Pith. sign in

REVIEW 3 major objections 5 minor 73 references

Provable Maximum Entropy Manifold Exploration via Diffusion Models

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Exploration of a learned design space can be posed as entropy maximization over a diffusion model's implicit manifold; sequential fine-tuning with the model's own score provably converges to the maximum-entropy distribution.

desk verdict A genuinely useful exploration idea built on the score-as-entropy-gradient identity, but the convergence theorem is shakier than the abstract suggests and needs a corrected statement. read the letter →

arxiv 2506.15385 v1 pith:ASAEAEGZ submitted 2025-06-18 cs.LG

Riccardo De Santi , Marin Vlastelica , Ya-Ping Hsieh , Zebang Shen , Niao He , Andreas Krause This is my paper · ORCID
classification cs.LG
keywords maximumentropyexplorationdiffusionmodelsscorefunctionmirrordescentsequentialfine-tuningmanifoldhypothesisstochasticoptimalcontrolmaximization
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

The paper attempts to turn exploration—generating genuinely novel designs rather than near-copies of the training data—into a well-defined optimization problem. Its claim is that a pre-trained diffusion model can be fine-tuned so that its samples spread out uniformly over the design manifold the model has implicitly learned, and that this is exactly the solution of a maximum-entropy problem over that manifold. The key move is an identity: $\nabla_x\delta H(p^\pi_T) = -\nabla_x\log p^\pi_T$, so the gradient of the entropy's first variation is the negative score function, a quantity the diffusion model already predicts, and the algorithm never has to estimate densities. The paper proves that with an exact score and an exact optimization oracle a single fine-tuning step is optimal, and that under realistic noise and bias conditions the sequential S-MEME algorithm converges weakly almost surely to the maximum-entropy distribution. This matters because it offers a principled, scalable route to novel valid designs in domains such as molecule and material discovery, where mimicking existing data is not enough.

What carries the argument

The argument is carried by three linked objects. First, the score-entropy identity $\nabla_x\delta H(p^\pi_T) = -\nabla_x\log p^\pi_T \simeq -s^\pi(\cdot,T)$: it converts the first variation of entropy, which naively requires the log-density, into the score function that the diffusion model is trained to output, removing density estimation from the loop. Second, the mirror-descent reading of fine-tuning, in which $D_{KL}(p^\pi_T, p^{k-1}_T)$ is the Bregman divergence of the entropic mirror map $Q = -H$, and the objective is 1-smooth and 1-strongly convex relative to itself, which makes an idealized single step optimal and grounds the iterative scheme. Third, the mirror-flow analysis: the dual variables $h_k = \delta H(p^k_T) = -\log p^k_T$ interpolate to an asymptotic pseudotrajectory of $\dot{h}_t = \delta H(p_t)$, $p_t = \delta(-H)^*(h_t)$, and the entropy $H$ acts as a strict Lyapunov function for this flow, so the unique internally chain-transitive set is the uniform distribution on $\Omega_{pre}$, to which the iterates converge almost surely.

What would settle it

Pre-train a diffusion model on samples lying exactly on a known lower-dimensional set, e.g. a circle in $\mathbb{R}^2$ or a 2D surface in $\mathbb{R}^{10}$, and run S-MEME. If the framework is sound, the fine-tuned model's samples should converge weakly to the uniform distribution on that set; if differential entropy is undefined on zero-measure supports, the iterates should fail to converge or spread unevenly along the intrinsic geometry. A complementary check is to measure, on the paper's own 2D example, the oracle bias $b_k = \mathbb{E}[\delta H(p^{\pi_k}_T) - \delta H(p^k_\sharp)\mid\mathcal{G}_k]$ as the linear solver's inner iterations grow, since Assumption 7.3 requires this bias to decay to zero.

Watch

Extended reading notes

Core claim

The paper's central claim is that maximum-entropy exploration over the support of a pre-trained diffusion model can be solved as mirror descent over probability measures, with the KL divergence as the Bregman distance and the model's own score as the gradient direction. Linearizing the entropy functional $H$ at the current iterate gives $\delta H(p^\pi_T) = -\log p^\pi_T$, and since $\nabla_x\delta H(p^\pi_T) = -\nabla_x\log p^\pi_T \simeq -s^\pi(\cdot,T)$, the reward gradient for fine-tuning is the negative score the diffusion model already computes. Each step of the proposed S-MEME algorithm (Score-based Maximum Entropy Manifold Exploration) solves $\arg\max_\pi \langle\delta H(p^{k-1}_T), p^\pi_T\rangle - \alpha_k D_{KL}(p^\pi_T, p^{k-1}_T)$, which is a mirror descent step with entropic mirror map $Q = -H$; because $F = -H$ is 1-smooth and 1-strongly convex relative to $Q$, a single such step is optimal when scores and optimization are exact. In the realistic case the discrete iterates form an asymptotic pseudotrajectory of the continuous-time mirror flow $\dot{h}_t = \delta H(p_t)$, $p_t = \delta(-H)^*(h_t)$, whose only internally chain-transitive set is the uniform measure on $\Omega_{pre}$. Theorem 7.1 concludes that the induced marginals converge weakly almost surely to $p^*_T \in \arg\max_{p_T\in\mathcal{P}(\Omega_{pre})} H(p_T)$, the maximum-entropy distribution on the pre-trained manifold.

