Pith. sign in

REVIEW 4 major objections 5 minor 31 references

Diversity By Design: Leveraging Distribution Matching for Offline Model-Based Optimization

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read DynAMO makes offline model-based optimization produce diverse designs without sacrificing the best candidate.

desk verdict DynAMO is a useful empirical plug-in for diversity in offline MBO, but the theoretical justification is rougher than the paper claims. read the letter →

arxiv 2501.18768 v2 pith:FH6NXCWK submitted 2025-01-30 cs.LG cs.AI

Michael S. Yao , James C. Gee , Osbert Bastani This is my paper · ORCID
classification cs.LGcs.AI
keywords offlinemodel-basedoptimizationdesigndiversitydistributionmatchingKLdivergenceLagrangiandualitysourcecriticgenerativeBayesian
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 proposes DynAMO, a way to modify any offline model-based optimization (MBO) formulation so that the final batch of proposed designs is diverse while still containing high-quality candidates. The central idea is to add a KL-divergence penalty that pushes the distribution of generated designs toward a temperature-weighted distribution of the offline dataset, thereby inheriting the dataset's diversity. The paper derives an explicit, concave lower bound for the Lagrangian dual function using the Fenchel conjugate of $f(u) = u \log u$, which makes the constrained objective tractable with any standard optimizer. Across six benchmark tasks and six backbone optimizers, DynAMO consistently improves pairwise diversity while keeping Best@128 quality competitive. If the paper is right, design diversity can be treated as a drop-in objective term rather than a separate algorithm.

What carries the argument

The load-bearing objects are the $\tau$-weighted distribution $p^\tau_D(x) \propto \exp(\tau r(x))$ over the offline dataset, which serves as the reference distribution whose diversity the policy is pushed to match, and the KL-divergence penalty $D_{\mathrm{KL}}(q_\pi \| p^\tau_D)$, which is the diversity term. The proof machinery is the Fenchel conjugate of the KL generator $f_{\mathrm{KL}}(u) = u \log u$, namely $f^*_{\mathrm{KL}}(v) = e^{v-1}$, which converts the Lagrangian dual into a closed-form concave function of $\lambda$. A 1-Lipschitz source critic $c^*$ supplies the out-of-distribution constraint, and together these pieces make the Lagrangian explicit enough that any standard optimization algorithm can be applied.

What would settle it

On a continuous task such as Superconductor or D'Kitty, compute the objective in Eq. (5) literally with the empirical $p^\tau_D$ as a sum of Dirac masses and a Gaussian or neural policy $q_\pi$; the KL divergence will be $+\infty$ for any continuous $q_\pi$, showing the algorithm as stated cannot run without additional smoothing. Separately, train the source critic to near-optimality with $W_0 = 0$ and check whether the constraint $\mathbb{E}_{p^\tau_D}[c^*] - \mathbb{E}_{q_\pi}[c^*] \le 0$ forces the Wasserstein-1 distance to zero, which would imply $q_\pi = p^\tau_D$ and eliminate reward optimization.

Watch

Extended reading notes

Core claim

The central claim is that the offline MBO objective $J(\pi) = \mathbb{E}_{q_\pi}[r_\theta(x)] - \frac{\beta}{\tau} D_{\mathrm{KL}}(q_\pi \| p^\tau_D)$, subject to a source-critic constraint $\mathbb{E}_{p^\tau_D}[c^*(x)] - \mathbb{E}_{q_\pi}[c^*(x)] \le W_0$, admits an explicit dual lower bound $g_\ell(\lambda) = \beta\big[\lambda(\mathbb{E}_{p^\tau_D}[c^*(x)] - W_0) - \mathbb{E}_{p^\tau_D} e^{\lambda c^*(x) - 1}\big]$. Because $g_\ell(\lambda)$ is concave in $\lambda$, the optimal Lagrange multiplier can be found by gradient ascent, leaving any backbone optimizer free to solve the resulting unconstrained Lagrangian. The paper argues that maximizing this modified objective makes a generative policy capture the diversity of high-scoring designs in the offline dataset, and it reports experiments showing that DynAMO-augmented optimizers discover both diverse and high-quality candidates.

Load-bearing premise

The derivation assumes the KL divergence between a continuous generative policy and the finite empirical $\tau$-weighted dataset distribution is a well-defined finite quantity, but as written that divergence is infinite or undefined for continuous design spaces without an explicit density estimator; a second fragile premise is that the source-critic constraint with $W_0=0$ does not collapse the solution to $q_\pi = p^\tau_D$.

