Pith. sign in

REVIEW 2 major objections 4 minor 33 references

Decision Making under the Exponential Family: Distributionally Robust Optimisation with Bayesian Ambiguity Sets

T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read For conjugate exponential family models, the DRO-BAS(PE) worst-case risk equals a KL-divergence DRO dual centered at the posterior-mean parameter with radius reduced by a posterior-variance gap—so the robust decision problem is a…

desk verdict BASPE is a genuine, citable reduction; BASPP is oversold in the abstract and its flagship experiments solve an unquantified SAA surrogate. read the letter →

arxiv 2411.16829 v2 pith:RBBV3WD4 submitted 2024-11-25 cs.LG math.OC

Charita Dellaporta , Patrick O'Hara , Theodoros Damoulas This is my paper · ORCID
classification cs.LGmath.OC
keywords distributionallyrobustoptimisationBayesianambiguitysetsconjugateexponentialfamilyKullback-Leiblerdivergenceworst-caserisksingle-stagestochasticprogramNewsvendorproblemportfolio
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 introduces DRO-BAS, a way to hedge decisions against both data noise and parameter uncertainty by replacing the Bayesian expected risk with a worst-case risk over an ambiguity set built from the posterior. Two sets are proposed: one of distributions within KL distance $\epsilon$ of the posterior predictive (DRO-BASPP), and one of distributions whose posterior-averaged KL distance to the model is at most $\epsilon$ (DRO-BASPE). The main result is that for conjugate exponential family models the DRO-BASPE problem has strong duality: the worst-case risk equals a classical KL-DRO dual centered at the posterior-mean parameter, with tolerance reduced by a closed-form posterior variance gap $G$. This makes the robust problem a single-stage stochastic program, and with a linear objective and Gaussian likelihood it is fully closed form and requires no sampling at all. On the Newsvendor and Portfolio problems the resulting formulations match or improve the out-of-sample mean-variance profile of existing Bayesian DRO while solving much faster.

What carries the argument

The load-bearing object is the closed-form decomposition of the posterior-expected KL divergence (Lemma 3.5): $$\mathbb{E}_{\eta\sim\Pi}[d_{\mathrm{KL}}(Q\|P_\eta)] = d_{\mathrm{KL}}(Q,P_{\hat\eta}) + G(\check\tau,\check\nu),$$ where $G$ is the non-negative gap between the posterior expectation of the log-partition function and the log-partition function at the posterior mean, non-negative because $A$ is convex. This identity converts a posterior-averaged divergence constraint into a single-divergence constraint about $P_{\hat\eta}$, which makes the conjugate of the ambiguity functional equal to that of an ordinary KL ball. The classical KL-DRO dual then applies unchanged, and $\epsilon_{\min} = G$ emerges as the minimum non-empty tolerance. The same identity drives the closed-form Gaussian-linear case and the closed-form formulas for the optimal tolerance level in well-specified models.

What would settle it

Take the univariate Normal likelihood with a Normal-Gamma prior and linear payoff $f_x(\xi)=\xi^\top x$, where Theorem 3.6 gives the closed form $\hat\mu x + \sqrt{2(\epsilon-G)}\,\sqrt{x^2 \hat\sigma^2}$; then solve the original primal worst-case risk (11) directly by numerical optimization over candidate distributions (for example tilted Gaussians with varying mean and variance) and compare the two values for several $x$, $\epsilon>G$, and posterior hyperparameters—any persistent gap would disprove strong duality.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 3.6: for a conjugate exponential family likelihood with posterior $\Pi(\eta \mid \check\tau, \check\nu)$ satisfying the differentiability conditions of Lemma 3.5 and for $\epsilon \ge \epsilon_{\min}(n)$, the worst-case risk over the posterior-expectation ambiguity set is exactly $$R_{A_\epsilon(\Pi)}(f_x) = \inf_{\gamma\ge 0}\left\{ \gamma(\epsilon - G(\check\tau,\check\nu)) + \gamma \ln \mathbb{E}_{p(\xi\mid \hat\eta)}\left[\exp(f_x(\xi)/\gamma)\right]\right\},$$ where $\hat\eta = \mathbb{E}_{\Pi}[\eta]$ and $G = \mathbb{E}_{\Pi}[A(\eta)] - A(\hat\eta)\ge 0$. The proof uses Lemma 3.5, which decomposes the posterior-averaged KL divergence as $d_{\mathrm{KL}}(Q,P_{\hat\eta}) + G$, so the ambiguity set collapses, per Corollary 3.7, to the ordinary KL ball of radius $\epsilon - G$ around the posterior-mean distribution $P_{\hat\eta}$. The authors also prove the analogous posterior-predictive dual (Proposition 3.2), give the worst-case distribution in both cases, derive closed-form tolerance levels in well-specified settings, and exhibit a closed-form DRO-BAS(PE) solution for linear objectives with a Normal likelihood.