Load-bearing premise

The entire problem is only well-defined if every distribution $p^{\pi_k}_T$ has an ordinary density with respect to Lebesgue measure on $\mathbb{R}^d$, because differential entropy and the KL regularization in the objective are defined only for such densities; if the learned manifold is truly lower-dimensional, the maximum-entropy problem and its convergence guarantee rest on a reference measure that the paper never supplies.

Editorial extensions

If this is right

  • Exploration needs no uncertainty quantification and no density estimator: the already-trained score network supplies the reward gradient, which is what makes the method scalable to high-dimensional text-to-image models.
  • With an exact score and an exact linear solver, a single fine-tuning step already produces the maximum-entropy model (Theorem 5.2); the sequential loop exists to correct the bias and noise of realistic solvers.
  • The regularization coefficient $\alpha$ trades validity against exploration: large $\alpha$ stays near the pre-trained model, small $\alpha$ reaches low-density regions, enabling guided data augmentation when a validity checker is available.
  • When the noise-and-bias conditions of Assumption 7.3 hold with the Robbins-Monro step rule, convergence is almost sure and in the weak topology, at a rate of order $\tilde{O}((\log\log k)^{-1})$.

Reading between the lines

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

  • Beyond Shannon entropy, the same sequential fine-tuning loop should work for any distributional objective whose first variation is available through the score—for instance Rényi entropies or submodular exploration measures—since only the first variation enters the linearized step.
  • In the strictly low-dimensional regime, the natural reformulation would maximize entropy with respect to the manifold's intrinsic volume rather than Lebesgue measure; a testable consequence is that naive S-MEME would spread samples unevenly along a curve or surface.
  • A practical diagnostic suggested by the analysis is the gap between the linearized iterate and the exact mirror descent step; if that gap does not shrink when the inner solver is run longer, the solver's bias, not the score error, is the binding constraint.
  • The image results hint at a general-purpose recipe—moderate $\alpha$, a few outer iterations, and an external validity checker—for turning any pre-trained generative model into a data-augmentation engine for design domains beyond images.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. This manuscript proposes a framework for exploration with pre-trained diffusion models. The maximum-entropy manifold exploration problem (Problem (7)) seeks a policy whose terminal marginal density p^π_T maximizes differential entropy H(μ) = −∫ dμ log(dμ/dx) over the approximate data manifold Ω_pre := supp(p^pre_T) induced by a pre-trained model. The key identity (12), ∇_x δH(p^π_T) = −∇_x log p^π_T ≃ −s^π(·, T), turns entropy-based exploration into score-based fine-tuning, avoiding explicit density estimation. The authors introduce S-MEME (Algorithm 1), a mirror-descent scheme in which each iteration KL-regularizes the previous model by fine-tuning with the negative score as reward gradient, and prove a one-step optimality result under exact score estimation and exact optimization (Theorem 5.2). Under Assumptions 7.1–7.3 (support compatibility, precompactness, and noise/bias conditions on the fine-tuning oracle), Theorem 7.1 claims almost-sure weak convergence of the induced marginal densities to the maximum-entropy distribution on Ω_pre. Experimental results on a synthetic task and on text-to-image generation with Stable Diffusion support the practical claim of increased sample diversity.

Significance. The strength of the paper is conceptual: Eq. (12) is a simple, parameter-free identity that connects the entropy first variation to the score, and the mirror-descent interpretation of sequential fine-tuning gives a principled escalation from the idealized one-step result to a stochastic-approximation analysis. The paper builds on the right external machinery (relative smoothness, mirror flows, and Benaim-style asymptotic pseudotrajectories) and is transparent about the gap between idealized and realistic assumptions, which is commendable. If the theory were fully rigorous, the contribution would be significant: a provable, density-estimation-free method for max-entropy exploration with diffusion models. However, three load-bearing gaps currently prevent that: the well-posedness of the entropy objective on a possibly lower-dimensional support is not established, Assumption 7.1 refers to an undefined set, and Assumption 7.3 is stated at the level of the oracle error and not verified for the proposed solver.