Editorial extensions

If this is right

  • Any existing MBO pipeline can be augmented with DynAMO by replacing its objective with the derived Lagrangian; the paper reports diversity gains across gradient ascent, Adam, CMA-ES, CoSyNE, BO-qEI, and BO-qUCB.
  • The final batch of 128 candidates retains competitive Best@128 oracle scores, so the diversity improvement does not come at the cost of best-candidate quality in the tested benchmarks.
  • The method applies to both discrete design spaces, through a VAE latent space, and continuous design spaces, making it portable across scientific domains.
  • The framework extends to mixed $f$-divergences; the appendix derives an analogous dual lower bound for a mixed $\chi^2$-KL divergence, though the experiments show KL alone is non-inferior.
  • A more diverse batch enables better downstream evaluation of secondary objectives such as manufacturing cost, toxicity, or DNA GC content, as demonstrated on vehicle safety, welded beam, and UTR tasks.

Reading between the lines

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

  • The 'provably optimal' claim should be read as optimal for the lower-bounded surrogate dual, not necessarily for the original non-convex primal, since the derivation only guarantees a lower bound on the dual function.
  • A reader checking the theory should look for the missing density estimator: with a finite empirical $p^\tau_D$ and a continuous generative policy, $D_{\mathrm{KL}}(q_\pi \| p^\tau_D)$ is infinite or undefined as written, so the derivation requires an implicit smoothing step to hold literally.
  • A direct testable extension is to use a synthetic design space with known modes and measure whether the number and location of modes recovered by DynAMO match the high-weight modes of $p^\tau_D$; the Branin figure suggests this behavior but does not quantify it.
  • The source-critic constraint with $W_0 = 0$ pins the Wasserstein-1 distance between $q_\pi$ and $p^\tau_D$ to zero when the critic is near-optimal, which would force $q_\pi = p^\tau_D$ and contradict reward maximization; the algorithm's practical success likely depends on the critic not being trained to optimality, an implicit gap between theory and implementation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes DynAMO, a wrapper for offline model-based optimization (MBO) that adds a KL-divergence-based distribution-matching term to the surrogate objective, encouraging the sampling distribution to match a τ-weighted empirical distribution over the offline dataset, and adds an adversarial source-critic constraint to limit out-of-distribution queries. The authors derive DynAMO as a solution to a Lagrangian-dual reformulation of this objective, claiming a provably optimal and optimizer-agnostic method. Experiments on six MBO tasks across six backbone optimizers (gradient ascent, Adam, CMA-ES, CoSyNE, BO-qEI, BO-qUCB) report that DynAMO substantially improves pairwise diversity while keeping Best@128 competitive with existing MBO baselines.

Significance. If the claims held, this would be a practically attractive contribution: a simple, optimizer-agnostic modification of the MBO objective that yields more diverse candidate batches without sacrificing the best oracle-scored design. The empirical study is extensive and well executed: six tasks spanning discrete and continuous design spaces, six backbone optimizers, ten seeds, comparisons to both model-based and model-free baselines, ablations of the two algorithmic components, hyperparameter and budget sensitivity analyses, and a demonstration of downstream secondary-objective exploration. The authors also release code. These strengths make the empirical claims credible even though the theoretical derivation, as written, has load-bearing gaps that prevent the 'provably optimal' claim from being supported by the presented mathematics.