Load-bearing premise

The load-bearing premise is that the payoff's moment-generating function is finite under the nominal distribution: without it the KL-dual formula diverges, and DRO-BASPP's exact dual is already unavailable for the common Normal-Gamma model because the Student-t posterior predictive has an infinite moment-generating function.

Editorial extensions

If this is right

  • For any conjugate exponential family model, DRO-BASPE is exactly a KL-DRO problem with nominal $P_{\hat\eta}$ and radius $\epsilon - G$, so existing KL-DRO theory and solvers apply directly after the Bayesian correction.
  • With a linear objective $f_x(\xi)=\xi^\top x$ and a Gaussian likelihood, the worst-case risk equals $\hat\mu^\top x + \sqrt{2(\epsilon-G)}\,\sqrt{x^\top \hat\Sigma x}$, a convex closed form with $O(D)$ variables and $O(D^2)$ input that needs no sampling.
  • DRO-BASPP's dual is also single-stage whenever the posterior predictive satisfies the finite-moment-generating-function property; otherwise, as with a Student-t predictive, the implemented sample-average approximation solves a finite-sample surrogate rather than the exact primal problem.
  • For well-specified models the optimal tolerance $\epsilon^*_{\mathrm{PE}}(n)$ has a closed form, and choosing $\epsilon \ge \epsilon^*_{\mathrm{PE}}(n)$ guarantees the data-generating process lies in the ambiguity set, so the worst-case risk upper-bounds the true risk.
  • Because $A_\epsilon(\Pi) \subseteq B_\epsilon(P_n)$ for fixed $\epsilon$, DRO-BASPE is less conservative than DRO-BASPP whenever both ambiguity sets contain the data-generating process.

Reading between the lines

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

  • Extension: the $G$ correction can be read as an automatic small-sample adjustment—posterior spread inflates $G$ and shrinks the usable radius, so practitioners may need cross-validation-based $\epsilon$ selection precisely in the data-poor regime where Bayesian robustness matters most.
  • Extension: since DRO-BASPE is literally a KL ball around the posterior-mean member, any future data-driven radius calibration for classical KL-DRO transfers to the Bayesian setting once $G$ is subtracted; this gives a ready-made roadmap for finite-sample coverage guarantees.
  • Extension: for heavy-tailed posterior predictives (Student-t from Normal-Gamma, Lomax from Gamma-exponential), the paper's own examples show DRO-BASPP violates the dual's key condition, so DRO-BASPE is the variant with a provable exact dual; this predicts that BASPE will be the safer choice in such models.
  • Extension: extending the decomposition to general $\phi$-divergences requires a closed-form posterior expectation of the divergence; the KL case works because convexity of the log-partition makes $G$ tractable, suggesting divergences with affine or quadratic score functions as the next feasible class.
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

2 major / 4 minor

Summary. The paper introduces Distributionally Robust Optimisation with Bayesian Ambiguity Sets (DRO-BAS), proposing two posterior-informed ambiguity sets: one based on the posterior predictive distribution (BASPP) and one based on the posterior expected KL divergence to the model family (BASPE). For BASPE restricted to conjugate exponential family models, the paper proves Lemma 3.5, which decomposes the expected posterior KL divergence into a KL divergence to the posterior-mean model plus a nonnegative gap G(τ,ν), and Theorem 3.6, which establishes a strong dual representation of the worst-case risk as a single-stage stochastic program centered at the posterior-mean model with radius reduced by G(τ,ν). For BASPP, Proposition 3.2 gives a standard KL-ball dual under a finite moment-generating function assumption on the objective with respect to the posterior predictive. The paper evaluates both formulations on the Newsvendor and Portfolio problems, reporting empirical Pareto dominance over Bayesian DRO and faster solve times on the Portfolio problem.