major comments (3)
  1. [Section 3, Eqs. (7)-(8); Appendix C, Eq. (MF)] The maximum-entropy objective in Problem (7) is not well defined for the manifold support that motivates it. Eq. (8) defines H(μ) = −∫ dμ log(dμ/dx) with all integrals taken with respect to Lebesgue measure (Section 2), which requires every admissible μ ∈ P(Ω_pre) to have a density with respect to Lebesgue; however, Section 3 and Fig. 1 motivate Ω_pre as an approximation of a low-dimensional data manifold (m ≪ d), and Proposition 1 proves only compactness, which does not imply positive Lebesgue measure. If Ω_pre has Lebesgue measure zero, no μ ∈ P(Ω_pre) admits a density with respect to Lebesgue, H(μ) is undefined (or −∞), the maximizer p*_T asserted in Theorem 7.1 need not exist as a density, and the mirror flow in (MF), p_t = e^{h_t}/∫_Ω e^{h_t}, is undefined because the normalizing Lebesgue integral vanishes; the KL penalty in Eq. (9) similarly lacks a common reference measure on a lower-dimensional Ω_pre. The manuscript needs an explicit regularity assumption (e.g., Ω_pre has nonempty interior and positive Lebesgue measure, and each p^{π_k}_T is absolutely continuous with respect to Lebesgue) or a reference measure on the manifold; the current tension between the manifold motivation and the Lebesgue-density formalism affects Problem (7), Theorem 5.2, and Theorem 7.1.
  2. [Section 7.1, Assumption 7.1] Assumption 7.1 refers to a set Ω-tilde that is nowhere defined. This is load-bearing rather than cosmetic: the assumptions supp(p^{π_k}_T) ⊂ Ω-tilde for all k and supp(p^{π_k}_j) = Ω-tilde for some j provide the common integration domain on which the dual variables h_k = δH(p^{π_k}_T) = −log p^{π_k}_T are well defined in the proof of Theorem 7.1, and the equality condition implicitly carries the requirement that the relevant densities be absolutely continuous with respect to a common full-dimensional reference measure. As written, the assumption is vacuous; the authors must define Ω-tilde and state its regularity properties, and should justify that the equality supp(p^{π_k}_j) = Ω-tilde is achievable for Algorithm 1.
  3. [Section 7.1, Assumption 7.3; Appendix C, proof of Theorem 7.1] The main convergence guarantee is conditional on Assumption 7.3, which is asserted directly on the oracle errors b_k and U_k, but the paper offers no analysis showing that the proposed instantiation of LINEARFINETUNINGSOLVER (Algorithm 2, based on Adjoint Matching) satisfies the bias-decay and summability conditions in terms of its inner-loop parameters (integration steps, gradient steps, batch sizes). Consequently, Theorem 7.1 is close to a restatement of the assumption that the solver tracks the idealized mirror iterates (MD_k): the same assumption also absorbs the unresolved gap between the densities reachable by diffusion policies and the full space P(Ω_pre) over which (MD_k) optimizes. In the proof, the step asserting that the interpolated process is an asymptotic pseudotrajectory of (MF) is delegated to 'standard arguments' without showing how Assumptions 7.1-7.3 control the discretization error for the specific nonlinear map h ↦ −log(e^h/∫ e^h). The theorem would be strengthened by either deriving verifiable bounds on the bias and noise of the concrete solver or restating the result as an explicitly conditional guarantee.
minor comments (5)
  1. [Section 7, Remark after Theorem 7.1] The claimed convergence rate of order Õ((log log k)^{−1}) is stated without derivation or a precise reference; either the rate proof should be included or the remark should be removed.
  2. [Appendix B, Eqs. (24)-(27)] The proof of Theorem 5.2 uses inconsistent symbols (a free 'μ' in Eq. (27) and in the expression '1 + kμ/(L−μ)', with 'l', 'k', and 'K' conflated), which obscures the argument; the variables should be made consistent throughout the proof.
  3. [Section 8, Tables 1 and 2] The last row of Table 1 is labeled 'H(p, p^pre_T)' in the header but is described in the text as Gaussian cross-entropy, and the caption of Table 2 says 'FID, CLIP and cross-entropy evaluation' although the table reports Vendi and CLIP scores; the captions and row labels should be corrected.
  4. [Section 7.1 and Algorithm 1] The Robbins-Monro conditions are stated for γ_k, but Algorithm 1 exposes only the regularization coefficients α_k = 1/γ_k, and the theoretical schedule γ_k → 0 with summable squares corresponds to α_k → ∞, which does not match the fixed-regularization experimental protocol described in Section 8; the paper should specify the recommended α_k schedule and clarify the relationship between the theoretical step-size conditions and the experiments.
  5. [Section 8 and Appendix D] The implementation scales the reward gradient by λ (λ = 0.1 in both experiments), while the theory in Sections 5 and 7 uses the unscaled score gradient −s^{k−1}(·, T); the paper should explain how this scaling is absorbed in the analysis (for example, into the step size or the bias/noise terms).

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the score–entropy identity (Eq. 12) is exact, and the convergence proof applies external mirror-flow/stochastic-approximation theorems rather than assuming the target.

full rationale

The paper's derivation chain is self-contained. The score–entropy connection (Eq. 12) is an exact identity: since δH(p) = −log p (up to an additive constant), ∇_x δH(p) = −∇_x log p = −score; this is calculus, not a fitted or assumed relation. Theorem 5.2 follows from the closed-form solution of the KL-regularized linearization (Eq. 9) with α = 1, which is p ∝ p_pre · exp(−log p_pre) ∝ 1, i.e., the uniform (maximum-entropy) distribution on Ω_pre; the proof's relative-smoothness formalism is a standard mirror-descent argument and does not inject the conclusion as an input. Theorem 7.1 is proved by showing the iterates are an asymptotic pseudotrajectory of the mirror flow (MF) and that the flow's only internally chain-transitive set is the entropy maximizer; these steps invoke published external results (Benaïm 2006; Benaïm & Hirsch 1996) and published mirror-flow analyses (Hsieh et al. 2019; Karimi et al. 2024). Those cited works are mathematical theorems with stated assumptions that do not include the present theorem; the co-authorship overlap does not make the support circular under the rules. Assumption 7.3 requires the fine-tuning oracle's bias b_k and noise U_k to vanish in a summable sense; it is a strong realizability condition on the solver, but it does not define p*_T in terms of the iterates nor assume convergence to p*_T—the convergence is derived from the APT/ICT machinery. The possible ill-posedness of differential entropy on a measure-zero manifold (Sec. 3, Eq. 8) is a correctness/well-posedness concern, not a circularity. No prediction in the paper reduces by construction to a fitted input or to a self-citation chain.