major comments (4)
  1. [§3.2, Definition 3.2, Eq. (5), Algorithm 1] The central objective DKL(qπ||pτ_D) is not well-defined for the setting actually used in the experiments. pτ_D is defined as the empirical τ-weighted distribution over the finite dataset D, while optimization is performed in the continuous VAE latent space (Appendix B). For any continuous generative distribution qπ, DKL(qπ||pτ_D) is infinite because log pτ_D(x) = -∞ for x outside the finite support of D; if qπ is restricted to dataset atoms, the policy cannot propose the novel designs reported in Table 1. The main text and Algorithm 1 never specify a density estimator, kernel smoother, or density-ratio surrogate for pτ_D. Since Lemmas 3.3 and 3.4 and the λ update in Algorithm 1 are all built on this KL term, the theoretical derivation does not currently apply to the algorithm that produced the empirical results. The manuscript must either specify a smoothed/estimated reference density and prove the same dual results for it, or clearly state that the implemented algorithm uses a different surrogate for the KL term.
  2. [§3.4, Lemma 3.4 and the following paragraph] Lemma 3.4 establishes only a lower bound gℓ(λ) ≤ g(λ), and the text correctly says that maximizing gℓ bounds the dual maximum from below. However, the surrounding claims then call the resulting λ the 'globally optimal λ' and describe the overall procedure as a 'provably optimal solution' to (7). A lower-bound dual function does not in general yield the optimal Lagrange multiplier nor a certificate of primal optimality, especially because the primal is non-convex and Slater's condition is not assumed. The abstract and Section 3.5 overstate the formal result; the paper should either strengthen the derivation to an exact dual solution under stated conditions or temper the optimality claims accordingly.
  3. [§4, implementation of W0=0, Algorithm 1] Setting W0 = 0 in the constraint Epτ_D[c*(x)] - Eqπ[c*(x)] ≤ W0 is theoretically problematic. The source critic c* is trained, as in Eq. (4), to maximize exactly this difference subject to a Lipschitz bound, so if c* is close to optimal the constraint forces W1(pτ_D, qπ) ≤ 0, i.e., qπ = pτ_D almost surely. That would preclude discovering designs beyond the offline dataset, contradicting the method's goal and its reported Best@128 improvements. Since W0 = 0 is fixed for all experiments, the paper should explain why the constraint is not vacuous or binding in practice (e.g., because the critic is only approximately trained, or because the constraint is enforced over a finite batch rather than the full distribution) and reconcile this with the theoretical formulation.
  4. [Algorithm 1, line 'Evaluate and cache the candidates according to (9)'] The Lagrangian in Eq. (9) is defined over distributions π, with terms Eqπ[rθ(x)] and Eqπ[c*(x)] and the KL divergence DKL(qπ||pτ_D). Algorithm 1, however, evaluates a per-candidate score -L(xnew; λ) and stores it in Dgen, without specifying how the distribution-level KL term is converted into a pointwise objective. A per-point evaluation of the KL divergence is not defined unless qπ is given a density and pτ_D is smoothed. This is not merely a presentation issue: the final top-k selection in Algorithm 1 is based on these per-candidate scores, so the reader cannot reproduce the method without guessing the per-point surrogate. The algorithm must state the exact per-candidate objective used.
minor comments (5)
  1. [Figure 1 caption] There is a typo: 'singla optima' should read 'single optimum' or 'single optima'.
  2. [§3.4, Eq. (10)] Weak duality gives a lower bound on the primal optimum; the text says it provides a 'tight lower bound,' which is not guaranteed. Please rephrase to 'a lower bound'.
  3. [Lemma 3.3 proof, Eq. (22)-(23)] The proof of Lemma 3.3 relies on the approximation rθ(x) ≈ r(x) over the dataset to replace e^{τ rθ(x)}/Zτ with pτ_D(x). The lemma statement should explicitly state this assumption, since the 'equivalence' is not algebraic but holds only under that approximation.
  4. [Appendix D.4, Theorem D.9, Eqs. (81) and (83)] There are typographical errors in the proof: in Eq. (81), the absolute difference should be between DKL(qπ||pτ(x)) and DKL(qπ||pτ_D(x)), not between two identical terms; and 'log(|4Π|/δ)' in Eq. (83) should be 'log(4|Π|/δ)' for dimensional consistency with Eq. (81).
  5. [Definition 3.2, Eq. (6)] The definition of pτ(x) uses an integral over X; for the discrete tasks (TFBind8, UTR, ChEMBL, Molecule), this should be a sum over the discrete design space, or the paper should clarify that the integral notation is understood in a measure-theoretic sense.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DynAMO's diversity objective is an explicit design choice, its dual derivation is algebraic, and empirical results are judged against external oracles with fixed hyperparameters.