Significance. If the central BASPE result holds, it is a valuable theoretical reduction: it connects posterior-informed ambiguity sets to classical KL-based DRO, yields a closed-form solution for linear objectives with Gaussian likelihoods, and provides a single-stage convex program in the exponential family. The paper also gives an explicit nonnegative correction G(τ,ν) arising from posterior uncertainty, which is a useful and non-obvious quantity. The claimed computational advantages over two-stage Bayesian DRO are plausible and supported by the reported timings. However, the paired-contribution claim is weakened by the BASPP side: the exact dual in Proposition 3.2 is inapplicable to the very conjugate models showcased in the experiments, and the implemented BASPP is an SAA surrogate whose fidelity to the original problem is not quantified.

major comments (2)
  1. [§3.1, Proposition 3.2, and §4.1–4.2] The exact dual for DRO-BASPP requires Property 3.1, i.e., finiteness of the moment-generating function of the objective under the posterior predictive P_n. For the two conjugate models used in the experiments (Normal likelihood with Normal-Gamma prior, giving a Student-t predictive in Appendix A.1, and Exponential likelihood with Gamma prior, giving a Lomax predictive in Appendix A.2), P_n has algebraic tails, so for the piecewise-linear Newsvendor objective E_{P_n}[exp(t f_x)] is infinite for every t>0. The paper acknowledges this in Section 3.1 and replaces P_n with the empirical measure of M posterior-predictive draws, then applies the dual to the resulting SAA problem. However, the SAA ambiguity set B_epsilon(hat P_{n,M}) is not the BASPP set B_epsilon(P_n), and the paper provides no error bound, consistency result, or convergence guarantee showing that the SAA-optimal value or decision is a faithful surrogate for the original DRO-BASPP objective in (6). Consequently, the BASPP experimental curves in Figures 2–3 and the abstract-level statement that both formulations admit strong duals are not supported in the showcased settings; the evaluated BASPP is a heuristic SAA surrogate rather than an exact solution of the proposed problem. This does not invalidate Theorem 3.6 for BASPE, but it undercuts the paired-contribution claim and needs to be addressed, either by proving SAA guarantees for this setting or by explicitly qualifying the BASPP claims.
  2. [Appendix B.4] In the strict-feasibility argument for Theorem 3.6, the text states that for Q̂ = p(ξ|η̂), E_{θ∼Π}[d_KL(Q̂∥P_θ)] = 0 < ϵ − G(τ,ν). This is incorrect: Lemma 3.5 gives E_{θ∼Π}[d_KL(Q̂∥P_θ)] = d_KL(Q̂, P_η̂) + G(τ,ν) = G(τ,ν). The conclusion of strict feasibility when ϵ > G(τ,ν) remains valid, so this is a proof-writing error rather than a substantive gap, but it should be corrected because the displayed equality is false.
minor comments (4)
  1. [§3.5] The sentence 'does not allow an exact dual formulation in Theorem 3.2' refers to Proposition 3.2, not Theorem 3.2; the numbering should be fixed.
  2. [§3.6] In the expression for the BASPE worst-case distribution, the functions p̃(ξ) and the subsequent density use 'fx(ξ′)' and 'fx(ξ)' inconsistently, and the normalization is written informally; please clarify the notation.
  3. [Figure 7 caption] The caption reads 'The Portfoliocumulativereturn' and is missing a space; also consider defining G in the legend for DRO-BASPE entries, since G is not introduced in the figure caption.
  4. [Corollary A.2] The displayed expression for η̂ contains an extra κ in the denominator of the second component (…,−κ(κ+1)/(4βκ)); this appears to be a typo and should be cleaned up.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Theorem 3.6 reduces a posterior-expected KL set to a standard KL ball via a derived identity, and the duals are external results.