Assumptions & free parameters 2 free parameters · 6 assumptions · 1 invented entities

The central derivation rests on the regularity of the densities (potentially conflicting with the manifold motivation), on the exactness of the score identity, and on convergence conditions assumed for the solver. The approximate data manifold is a new implicit object with no independent evidence. No new physical entities are introduced.

free parameters (2)
  • KL regularization coefficient alpha_k (mirror descent step size gamma_k = 1/alpha_k) = lambda = 0.1 in experiments; no concrete schedule for the general theorem
    Controls the trade-off between exploration and staying close to the pre-trained model; required as input to Algorithm 1, and only abstract Robbins-Monro conditions are stated in Theorem 7.1.
  • Reward gradient scale lambda in LINEARFINETUNINGSOLVER = 0.1
    Multiplies the score-based reward gradient in the Adjoint Matching implementation; chosen by hand and not derived from the optimization problem.
assumptions (6)
  • domain assumption The marginals p_pre_T and p^pi_T admit densities with respect to Lebesgue measure on R^d, so differential entropy and KL divergence are finite and well-defined.
    Problem (7) and Eq. (10) require Lebesgue densities; the low-dimensional manifold motivation (Fig. 1) suggests the support could have measure zero, and no reference measure on Omega_pre is provided.
  • domain assumption The score function s^pi(·,T) equals grad log p^pi_T and can be used as the reward gradient in Eq. (12).
    This identity is the algorithmic core; in practice the score is a neural network approximation, and the theory only handles approximation error through Assumption 7.3.
  • ad hoc to paper Assumption 7.1: supp(p^pi_k_T) subset Omega-tilde for all k and supp(p^pi_k_j) = Omega-tilde for some j, for a set Omega-tilde.
    The set Omega-tilde is never defined and its relation to Omega_pre is unspecified, so the assumption cannot be checked.
  • ad hoc to paper Assumption 7.3: the oracle bias b_k and noise U_k satisfy ||b_k||_inf -> 0, sum E[gamma_k^2(||b_k||^2_inf + ||U_k||^2_inf)] < inf, and sum gamma_k ||b_k||_inf < inf.
    These conditions assume the linear fine-tuning solver becomes asymptotically unbiased at a sufficient rate; the paper does not prove or empirically verify them for the Adjoint Matching solver.
  • standard math Mirror flow and stochastic approximation background: Benaïm (2006), Hsieh et al. (2019, 2021), Lu et al. (2018), Karimi et al. (2024).
    Basis for Theorem 7.1; some are co-authored by Hsieh and Krause, but they are published standard results.
  • standard math Lemma 5.1: differential entropy is 1-smooth and 1-strongly convex relative to -H, so the KL term is the Bregman divergence D_{-H}.
    Used to turn a fine-tuning step into one step of mirror descent; follows from definitions but is load-bearing for the one-step theorem.
invented entities (1)
  • Approximate data manifold Omega_pre := supp(p_pre_T)
    purpose: Defines the constraint set for maximum entropy exploration; all fine-tuned distributions are required to be supported on it.
    The manifold is defined only implicitly through the pre-trained model and its low-dimensional nature is assumed from the manifold hypothesis; no independent measurement or falsifiable prediction is tied to it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Provable Maximum Entropy Manifold Exploration via Diffusion Models." pith.science (2026). https://pith.science/paper/ASAEAEGZ

@misc{pith2026250615385,
  author       = {Pith},
  title        = {Pith review of: Provable Maximum Entropy Manifold Exploration via Diffusion Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ASAEAEGZ}},
  note         = {Machine review of arXiv:2506.15385}
}
read the original abstract

Exploration is critical for solving real-world decision-making problems such as scientific discovery, where the objective is to generate truly novel designs rather than mimic existing data distributions. In this work, we address the challenge of leveraging the representational power of generative models for exploration without relying on explicit uncertainty quantification. We introduce a novel framework that casts exploration as entropy maximization over the approximate data manifold implicitly defined by a pre-trained diffusion model. Then, we present a novel principle for exploration based on density estimation, a problem well-known to be challenging in practice. To overcome this issue and render this method truly scalable, we leverage a fundamental connection between the entropy of the density induced by a diffusion model and its score function. Building on this, we develop an algorithm based on mirror descent that solves the exploration problem as sequential fine-tuning of a pre-trained diffusion model. We prove its convergence to the optimal exploratory diffusion model under realistic assumptions by leveraging recent understanding of mirror flows. Finally, we empirically evaluate our approach on both synthetic and high-dimensional text-to-image diffusion, demonstrating promising results.