full rationale

I find no circular dependency in the claimed derivation. Eq. (5) explicitly introduces a modified objective; Lemma 3.3 is an algebraic rewrite up to the stated surrogate-calibration assumption, and Lemma 3.4 is a Fenchel-duality lower bound rather than an imported conclusion. The central empirical claim (diversity improvement without loss of Best@128) is tested against external Design-Bench oracles and model-free baselines, with beta=1, tau=1, and W0=0 fixed a priori; no parameter is fitted to the reported PD or Best@128 numbers. The reuse of the source-critic regularizer from the authors' GAMBO paper (Yao et al. 2024) supplies one component, but the KL distribution-matching term and its derivation are independent, and the citation is to a public prior implementation, so it does not carry the argument by itself. The paper's own Limitations section candidly reports weaker Median@128 performance, which is a quality trade-off, not circularity. The main weaknesses are formal: DKL(q_pi || p_tau_D) with an empirical p_tau_D over a continuous VAE latent space requires a density estimator that is never specified, and W0=0 formally forces the W1 constraint to q_pi = p_tau_D; these affect soundness of the theory-to-algorithm link, but they are not cases where an output is equivalent to an input by construction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim relies on three hand-set hyperparameters (β, τ, W0) and on the well-definedness of the KL divergence between a continuous policy and an empirical reference distribution. Theoretical guarantees further assume bounded importance weights and a bounded surrogate error. No new physical entities are introduced; the τ-weighted reference distribution is a modeling construct.

free parameters (3)
  • β (KL regularization strength) = 1.0 (fixed for all tasks)
    Chosen hyperparameter controlling the weight of the KL distribution-matching penalty; not tuned per task, but directly controls the diversity-quality trade-off.
  • τ (temperature of reference distribution) = 1.0 (fixed for all tasks)
    Controls how strongly high-reward designs are weighted in pτ_D; affects what distribution the generator is encouraged to match.
  • W0 (source critic constraint bound) = 0 (fixed for all tasks)
    Constraint bound in Eq. (7). With W0=0 and an optimal 1-Lipschitz critic, the constraint would require W1(pτ_D, qπ) <= 0, a very strong condition.
assumptions (4)
  • domain assumption The learned surrogate rθ approximates the oracle r on the dataset and on the distributions visited under the critic constraint (Assumption D.4).
    Used in Lemma 3.3 and Theorem D.9 to replace rθ by r and to bound prediction error; if false, the distribution-matching objective does not track true reward.
  • ad hoc to paper DKL(qπ||pτ_D) is well-defined and computable for continuous qπ and the finite empirical pτ_D.
    Definition 3.2 and Eq. (5) require this KL; the paper does not specify a density estimator or smoothing, so the objective is formally undefined.
  • domain assumption Bounded importance weights M exist for qπ/p and qπ/pτ_D (Assumption D.6), and optimal policies are in the finite policy class Π (Assumption D.5).
    Needed for the finite-sample bounds in Appendix D.4.
  • standard math Fenchel conjugate of u log u is e^{v-1}, and weak Lagrangian duality applies.
    Standard calculus and convex analysis used in Lemma 3.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diversity By Design: Leveraging Distribution Matching for Offline Model-Based Optimization." pith.science (2026). https://pith.science/paper/FH6NXCWK

@misc{pith2026250118768,
  author       = {Pith},
  title        = {Pith review of: Diversity By Design: Leveraging Distribution Matching for Offline Model-Based Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FH6NXCWK}},
  note         = {Machine review of arXiv:2501.18768}
}
read the original abstract

The goal of offline model-based optimization (MBO) is to propose new designs that maximize a reward function given only an offline dataset. However, an important desiderata is to also propose a diverse set of final candidates that capture many optimal and near-optimal design configurations. We propose Diversity in Adversarial Model-based Optimization (DynAMO) as a novel method to introduce design diversity as an explicit objective into any MBO problem. Our key insight is to formulate diversity as a distribution matching problem where the distribution of generated designs captures the inherent diversity contained within the offline dataset. Extensive experiments spanning multiple scientific domains show that DynAMO can be used with common optimization methods to significantly improve the diversity of proposed designs while still discovering high-quality candidates.

Figures

Figures reproduced from arXiv: 2501.18768 by the authors.