full rationale

The paper's central derivation chain is self-contained. Lemma 3.5 proves, for conjugate exponential families, that E_{theta~Pi}[dKL(Q||P_theta)] = dKL(Q||P_hat_eta)+G, with G computed from posterior log-partition derivatives and non-negativity following from convexity of A (external Brown 1986). Theorem 3.6 substitutes this identity into the Lagrangian, so the BASPE worst-case risk is exactly the Hu-Hong dual centered at P_hat_eta with radius epsilon-G. No parameter in this chain is fitted to the quantity being predicted; hat_eta and G are posterior summaries, and epsilon is a user choice, with the closed-form epsilon* presented only as guidance and replaced by cross-validation in Section E.4. Proposition 3.2 for BASPP is the external Hu-Hong dual, and the paper explicitly acknowledges that Property 3.1 fails for Student-t and Lomax posterior predictives, so it does not claim an exact dual in those cases. That is a correctness/scope limitation, not circularity. No load-bearing self-citation or renamed known result appears; the core reduction is a genuinely derived equivalence rather than an assumed one.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The framework introduces no new entities. Its theoretical claims rest on classical KL-DRO duality and standard exponential-family identities; the new quantity G(tau, nu) is derived, not postulated. The hand-set inputs are the ambiguity radius epsilon and the experimental priors.

free parameters (2)
  • Ambiguity radius epsilon = grid 0.001 to 1.0, or cross-validated
    Controls the size of the KL ambiguity set. The paper derives epsilon*_PE(n) in the well-specified case using the unknown true DGP, but experiments sweep epsilon over a grid or select it by cross-validation (Sections 3.4 and E.4).
  • Prior hyperparameters = e.g., Normal-Gamma (0,1,1,1), NIW with mu=0, iota=D+1, kappa=iota+D+2, Psi=I
    Hand-chosen weakly informative priors in the experiments (Appendix E.1). These are inputs to the method, not fitted, but they influence the posterior and hence the ambiguity set.
assumptions (6)
  • domain assumption Conjugate exponential family likelihood with prior/posterior of the form in Definition 3.4, with natural parameter space Omega = {eta : A(eta) < infinity}.
    Restricts the DRO-BASPE strong-duality result to this model class (Section 3.3).
  • domain assumption Property 3.1: f_x has a finite moment-generating function under the nominal distribution (P_n for BASPP, P_eta_hat for BASPE).
    Needed for the KL-dual to be well-defined; the paper notes it fails for Student-t and Lomax predictives (Sections 3.1 and 3.5).
  • domain assumption Posterior density is differentiable in the hyperparameters (nu, tau) and differentiation can interchange with integration (Lemma 3.5).
    Technical regularity required for the closed-form expected KL identity (Appendix B.3).
  • standard math Log-partition function A(eta) of the exponential family is convex.
    Used to prove G(tau, nu) >= 0 and the Jensen step in Lemma 3.5 (Brown, 1986, Theorem 1.13).
  • standard math Strong duality for KL-divergence DRO with a fixed nominal distribution (Hu and Hong, 2013).
    Used as the external benchmark dual in Proposition 3.2 and Theorem 3.6 (Appendix B.1).
  • standard math Convex conjugate of the KL divergence (Agrawal and Horel, 2021, Lemma B.2).
    Gives the exponential form of the dual objective (Appendix B.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decision Making under the Exponential Family: Distributionally Robust Optimisation with Bayesian Ambiguity Sets." pith.science (2026). https://pith.science/paper/RBBV3WD4

@misc{pith2026241116829,
  author       = {Pith},
  title        = {Pith review of: Decision Making under the Exponential Family: Distributionally Robust Optimisation with Bayesian Ambiguity Sets},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RBBV3WD4}},
  note         = {Machine review of arXiv:2411.16829}
}
read the original abstract