Figures

Figures reproduced from arXiv: 2506.15385 by the authors.

Figure 1
Figure 1. A diffusion model π pre pre-trained on a set of points (white) implicitly learns a set Ωpre (orange) approx￾imating the true low-dimensional data manifold M ⊆ R m (black) with m ≪ d. The approximate data manifold Ωpre can be significantly smaller than R d (yellow). et al., 2023). Traditionally, generative models have been employed to capture the underlying data distribution in high￾dimensional spaces, facilitating p… view at source ↗
Figure 2
Figure 2. (2a) Diffusion processes and marginal densities corresponding to the pre-trained model π pre (green), and maximally explorative fine-tuned model π ∗ (violet). (2b) Fine-tuning a pre-trained diffusion model (green) via the surprise maximization principle in Eq. (9) one obtains a diffusion model (violet) able to sample low-density regions. the true score ∇x log pt(x) with sθ ∗ , leading to the process {X← t }t∈[0,T] .… view at source ↗
Figure 3
Figure 3. Illustrative example with unbalanced pre-trained model [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Generated images from π pre (top) and π3 (bottom) for a fixed set of initial noisy samples using the prompt ”A creative architecture.”. We observe an increase in complexity and originality of the S-MEME generated images while preserving semantic faithfulness, likely hi…
Figure 5
Figure 5. Figure 5: Generated images from π pre (top) and π3 (bottom) for a fixed set of initial noisy samples using the prompt ”A creative architecture.” [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]
Figure 6
Figure 6. Figure 6: Generated images from π pre with prompt ”A creative impressionist painting.” 20 [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: Generated images obtained via fine-tuning of [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: Generated images from π pre with prompt ”Creative furniture.” 21 [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Generated images obtained via fine-tuning of [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: Generated images from π pre with prompt ”An innovative car design.” 22 [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]
Figure 11
Figure 11. Figure 11: Generated images obtained via fine-tuning of [PITH_FULL_IMAGE:figures/full_fig_p023_11.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 31 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    and Sastry, S

    Achiam, J. and Sastry, S. Surprise-based intrinsic motivation for deep reinforcement learning. arXiv preprint arXiv:1703.01732, 2017

    arXiv 2017

  3. [3]

    Mirror descent with relative smoothness in measure spaces, with application to sinkhorn and em

    Aubin-Frankowski, P.-C., Korba, A., and L \'e ger, F. Mirror descent with relative smoothness in measure spaces, with application to sinkhorn and em. Advances in Neural Information Processing Systems, 35: 0 17263--17275, 2022

    work page 2022

  4. [4]

    Dynamics of stochastic approximation algorithms

    Bena \" m, M. Dynamics of stochastic approximation algorithms. In Seminaire de probabilites XXXIII, pp.\ 1--68. Springer, 2006

    work page 2006

  5. [5]

    and Hirsch, M

    Bena \" m, M. and Hirsch, M. W. Asymptotic pseudotrajectories and chain recurrent flows, with applications. Journal of Dynamics and Differential Equations, 8: 0 141--176, 1996

    work page 1996

  6. [6]

    Bilodeau, C., Jin, W., Jaakkola, T., Barzilay, R., and Jensen, K. F. Generative models for molecular discovery: Recent advances and challenges. Wiley Interdisciplinary Reviews: Computational Molecular Science, 12 0 (5): 0 e1608, 2022

    work page 2022

  7. [7]

    Score approximation, estimation and distribution recovery of diffusion models on low-dimensional data

    Chen, M., Huang, K., Zhao, T., and Wang, M. Score approximation, estimation and distribution recovery of diffusion models on low-dimensional data. In International Conference on Machine Learning, pp.\ 4672--4712. PMLR, 2023

    2023

  8. [8]

    Diffusion policy: Visuomotor policy learning via action diffusion

    Chi, C., Feng, S., Du, Y., Xu, Z., Cousineau, E., Burchfiel, B., and Song, S. Diffusion policy: Visuomotor policy learning via action diffusion. arXiv preprint arXiv:2303.04137, 2023

    arXiv 2023

Show all 73 references
  1. [9]

    Diffdock: Diffusion steps, twists, and turns for molecular docking

    Corso, G., St \"a rk, H., Jing, B., Barzilay, R., and Jaakkola, T. Diffdock: Diffusion steps, twists, and turns for molecular docking. arXiv preprint arXiv:2210.01776, 2022

    2022 arXiv

  2. [10]

    Particle guidance: non-iid diverse sampling with diffusion models

    Corso, G., Xu, Y., De Bortoli, V., Barzilay, R., and Jaakkola, T. Particle guidance: non-iid diverse sampling with diffusion models. arXiv preprint arXiv:2310.13102, 2023

    2023 arXiv

  3. [11]

    and Madre, J

    Coudert, O. and Madre, J. C. A unified framework for the formal verification of sequential circuits. In The Best of ICCAD: 20 Years of Excellence in Computer-Aided Design, pp.\ 39--50. Springer, 1990

    1990

  4. [12]

