefore unsupported by the stated argument. These issues concern core guarantees rather than optional future variance reduction.
- The intended objective and the evidence for the nonlinear contribution remain insufficiently connected.
Remark 2.4 calls the terminal penalty reverse KL, , but its exponential conjugate and power-law boundary condition do not follow from that probability-domain penalty. For the standard forward KL on probability measures, the conjugate is , rather than the exponential integral displayed in Equation (6). A suitable unnormalized entropy functional has a different domain and normalization convention. Those distinctions matter because the process law is a probability distribution and the soft terminal condition is used in training.
There is also a direct inconsistency in Equation (15): the second squared-measure term uses , as required by Equations (13) and (27), but its corresponding cross term uses . This may be a typographical error, but the paper does not establish which expression was implemented. Algorithm 1 further sums all minibatch pairs, including the diagonal, while reusing one-tree estimates. On a diagonal, squaring an unbiased tree estimate does not give an unbiased estimate of the squared potential; it introduces a parameter-dependent second-moment contribution. Thus, the expected minibatch objective also needs to be distinguished from the intended population MMD.
Empirically, the high-dimensional mixture, MNIST, and FFHQ experiments in the main text all use . They validate the classical specialization, not scalability of genuinely density-dependent branching. The experiment is two-dimensional and principally shows a penalty trajectory; the obstacle example uses and does not test dependence on the evolving density. A small set of quantitative nonlinear comparisons, together with endpoint error and tree-cost measurements, would substantially strengthen the central contribution. I am not requiring an exhaustive benchmark suite. The required distinction is between demonstrating an interesting nonlinear mechanism and establishing that it reliably solves the intended nonlinear bridge. Appendix O may contain additional relevant protocols or evidence, which could not be assessed from this copy.
Questions for authors
-
What additional conditions ensure existence, tree termination, and finite second moments in Proposition 3.1? In particular, how does Appendix L accommodate , , and constant terminal potentials equal to one, for which its absolute-moment majorant requires ? Please provide a corrected argument or restricted theorem, and identify which experimental configurations satisfy its conditions. This could materially change my soundness assessment.
-
Is Algorithm 2 actually intended to provide an unbiased score, or a biased but consistent ratio estimator? Please justify the claim for the implemented estimator or revise it and characterize the finite- bias. How are nonpositive empirical potentials caused by signed weights handled? Any clipping, tree truncation, or alternative coefficient decomposition should be stated, since it can change both the estimator and its guarantees.
-
Which potential multiplies the black-tree estimator in the implemented cross term: as required by Equations (13) and (27), or as printed in Equation (15)? Are diagonal minibatch products evaluated with independent trees, removed as in a U-statistic, or retained with their bias? Please state the objective whose gradient is actually estimated and whether correcting these discrepancies changes the reported results.
-
Is the intended penalty reverse KL, forward KL on probability measures, or an unnormalized entropy functional on finite measures? Please reconcile its definition, the conjugate in Equation (6), the boundary conditions in Equations (7)–(8), and mass normalization. Does the corrected formulation change the soft-constraint experiment or the interpretation of its reported terminal KL?
-
What quantitative results establish successful optimization with genuinely density-dependent , beyond the two-dimensional penalty plots? A representative comparison against an appropriate nonlinear solver, reporting the nonlinear objective, endpoint mismatch, and matched computational cost, would be particularly informative. Please also identify any supporting results in Appendix O and report how tree counts, runtime, and stability depend on , , , and dimension. This requests evidence for the central claim, rather than an exhaustive set of future applications.
2 comments
Thank you very much for the detailed response and clarifications. I have noted the chairs’ reminder regarding AI use. While the corrections are helpful, some nonlinear experiments remain outside the revised theoretical guarantees, and corrected results for the signed-weight objective are still pending. These limitations leave my main concerns only partially resolved. I therefore maintain my original rating of weak rejection.
Corrected branching conditions, consistent drift estimator, clarified objective
We thank the reviewer for the careful reading and the explicit counterexamples.