Figure 1
Figure 1. Overview of Diversity in Adversarial Model-based Optimization (DynAMO). Traditional model-based optimiza￾tion (MBO) (Trabucco et al., 2021) techniques can generate high￾scoring designs, although often at the expense of the diversity of proposed designs. Ideally, the final set of candidates should be of high quality while capturing multiple ‘modes of goodness’ within the design space. For example, although there are … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 20 canonical work pages

  1. [1]

    Candidate-Dataset Diversity: How novel is a proposed candidate compared to the real designs previously observed in the offline dataset?

    work page

  2. [2]

    (2023) and Jain et al

    Aggregate Diversity: How well does the batch of candidate designs collectively cover the possible search space? To evaluate (1), we follow prior work by Kim et al. (2023) and Jain et al. (2022) and evaluate theMinimum Novelty (MN) for a batch of k final proposed candidates with respect to the offline dataset D, defined as MN({xF i }k i=1; D) := ExF i min ...

    work page 2023

  3. [3]

    (2020) and Sample et al

    UTR is a discrete, 50-dimensional optimization problem from Angerm¨ueller et al. (2020) and Sample et al. (2019) with the goal of finding an optimal 50-bp DNA sequence that maximizes the gene expression from a 5’ UTR DNA sequence. This is an offline MBO problem from the Design-Bench benchmarking suite (Trabucco et al., 2022) that we use to evaluate offlin...

    work page 2020

  4. [4]

    Arjovsky, M., Chintala, S., and Bottou, L

    URL https://openreview.net/forum? id=HklxbgBKvr. Arjovsky, M., Chintala, S., and Bottou, L. Wasser- stein generative adversarial networks. In Proc ICML, volume 70, pp. 214–23. PMLR, 2017. URL https://proceedings.mlr.press/v70/ arjovsky17a.html. Astudillo, R. and Frazier, P. Bayesian optimization of com- posite functions. In Proc ICML, volume 97, pp. 354–6...

    arXiv 2017

  5. [6]

    Chemingui, Y ., Deshwal, A., Hoang, T

    URL https://proceedings.mlr.press/ v235/chakraborty24b.html. Chemingui, Y ., Deshwal, A., Hoang, T. N., and Doppa, J. R. Offline model-based optimization via policy-guided gradient search. In Proc AAAI, pp. 11230–9, 2024. doi: 10.1609/aaai.v38i10.29001. Chen, C., Zhang, Y ., Fu, J., Liu, X., and Coates, M. Bidi- rectional learning for offline infinite-wid...

    work page arXiv 2024

  6. [7]

    Deb, K., Sunda, J., Rao, U

    URL https://proceedings.mlr.press/ v235/dao24b.html. Deb, K., Sunda, J., Rao, U. B. N., and Chaudhuri, S. Ref- erence point based multi-objective optimization using evolutionary algorithms. Int J Comp Intell Res , 2(3): 273–86, 2006. Defazio, A., Yang, X. A., Khaled, A., Mishchenko, K., Mehta, H., and Cutkosky, A. The road less scheduled. In Proc NeurIPS,...

    work page 2006

  7. [8]

    Du, Y ., Jamasb, A

    URL https://proceedings.mlr.press/ v205/deka23a.html. Du, Y ., Jamasb, A. R., Guo, J., Fu, T., Harris, C., Wang, Y ., Duan, C., Li `o, P., Schwaller, P., and Blun- dell, T. L. Machine learning-aided generative molec- ular design. Nat Mach Intell , 6:589–604, 2024. doi: 10.1038/s42256-024-00843-5. Duci, J., Hazan, E., and Singer, Y . Adaptive subgradient m...

    doi:10.1038/s42256-024-00843-5 2024

  8. [11]

    Jain, M., Bengio, E., Hernandez-Garcia, A., Rector-Brooks, J., Dossou, B

    URL https://proceedings.mlr.press/ v235/hvarfner24a.html. Jain, M., Bengio, E., Hernandez-Garcia, A., Rector-Brooks, J., Dossou, B. F. P., Ekbote, C. A., Fu, J., Zhang, T., Kilgour, M., Zhang, D., Simine, L., Das, P., and Bengio, Y . Biological sequence design with GFlowNets. InProc ICML, volume 162 of Proc Mach Learn Res, pp. 9786–

    work page

Show all 31 references
  1. [13]

    Kingma, D

    doi: 10.5555/3666122.3669080. Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. In Proc ICLR, 2015. doi: 10.48550/arXiv. 1412.6980. Kingma, D. P. and Welling, M. Auto-encoding variational Bayes. In Proc ICLR, 2014. doi: 10.48550/arXiv.1312. 6114. Kong, D., P...

    2015

  2. [14]

    Korshunova, M., Huang, N., Capuzzi, S., Radchenko, D

    URL https://proceedings.mlr.press/ v216/kong23a.html. Korshunova, M., Huang, N., Capuzzi, S., Radchenko, D. S., Savych, O., Moroz, Y . S., Wells, C. I., Willson, T. M., Tropsha, A., and Isayev, O. Generative and reinforcement learning approaches for the automated de novo desig...