Decision making under uncertainty is challenging as the data-generating process (DGP) is often unknown. Bayesian inference proceeds by estimating the DGP through posterior beliefs on the model's parameters. However, minimising the expected risk under these beliefs can lead to suboptimal decisions due to model uncertainty or limited, noisy observations. To address this, we introduce Distributionally Robust Optimisation with Bayesian Ambiguity Sets (DRO-BAS) which hedges against model uncertainty by optimising the worst-case risk over a posterior-informed ambiguity set. We provide two such sets, based on posterior expectations (DRO-BAS(PE)) or posterior predictives (DRO-BAS(PP)) and prove that both admit, under conditions, strong dual formulations leading to efficient single-stage stochastic programs which are solved with a sample average approximation. For DRO-BAS(PE) this covers all conjugate exponential family members while for DRO-BAS(PP) this is shown under conditions on the predictive's moment generating function. Our DRO-BAS formulations outperform existing Bayesian DRO on the Newsvendor problem and achieve faster solve times with comparable robustness on the Portfolio problem.

Figures

Figures reproduced from arXiv: 2411.16829 by the authors.

Figure 1
Figure 1. Visualisation of our BASPE and BASPP ambiguity sets versus BDRO (Shapiro et al., 2023) for Normal likelihood with Normal-Gamma posterior. BASPE contains distributions with ex￾pected (under the posterior Π) KL divergence from the model of at most ϵ. BASPP contains distributions with KL divergence at most ϵ from the posterior predictive distribution. BDRO minimises the expected worst-case risk between each KL-based am… view at source ↗
Figure 2
Figure 2. The out-of-sample mean-variance tradeoff for exponential DGP (top row), 5D multivariate normal DGP (middle row), and truncated normal DGP (bottom row). We vary ϵ when the total number of SAA samples is 25 (left), 100 (middle), and 900 (right) and plot the markers in bold. For illustration purposes, blurred markers/lines show the other smaller values of M; for example, the right column shows M = 900 in bold and M = 2… view at source ↗
Figure 3
Figure 3. Mean-variance tradeoff of the Portfolio objective across all OOS time windows. Markers are filled similarly to [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The out-of-sample mean-variance tradeoff for normal DGP (top row) and Contaminated Exponential DGP (bottom row). We vary ϵ for DRO-BASPE, DRO-BASPP, and BDRO when the total number of samples from the model is 25 (left), 100 (middle), and 900 (right) and plot the marker…
Figure 5
Figure 5. Figure 5: The out-of-sample mean-variance tradeoff for Exponential DGP when the total number of observations n is 5 (left), 20 (middle), and 100 (right). For illustration purposes, blurred markers/lines show the previous two values of n; for example, the right column shows n = 1…
Figure 6
Figure 6. Figure 6: The first three plots showcase the mean solution (decision variable x) with the associated variance (error bars) obtained by the methods applied to the Newsvendor problem with a Normal DGP for M = 25, 100, 900. The last plot shows the true risk (expected cost under the…
Figure 8
Figure 8. Figure 8: Comparison of the OOS mean-variance tradeoff on the Newsvendor for Empirical KL DRO and Empirical Wasserstein DRO vs Bayesian model-based DRO approaches on the Newsvendor Problem with Normal DGP (top) and Truncated Normal DGP (bottom). The results in [PITH_FULL_IMAGE:…
Figure 9
Figure 9. Figure 9: The cross-validation (CV) mean-variance tradeoff on the Newsvendor Problem with Normal DGP and n = 100 training observations. We highlight three ways the decision maker can choose ϵ: firstly, minimise the CV-mean (cross markers); secondly, minimise the CV-var (star mar…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 24 canonical work pages

  1. [1]

    A rewriting system for convex optimization problems

    Agrawal, A., Verschueren, R., Diamond, S., and Boyd, S. A rewriting system for convex optimization problems. Journal of Control and Decision, 5 0 (1): 0 42--60, 2018

    2018

  2. [2]

    and Horel, T

    Agrawal, R. and Horel, T. Optimal bounds between f-divergences and integral probability metrics. Journal of Machine Learning Research, 22 0 (128): 0 1--59, 2021

    work page 2021

  3. [3]