Q1. The reviewer is right: in App. L, the constancy of U(t)Uˉ(T−t)=r does not ensure that the majorant meets both endpoint conditions. For the second-moment majorant these force r=Ψ(r):=A2B2exp(cT(Q(r)−1)), with A=∥φT∥∞, B=∥φˉT∥∞, Q(r)=∑k∥ak∥∞2rk/pk, and a bounded solution exists iff this equation has a root. The corrected Prop. 3.1 assumes (i) cTKˉ<1, Kˉ=∑kkpk, and (ii) Ψ(r)≤r for some r>0. Under (i), the expected tree size is exactly 1+c(T−t)(1+2Kˉ)/(1−cTKˉ), so trees are a.s. finite; under (ii), supt,xE[Φ2]<∞ and E[Φ] is the unique mild solution in the dominated class. Your example: for F=κm and φT=φˉT≡1, the signed system has the solution e−cκm(T−t), m=W0(cκT)/(cκT), while r=ecκTr has a root iff ecκT≤1. As ∣Φ∣ equals the majorant functional for constant data, E∣Φ∣=∞ when ecκT>1, so (ii) cannot be dropped although the PDE solution exists. Experiments: for F∈{0,V}, K=0 and (i)-(ii) hold for every T (all non-Gross-Pitaevskii runs). For F=κm (T=2, p0=p1=21), (i) holds as cT≤0.6, and (ii) reads 2ecTκ2A2B2ecT≤1. At a solution, AB≥supμ0≥(2πe)−1 for whitened 2D data, so (ii) fails for (c,κ)=(0.3,10) and (0.3,100), which we now present as qualitative only; for κ=1 it reduces to an explicit bound on AB.
Q2. It is a consistent ratio estimator, not an unbiased one. Under smooth data and a strict version of (ii), φ and ∇φ are unbiased, and sM=∇φ/φ is consistent and asymptotically normal. For positive weights with ∣∇Φ∣≤LΦ, E[sM]=s+b1/M+O(M−2), b1=(sVarΦ−Cov(∇Φ,Φ))/φ2. Your heat example is such a case, with b1=s(1−e−2)/8>0. With signed weights we only claim consistency and the CLT. The exact potential is positive (Feynman-Kac), but the empirical average may not be; Algorithm 2 then sets the drift to zero. No clipping, tree truncation or alternative decomposition is used. The sentence on p. 8 is removed.
Q3. The cross term should contain A(Z,∙)φθ(Z), as in Eqs. (13) and (27); φˉη in Eq. (15) is a typo, and the code uses φθ (
WT = B_Z * phi_Z / mu_Zinsbnl_toy.py). The two factors of each quadratic term use independent reference points and trees, so no squared one-tree estimate appears; Algorithm 1 now writes the loss as a U-statistic over b=b′, unbiased for the MMD of Prop. 4.1, as is its gradient. Our runs, however, normalise the weights by ∑b∣Wb∣. With positive weights (F∈{0,V}), the zeros of this objective define the same bridge, so those results are unchanged. With the signed weights of the Gross-Pitaevskii runs, it is strictly positive, so these runs minimise a surrogate; we are re-running them with the unnormalised loss.Q4. The penalty is the forward KL: with f(u)=−logu−1, Df(μT,m)=∫m(logμTm−1)=KL(m∥μT)−1 on probability densities, so "reverse" in Remark 2.4 is wrong. Eq. (6) is the conjugate over nonnegative densities; over probability densities it is λlogEμT[ev/λ]+λ. Both agree at the optimum, and the mass constraint only adds a multiplier, absorbed by the gauge (φ,φˉ)↦(sφ,φˉ/s). Eqs. (7)-(8) and the soft-constraint experiment are unchanged; Fig. 4b reports KL(Law(XT)∥μT).
Q5. The paper does show the effect of non-classical running costs in dimension 2, on obstacle (F=V, Fig. 5, App. O.3) and congestion (F=κm, Fig. 4c) problems of the type studied by DeepGSB (Liu et al., 2022) and GSBM (Liu et al., 2024); for F=κm, larger c lowers the concentration at intermediate times. In high dimension, we chose to compare with SOTA SB methods (DSBM, LightSB, LightSBB-M, DeepGSB, BranchSBM) on their benchmarks, which are classical bridges, hence F=0 for MNIST and FFHQ. The 50D run (App. O.4) does use F=κm, c=0.1; the label F=0 in Sec. 5 is a typo. Whitened densities being small in d=50, the coupling is weak there, and this run mainly shows that training with branching trees is stable. The expected tree size (Q1) does not depend on d or κ, and a tree costs O(d). A quantitative 2D comparison against a grid solver (Guéant, 2012), DeepGSB and the explicit case ecκT≤1 of Q1 would further strengthen this evidence; Sec. 6 now mentions it.