    2022 doi

  3. [15]

    Krishnamoorthy, S., Mashkaria, S

    URL https://openreview.net/forum? id=Hyg-JC4FDr. Krishnamoorthy, S., Mashkaria, S. M., and Grover, A. Diffusion models for black-box optimiza- tion. In Proc ICML , volume 202, pp. 17842–57,

  4. [16]

    Kumar, A

    URL https://proceedings.mlr.press/ v202/krishnamoorthy23a.html. Kumar, A. and Levine, S. Model inversion networks for model-based optimization. In Proc NeurIPS, pp. 5126– 37, 2019. doi: 10.5555/3495724.3496155. Kushner, H. J. A new method of locating the maximum of an arbitrar...

    2019

  5. [17]

    doi: 10.1038/s41524-024-01243-4. Ma, Y . J., Yan, J., Jayaraman, D., and Bastani, O. Offline goal-conditioned reinforcement learning via f-advantage regression. In Proc NeurIPS, pp. 310–23, 2022. doi: 10.5555/3600270.3600293. Mashkaria, S. M., Krishnamoorthy, S., and Grover, A...

    2022

  6. [18]

    Matthews, A

    URL https://proceedings.mlr.press/ v202/mashkaria23a.html. Matthews, A. G., Hensman, J., Turner, R., and Ghahra- mani, Z. On sparse variational methods and the Kullback- Leibler divergence between stochastic processes. In Proc Int Conf Artif Intell Stats, volume 51, pp. 231–9. PMLR,

  7. [21]

    Papalexopoulos, T

    doi: 10.1038/s41598-022-08413-8. Papalexopoulos, T. P., Tjandraatmadja, C., Anderson, R., Vielma, J. P., and Belanger, D. Constrained discrete black-box optimization using mixed-integer program- ming. In Proc ICML, volume 162, pp. 17295–322. PMLR,

    doi

  8. [22]

    URL https://proceedings.mlr.press/ v162/papalexopoulos22a.html. Pele, O. and Werman, M. Fast and robust Earth mover’s distances. In Proc ICCV, pp. 460–7, 2009. doi: 10.1109/ ICCV .2009.5459199. Pogue, E. A., New, A., McElroy, K., Le, N. Q., Pekala, M. J., McCue, I., Gienger, E...

    2009

  9. [23]

    Todorov, E., Erez, T., and Tassa, Y

    URL https://proceedings.mlr.press/ v151/terjek22a.html. Todorov, E., Erez, T., and Tassa, Y . MuJoCo: A physics engine for model-based control. In Proc Int Conf Intell Rob Sys, pp. 5026–33, 2012. doi: 10.1109/IROS.2012. 6386109. Trabucco, B., Kumar, A., Geng, X., and Levine, S...

    2012 doi

  10. [24]

    Tripp, A., Daxberger, E., and Hern ´andez-Lobato, J

    URL https://proceedings.mlr.press/ v162/trabucco22a.html. Tripp, A., Daxberger, E., and Hern ´andez-Lobato, J. M. Sample-efficient optimization in the latent space of deep generative models via weighted retraining. In Proc NeurIPS, pp. 11259–72, 2020. doi: 10.5555/3495724. 349...

    2020 doi arXiv

  11. [29]

    (2008) to find an optimal set of car dimensions that minimize the total Mass of the vehicle

    Vehicle Safety is continuous, 5-dimensional optimization problem from Liao et al. (2008) to find an optimal set of car dimensions that minimize the total Mass of the vehicle. The problem initially stems from the multi-objective optimization literature (Blank & Deb, 2020; Liao ...

    2008

  12. [30]

    Welded Beam is a continuous, 4-dimensional optimization problem from Ray & Liew (2002) to find an optimal set of dimensions for a welded steel beam that minimizes the total manufacturing Cost. Similar to the Vehicle task, this problem was initially proposed in the multi-object...

    2002

  13. [801]

    URL https://proceedings

    PMLR, 2022. URL https://proceedings. mlr.press/v162/jain22a.html. Jang, S., Yoo, S., and Kang, N. Generative design by re- inforcement learning: Enhancing the diversity of topol- ogy optimization designs. Computer-Aided Design, 146: 103225, 2022. doi: 10.1016/j.cad.2022.103225...