    A., Liniger, N., Mutti, M., and Krause, A

    De Santi, R., Joseph, F. A., Liniger, N., Mutti, M., and Krause, A. Geometric active exploration in markov decision processes: the benefit of abstraction. arXiv preprint arXiv:2407.13364, 2024 a

    2024 arXiv

  5. [13]

    Global reinforcement learning: Beyond linear and convex rewards via submodular semi-gradient methods

    De Santi, R., Prajapat, M., and Krause, A. Global reinforcement learning: Beyond linear and convex rewards via submodular semi-gradient methods. arXiv preprint arXiv:2407.09905, 2024 b

    2024 arXiv

  6. [14]

    Domingo-Enrich, C., Drozdzal, M., Karrer, B., and Chen, R. T. Adjoint matching: Fine-tuning flow and diffusion generative models with memoryless stochastic optimal control. arXiv preprint arXiv:2409.08861, 2024

    2024 arXiv

  7. [15]

    Reinforcement learning in continuous time and space

    Doya, K. Reinforcement learning in continuous time and space. Neural computation, 12 0 (1): 0 219--245, 2000

    2000

  8. [16]

    and Zhu, J.-J

    Dvurechensky, P. and Zhu, J.-J. Analysis of kernel mirror prox for measure optimization. In International Conference on Artificial Intelligence and Statistics, pp.\ 2350--2358. PMLR, 2024

    2024

  9. [17]

    and Schuffenhauer, A

    Ertl, P. and Schuffenhauer, A. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions. Journal of cheminformatics, 1: 0 1--11, 2009

    2009

  10. [18]

    Fleming, W. H. and Rishel, R. W. Deterministic and stochastic optimal control, volume 1. Springer Science & Business Media, 2012

    2012

  11. [19]

    and Dieng, A

    Friedman, D. and Dieng, A. B. The vendi score: A diversity evaluation metric for machine learning. arXiv preprint arXiv:2210.02410, 2022

    2022 arXiv

  12. [20]

    D., Azar, M

    Guo, Z. D., Azar, M. G., Saade, A., Thakoor, S., Piot, B., Pires, B. A., Valko, M., Mesnard, T., Lattimore, T., and Munos, R. Geometric entropic exploration. arXiv preprint arXiv:2101.02055, 2021

    2021 arXiv

  13. [21]

    Halmos, P. R. Measure theory, volume 18. Springer, 2013

    2013

  14. [22]

    Provably efficient maximum entropy exploration

    Hazan, E., Kakade, S., Singh, K., and Van Soest, A. Provably efficient maximum entropy exploration. In International Conference on Machine Learning, 2019

    2019

  15. [23]

    Clipscore: A reference-free evaluation metric for image captioning

    Hessel, J., Holtzman, A., Forbes, M., Le Bras, R., and Choi, Y. Clipscore: A reference-free evaluation metric for image captioning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp.\ 7514--7528, 2021

    2021

  16. [24]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., and Hochreiter, S. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017

    2017

  17. [25]

    and Lemar \'e chal, C

    Hiriart-Urruty, J.-B. and Lemar \'e chal, C. Fundamentals of convex analysis. Springer Science & Business Media, 2004

    2004

  18. [26]

    and Salimans, T

    Ho, J. and Salimans, T. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022

    2022 arXiv

  19. [27]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    2020

  20. [28]

    G., Vignac, C., and Welling, M

    Hoogeboom, E., Satorras, V. G., Vignac, C., and Welling, M. Equivariant diffusion for molecule generation in 3d. In International conference on machine learning, pp.\ 8867--8887. PMLR, 2022

    2022

  21. [29]

    Finding mixed nash equilibria of generative adversarial networks

    Hsieh, Y.-P., Liu, C., and Cevher, V. Finding mixed nash equilibria of generative adversarial networks. In International Conference on Machine Learning, pp.\ 2810--2819. PMLR, 2019

    2019

  22. [30]

    The limits of min-max optimization algorithms: Convergence to spurious non-critical sets

    Hsieh, Y.-P., Mertikopoulos, P., and Cevher, V. The limits of min-max optimization algorithms: Convergence to spurious non-critical sets. In International Conference on Machine Learning, pp.\ 4337--4348. PMLR, 2021

    2021

  23. [31]

    and Zhou, X

    Jia, Y. and Zhou, X. Y. Policy evaluation and temporal-difference learning in continuous time and space: A martingale approach. Journal of Machine Learning Research, 23 0 (154): 0 1--55, 2022

    2022

  24. [32]

    L., Hosseinzadeh, R., Cresswell, J

    Kamkari, H., Ross, B. L., Hosseinzadeh, R., Cresswell, J. C., and Loaiza-Ganem, G. A geometric view of data complexity: Efficient local intrinsic dimension estimation with diffusion models. arXiv preprint arXiv:2406.03537, 2024

    2024 arXiv

  25. [33]

    R., Hsieh, Y.-P., and Krause, A

    Karimi, M. R., Hsieh, Y.-P., and Krause, A. Sinkhorn flow as mirror flow: A continuous-time framework for generalizing the sinkhorn algorithm. In International Conference on Artificial Intelligence and Statistics, pp.\ 4186--4194. PMLR, 2024

    2024

  26. [34]