    Bayesian Nonparametrics Meets Data-Driven Distributionally Robust Optimization

    Bariletto, N. and Ho, N. Bayesian nonparametrics meets data-driven robust optimization. arXiv preprint arXiv:2401.15771, 2024

    work page Pith review arXiv 2024

  4. [4]

    Data-Driven DRO and Economic Decision Theory: An Analytical Synthesis With Bayesian Nonparametric Advancements

    Bariletto, N., Nguyen, K., and Ho, N. Borrowing strength in distributionally robust optimization via hierarchical dirichlet processes. arXiv preprint arXiv:2405.13160, 2024

    work page Pith review arXiv 2024

  5. [5]

    and Love, D

    Bayraksan, G. and Love, D. K. Data-driven stochastic programming using phi-divergences. In The operations research revolution, pp.\ 1--19. INFORMS, 2015

    work page 2015

  6. [6]

    Data-driven robust optimization

    Bertsimas, D., Gupta, V., and Kallus, N. Data-driven robust optimization. Mathematical Programming, 167: 0 235--292, 2018

    work page 2018

  7. [7]

    Brown, L. D. Fundamentals of statistical exponential families with applications in statistical decision theory. Lecture Notes-Monograph Series, 9: 0 i--279, 1986. ISSN 07492170. URL http://www.jstor.org/stable/4355554

    arXiv 1986

  8. [8]

    Real-world datasets for portfolio selection and solutions of some stochastic dominance portfolio models

    Bruni, R., Cesarone, F., Scozzari, A., and Tardella, F. Real-world datasets for portfolio selection and solutions of some stochastic dominance portfolio models. Data in brief, 8: 0 858--862, 2016

    work page 2016

Show all 33 references
  1. [9]

    Uncertainty averse preferences

    Cerreia-Vioglio, S., Maccheroni, F., Marinacci, M., and Montrucchio, L. Uncertainty averse preferences. Journal of Economic Theory, 146 0 (4): 0 1275--1330, 2011

    2011

  2. [10]

    and Ylvisaker, D

    Diaconis, P. and Ylvisaker, D. Conjugate priors for exponential families. The Annals of statistics, pp.\ 269--281, 1979

    1979

  3. [11]

    and Boyd, S

    Diamond, S. and Boyd, S. CVXPY : A P ython-embedded modeling language for convex optimization. Journal of Machine Learning Research, 17 0 (83): 0 1--5, 2016

    2016

  4. [12]

    and Namkoong, H

    Duchi, J. and Namkoong, H. Variance-based regularization with convex objectives. Journal of Machine Learning Research, 20 0 (68): 0 1--55, 2019

    2019

  5. [13]

    A collection of identities for variational inference with exponential-family models

    Endres, D., Pabst, K., Eckert, A.-L., and Schween, R. A collection of identities for variational inference with exponential-family models. Technical report, Philipps-Universit\" a t Marburg, 2022

    2022

  6. [14]

    and Kleywegt, A

    Gao, R. and Kleywegt, A. Distributionally robust stochastic optimization with W asserstein distance. Mathematics of Operations Research, 48 0 (2): 0 603--655, 2023

    2023

  7. [15]

    J., and Lim, A

    Gotoh, J., Kim, M. J., and Lim, A. E. B. Calibration of distributionally robust empirical optimization models. Operations Research, 69 0 (5): 0 1630--1650, 2021. doi:10.1287/opre.2020.2041. URL https://doi.org/10.1287/opre.2020.2041

    2021

  8. [16]

    Near-optimal bayesian ambiguity sets for distributionally robust optimization

    Gupta, V. Near-optimal bayesian ambiguity sets for distributionally robust optimization. Management Science, 65 0 (9): 0 4242--4260, 2019

    2019

  9. [17]

    Moments for the canonical parameter of an exponential family under a conjugate distribution

    Gutierrez-Pena, E. Moments for the canonical parameter of an exponential family under a conjugate distribution. Biometrika, 84 0 (3): 0 727--732, 1997