    2022

  14. [1997]

    15 Diversity By Design: Leveraging Distribution Matching for Offline Model-Based Optimization Appendix Table of Contents A

    doi: 10.1145/279232.279236. 15 Diversity By Design: Leveraging Distribution Matching for Offline Model-Based Optimization Appendix Table of Contents A. Proofs 16 Proof for Lemma 3.1: Diversity Collapse in Reward Optimization Proof for Lemma 3.3: Entropy-Divergence Formulation ...

    2006

  15. [2004]

    We leverage this convergence guarantee to solve for⃗λ∗ na¨ıvely via gradient ascent in DynAMO (see Algorithm 1)

    that g(⃗λ) is convex; assuming that the gradient of g(⃗λ) has a bounded Lipschitz constant, we can therefore arrive in an ε-neighborhood around the optimal ⃗λ∗ within O(1/ε) time (Bubeck, 2015; Grimmer et al., 2023; Zhang et al., 2019). We leverage this convergence guarantee t...

    2017

  16. [2006]

    doi: 10.1007/ 978-0-387-31256-9

    ISBN 978-0-387-29570-1. doi: 10.1007/ 978-0-387-31256-9. 9 Diversity By Design: Leveraging Distribution Matching for Offline Model-Based Optimization Boyd, S. and Vandenberghe, L. Convex optimization . Cambridge university press, 2004. doi: 10.1017/ CBO978051180444. Branin, F....

    doi

  17. [2014]

    Grimmer, B., Shu, K., and Wang, A

    doi: 10.48550/arXiv.1406.2661. Grimmer, B., Shu, K., and Wang, A. L. Accelerated gradient descent via long steps. arXiv Preprint, 2023. doi: 10. 48550/arXiv.2309.09961. Haldar, R. and Mukhopadhyay, D. Levenshtein distance technique in dictionary lookup methods: An improved app...

    2023 doi arXiv

  18. [2016]

    URL https://proceedings.mlr.press/ v51/matthews16.html. Maus, N. T., Jones, H. T., Moore, J. S., Kusner, M. J., Brad- shaw, J., and Gardner, J. R. Local latent space Bayesian optimization over structured inputs. In Proc NeurIPS,

  19. [2020]

    Ament, S., Daulton, S., Eriksson, D., Balandat, M., and Bakshy, E

    URL https://proceedings.mlr.press/ v100/ahn20a.html. Ament, S., Daulton, S., Eriksson, D., Balandat, M., and Bakshy, E. Unexpected improvements to expected im- provement for Bayesian optimization. In Proc NeurIPS,

  20. [2021]

    Eriksson, D., Pearce, M., Gardner, J., Turner, R

    URL https://proceedings.mlr.press/ v161/eriksson21a.html. Eriksson, D., Pearce, M., Gardner, J., Turner, R. D., and Poloczek, M. Scalable global optimization via local Bayesian optimization. In Proc NeurIPS, volume 32, 2019. 10 Diversity By Design: Leveraging Distribution Matc...

    2019 doi arXiv

  21. [2022]

    doi: 10.48550/arXiv.2201.11872. Maus, N. T., Wu, K., Eriksson, D., and Gardner, J. R. Dis- covering many diverse solutions with Bayesian optimiza- tion. In Proc Int Conf AI Stats, volume 206, pp. 1779–98. PMLR, 2023. URL https://proceedings.mlr. press/v206/maus23a.html. Mullis...

    2023 doi arXiv

  22. [2023]

    URL https://openreview.net/forum? id=1vyAG6j9PE. Amos, B. On amortizing convex conjugates for opti- mal transport. In Proc ICLR , 2023. URL https: //openreview.net/forum?id=TQ5WUwS_4ai. Angerm¨ueller, C., Belanger, D., Gane, A., Mariet, Z., Dohan, D., Murphy, K., Colwell, L., ...

    2023

  23. [2024]

    Agrawal, R

    doi: 10.5555/3666122.3666652. Agrawal, R. and Horel, T. Optimal bounds between f- divergences and integral probability metrics. J Mach Learn Res, 22(128):1–59, 2021. doi: 10.48550/arXiv. 2006.05973. Ahn, M., Zhu, H., Hartikainen, K., Ponte, H., Gupta, A., Levine, S., and Kumar...

    2021

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.