    Variational diffusion models

    Kingma, D., Salimans, T., Poole, B., and Ho, J. Variational diffusion models. Advances in neural information processing systems, 34: 0 21696--21707, 2021

    2021

  27. [35]

    Sparse repellency for shielded generation in text-to-image diffusion models

    Kirchhof, M., Thornton, J., Ablin, P., B \'e thune, L., Ndiaye, E., and Cuturi, M. Sparse repellency for shielded generation in text-to-image diffusion models. arXiv preprint arXiv:2410.06025, 2024

    2024 arXiv

  28. [36]

    Convergence for score-based generative modeling with polynomial complexity

    Lee, H., Lu, J., and Tan, Y. Convergence for score-based generative modeling with polynomial complexity. Advances in Neural Information Processing Systems, 35: 0 22870--22882, 2022

    2022

  29. [37]

    Efficient exploration via state marginal matching

    Lee, L., Eysenbach, B., Parisotto, E., Xing, E., Levine, S., and Salakhutdinov, R. Efficient exploration via state marginal matching. arXiv preprint arXiv:1906.05274, 2019

    1906 arXiv

  30. [38]

    A gradient descent perspective on sinkhorn

    L \'e ger, F. A gradient descent perspective on sinkhorn. Applied Mathematics & Optimization, 84 0 (2): 0 1843--1855, 2021

    2021

  31. [39]

    Diffusion model for data-driven black-box optimization

    Li, Z., Yuan, H., Huang, K., Ni, C., Ye, Y., Chen, M., and Wang, M. Diffusion model for data-driven black-box optimization. arXiv preprint arXiv:2403.13219, 2024

    2024 arXiv

  32. [40]

    Information directed reward learning for reinforcement learning

    Lindner, D., Turchetta, M., Tschiatschek, S., Ciosek, K., and Krause, A. Information directed reward learning for reinforcement learning. Advances in Neural Information Processing Systems, 34: 0 3850--3862, 2021

    2021

  33. [41]

    and Abbeel, P

    Liu, H. and Abbeel, P. Behavior from the void: Unsupervised active pre-training. Advances in Neural Information Processing Systems, 34: 0 18459--18473, 2021

    2021

  34. [42]

    M., and Nesterov, Y

    Lu, H., Freund, R. M., and Nesterov, Y. Relatively smooth convex optimization by first-order methods, and applications. SIAM Journal on Optimization, 28 0 (1): 0 333--354, 2018

    2018

  35. [43]

    A unified stochastic approximation framework for learning in games

    Mertikopoulos, P., Hsieh, Y.-P., and Cevher, V. A unified stochastic approximation framework for learning in games. Mathematical Programming, 203 0 (1): 0 559--609, 2024

    2024

  36. [44]

    Training diffusion models towards diverse image generation with reinforcement learning

    Miao, Z., Wang, J., Wang, Z., Yang, Z., Wang, L., Qiu, Q., and Liu, Z. Training diffusion models towards diverse image generation with reinforcement learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 10844--10853, 2024

    2024

  37. [45]

    Active exploration via experiment design in M arkov chains

    Mutny, M., Janik, T., and Krause, A. Active exploration via experiment design in M arkov chains. In International Conference on Artificial Intelligence and Statistics, 2023

    2023

  38. [46]

    Task-agnostic exploration via policy gradient of a non-parametric state entropy estimate

    Mutti, M., Pratissoli, L., and Restelli, M. Task-agnostic exploration via policy gradient of a non-parametric state entropy estimate. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pp.\ 9028--9036, 2021

    2021

  39. [47]

    Challenging common assumptions in convex reinforcement learning

    Mutti, M., De Santi, R., De Bartolomeis, P., and Restelli, M. Challenging common assumptions in convex reinforcement learning. Advances in Neural Information Processing Systems, 35: 0 4489--4502, 2022 a

    2022

  40. [48]

    The importance of non-markovianity in maximum state entropy exploration

    Mutti, M., De Santi, R., and Restelli, M. The importance of non-markovianity in maximum state entropy exploration. In International Conference on Machine Learning, pp.\ 16223--16239. PMLR, 2022 b

    2022

  41. [49]

    Convex reinforcement learning in finite trials

    Mutti, M., De Santi, R., De Bartolomeis, P., and Restelli, M. Convex reinforcement learning in finite trials. Journal of Machine Learning Research, 24 0 (250): 0 1--42, 2023

    2023

  42. [50]

    Nemirovskij, A. S. and Yudin, D. B. Problem complexity and method efficiency in optimization. 1983

    1983

  43. [51]

    Score-based generative models detect manifolds

    Pidstrigach, J. Score-based generative models detect manifolds. Advances in Neural Information Processing Systems, 35: 0 35852--35865, 2022

    2022

  44. [52]

    N., and Krause, A

    Prajapat, M., Mutn \`y , M., Zeilinger, M. N., and Krause, A. Submodular reinforcement learning. arXiv preprint arXiv:2307.13372, 2023

    2023 arXiv

  45. [53]

    High-resolution image synthesis with latent diffusion models, 2021

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. High-resolution image synthesis with latent diffusion models, 2021

    2021

  46. [54]

    Sadat, S., Buhmann, J., Bradley, D., Hilliges, O., and Weber, R. M. Cads: Unleashing the diversity of diffusion models through condition-annealed sampling, 2024. URL https://arxiv.org/abs/2310.17347

    2024 arXiv

  47. [55]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., Coombes, T., Katta, A., Mullis, C., Wortsman, M., et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in Neural Information Processing Systems, 35: 0...