    1997

  10. [18]

    M., Consonni, G., Veronese, P., George, E., Gir \'o n, F., Mart \' nez, M., Letac, G., and Morris, C

    Guti \'e rrez-Pe \ n a, E., Smith, A., Bernardo, J. M., Consonni, G., Veronese, P., George, E., Gir \'o n, F., Mart \' nez, M., Letac, G., and Morris, C. N. Exponential and bayesian conjugate families: review and extensions. Test, 6: 0 1--90, 1997

    1997

  11. [19]

    and Hong, L

    Hu, Z. and Hong, L. J. Kullback- L eibler divergence constrained distributionally robust optimization. Available at Optimization Online, 1 0 (2): 0 9, 2013

    2013

  12. [20]

    Hedging against complexity: D istributionally robust optimization with parametric approximation

    Iyengar, G., Lam, H., and Wang, T. Hedging against complexity: D istributionally robust optimization with parametric approximation. In International Conference on Artificial Intelligence and Statistics, pp.\ 9976--10011. PMLR, 2023

    2023

  13. [21]

    M., Nguyen, V

    Kuhn, D., Esfahani, P. M., Nguyen, V. A., and Shafieezadeh-Abadeh, S. Wasserstein distributionally robust optimization: Theory and applications in machine learning. In Operations Research & Management Science in the Age of Analytics, chapter 6, pp.\ 130--166. Informs, 2019. do...

    2019

  14. [22]

    Portfolio selection

    Markowitz, H. Portfolio selection. The Journal of Finance, 7 0 (1): 0 77--91, 1952. doi:https://doi.org/10.1111/j.1540-6261.1952.tb01525.x. URL https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1540-6261.1952.tb01525.x

    1952

  15. [23]

    Modeling the second player in distributionally robust optimization

    Michel, P., Hashimoto, T., and Neubig, G. Modeling the second player in distributionally robust optimization. arXiv preprint arXiv:2103.10282, 2021

    2021 arXiv

  16. [24]

    Distributionally robust models with parametric likelihood ratios

    Michel, P., Hashimoto, T., and Neubig, G. Distributionally robust models with parametric likelihood ratios. arXiv preprint arXiv:2204.06340, 2022

    2022 arXiv

  17. [25]

    Murphy, K. P. Probabilistic machine learning: Advanced topics. MIT press, 2023

    2023

  18. [26]

    Petersen, K. B. and Pedersen, M. S. The matrix cookbook, October 2008. URL http://www2.imm.dtu.dk/pubdb/p.php?3274. Version 20081110

    2008

  19. [27]

    Lectures on stochastic programming: modeling and theory

    Shapiro, A., Dentcheva, D., and Ruszczynski, A. Lectures on stochastic programming: modeling and theory. SIAM, 2021

    2021

  20. [28]

    Bayesian distributionally robust optimization

    Shapiro, A., Zhou, E., and Lin, Y. Bayesian distributionally robust optimization. SIAM Journal on Optimization, 33 0 (2): 0 1279--1304, 2023

    2023

  21. [29]

    and Jegelka, S

    Staib, M. and Jegelka, S. Distributionally robust optimization and generalization in kernel methods. Advances in Neural Information Processing Systems, 32, 2019

    2019

  22. [30]

    Walker, S. G. Bayesian inference with misspecified models. Journal of statistical planning and inference, 143 0 (10): 0 1621--1633, 2013

    2013

  23. [31]

    Learning against distributional uncertainty: On the trade-off between robustness and specificity

    Wang, S., Wang, H., and Honorio, J. Learning against distributional uncertainty: On the trade-off between robustness and specificity. arXiv preprint arXiv:2301.13565, 2023

    2023 arXiv

  24. [32]

    A B ayesian risk approach to data-driven stochastic optimization: Formulations and asymptotics

    Wu, D., Zhu, H., and Zhou, E. A B ayesian risk approach to data-driven stochastic optimization: Formulations and asymptotics. SIAM Journal on Optimization, 28 0 (2): 0 1588--1612, 2018

    2018

  25. [33]

    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 gl...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.