    2022

  48. [56]

    State entropy maximization with random encoders for efficient exploration

    Seo, Y., Chen, L., Shin, J., Lee, H., Abbeel, P., and Lee, K. State entropy maximization with random encoders for efficient exploration. In International Conference on Machine Learning, pp.\ 9443--9454. PMLR, 2021

    2021

  49. [57]

    J., Tong, A., and Neklyudov, K

    Skreta, M., Atanackovic, L., Bose, A. J., Tong, A., and Neklyudov, K. The superposition of diffusion models using the it \^ o density estimator. arXiv preprint arXiv:2412.17762, 2024

    2024 arXiv

  50. [58]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Sohl-Dickstein, J., Weiss, E., Maheswaranathan, N., and Ganguli, S. Deep unsupervised learning using nonequilibrium thermodynamics. In International conference on machine learning, pp.\ 2256--2265. PMLR, 2015

    2015

  51. [59]

    and Ermon, S

    Song, Y. and Ermon, S. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019

    2019

  52. [60]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456, 2020

    2011 arXiv

  53. [61]

    P., Batzolis, G., Deveney, T., and Sch \"o nlieb, C.-B

    Stanczuk, J. P., Batzolis, G., Deveney, T., and Sch \"o nlieb, C.-B. Diffusion models encode the intrinsic dimension of data manifolds. In Forty-first International Conference on Machine Learning, 2024

    2024

  54. [62]

    Fine-tuning of diffusion models via stochastic control: entropy regularization and beyond

    Tang, W. Fine-tuning of diffusion models via stochastic control: entropy regularization and beyond. arXiv preprint arXiv:2403.06279, 2024

    2024

  55. [63]

    and Zhao, H

    Tang, W. and Zhao, H. Contractive diffusion probabilistic models. arXiv preprint arXiv:2401.13115, 2024

    2024 arXiv

  56. [64]

    L., Tseng, A

    Uehara, M., Zhao, Y., Black, K., Hajiramezanali, E., Scalia, G., Diamant, N. L., Tseng, A. M., Biancalani, T., and Levine, S. Fine-tuning of continuous-time diffusion models as entropy-regularized control. arXiv preprint arXiv:2402.15194, 2024 a

    2024 arXiv

  57. [65]

    L., Tseng, A

    Uehara, M., Zhao, Y., Black, K., Hajiramezanali, E., Scalia, G., Diamant, N. L., Tseng, A. M., Levine, S., and Biancalani, T. Feedback efficient online fine-tuning of diffusion models. arXiv preprint arXiv:2402.16359, 2024 b

    2024 arXiv

  58. [66]

    and Ye, J

    Um, S. and Ye, J. C. Self-guided generation of minority samples using diffusion models. In European Conference on Computer Vision, pp.\ 414--430. Springer, 2025

    2025

  59. [67]

    Um, S., Lee, S., and Ye, J. C. Don't play favorites: Minority guidance for diffusion models. arXiv preprint arXiv:2301.12334, 2023

    2023 arXiv

  60. [68]

    A connection between score matching and denoising autoencoders

    Vincent, P. A connection between score matching and denoising autoencoders. Neural computation, 23 0 (7): 0 1661--1674, 2011

    2011

  61. [69]

    Wang, H., Zariphopoulou, T., and Zhou, X. Y. Reinforcement learning in continuous time and space: A stochastic control approach. Journal of Machine Learning Research, 21 0 (198): 0 1--34, 2020

    2020

  62. [70]

    Exploring low-toxicity chemical space with deep learning for molecular generation

    Yang, Y., Wu, Z., Yao, X., Kang, Y., Hou, T., Hsieh, C.-Y., and Liu, H. Exploring low-toxicity chemical space with deep learning for molecular generation. Journal of Chemical Information and Modeling, 62 0 (13): 0 3191--3199, 2022

    2022

  63. [71]

    Mattergen: a generative model for inorganic materials design

    Zeni, C., Pinsler, R., Z \"u gner, D., Fowler, A., Horton, M., Fu, X., Shysheya, S., Crabb \'e , J., Sun, L., Smith, J., et al. Mattergen: a generative model for inorganic materials design. arXiv preprint arXiv:2312.03687, 2023

    2023 arXiv

  64. [72]

    D., and Tang, W

    Zhao, H., Chen, H., Zhang, J., Yao, D. D., and Tang, W. Scores as actions: a framework of fine-tuning diffusion models by continuous-time reinforcement learning. arXiv preprint arXiv:2409.08400, 2024

    2024 arXiv

  65. [73]

    Toward understanding generative data augmentation

    Zheng, C., Wu, G., and Li, C. Toward understanding generative data augmentation. Advances in neural information processing systems, 36: 0 54046--54060, 2023

    2023

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.