pith. sign in

arxiv: 2502.16607 · v2 · submitted 2025-02-23 · 🧮 math.OC

Risk-averse Decision Making with Contextual Information: Model, Sample Average Approximation, and Kernelization

Yuan Tao , Erick Delage , Huifu Xu This is my paper

Pith reviewed 2026-05-23 02:21 UTC · model grok-4.3

classification 🧮 math.OC
keywords risk-averse optimizationcontextual uncertaintysample average approximationkernel methodsrisk measuresstochastic programmingnewsvendor problemportfolio selection
0
0 comments X

The pith

The nested risk minimization over problem data uncertainty given context is equivalent to joint risk minimization over both uncertainties when risk measures are chosen appropriately.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper studies risk-averse decisions under two uncertainties: problem data uncertainty and the contextual uncertainty that accompanies it. The decision maker first picks a policy that minimizes risk from the data uncertainty conditional on an observed context, then evaluates the risk of that policy over the distribution of contexts. The central question is whether this two-step process produces the same decisions and values as minimizing a single risk measure defined jointly over both uncertainties at once. Equivalence holds when the risk measures for each layer are paired correctly, and under additional conditions the resulting policies do not change when the contextual risk measure is replaced by another. The equivalence converts the original nested problem into an ordinary one-stage risk minimization problem that can be solved by sample average approximation, and the same reduction yields consistent approximations when the problem is lifted into a reproducing kernel Hilbert space.

Core claim

Appropriate choices of risk measures make the nested contextual risk process equivalent to joint risk minimization over problem data uncertainty and contextual uncertainty simultaneously; under further conditions the optimal policies are independent of the particular risk measure chosen for the contextual uncertainty.

What carries the argument

The mapping between a nested risk structure (minimize risk of PDU conditional on CU, then assess risk of that policy over CU) and a joint risk measure over both uncertainties, realized by specific pairings of risk measures.

If this is right

  • The nested contextual problem reduces to an ordinary one-stage risk minimization problem that admits direct sample average approximation.
  • Under the stated conditions the optimal decision does not depend on which risk measure is used to evaluate contextual uncertainty.
  • Consistency of optimal values and policies holds for sample average approximations when the problem is posed in a reproducing kernel Hilbert space.
  • The same equivalence supplies tractable formulations for standard risk measures in newsvendor and portfolio selection problems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The reduction technique could apply to other multi-layer risk problems in stochastic programming whenever the inner and outer risk measures can be paired appropriately.
  • Policy independence from the contextual risk measure would simplify modeling whenever context is observed but its risk cannot be quantified precisely.
  • Kernel consistency results may extend to other nonparametric representations of contextual uncertainty beyond the reproducing kernel Hilbert space setting examined here.

Load-bearing premise

The decision process must be structured as first minimizing risk from problem data uncertainty conditional on a contextual observation and only afterward assessing the risk of the resulting policy against contextual uncertainty.

What would settle it

An explicit pair of risk measures together with a joint distribution over PDU and CU such that the policy obtained from the nested process differs from every policy obtained by joint risk minimization, or a concrete instance in which the optimal policy changes when the contextual risk measure is altered while the paper's stated conditions still hold.

Figures

Figures reproduced from arXiv: 2502.16607 by Erick Delage, Huifu Xu, Yuan Tao.

Figure 1
Figure 1. Figure 1: The policies for newsvendor problem under different risk measures. [PITH_FULL_IMAGE:figures/full_fig_p024_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Relative distance of conditional expected loss, conditional CVaR, and conditional [PITH_FULL_IMAGE:figures/full_fig_p024_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Optimal policy and SAA policies (sample size [PITH_FULL_IMAGE:figures/full_fig_p025_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: In risk-averse contextual optimization problem with expected CVaR, relative average [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The policies under different sample sizes [PITH_FULL_IMAGE:figures/full_fig_p054_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Relative distances of policies compared to optimal policy under different sample sizes [PITH_FULL_IMAGE:figures/full_fig_p054_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Optimal policy and SAA policies (sample size N = 800) for risk-averse contextual [PITH_FULL_IMAGE:figures/full_fig_p054_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: In risk-averse contextual optimization problem with entropic risk measure, relative [PITH_FULL_IMAGE:figures/full_fig_p055_8.png] view at source ↗
read the original abstract

We consider risk-averse contextual optimization problems where the decision maker (DM) faces two types of uncertainties: problem data uncertainty (PDU) and contextual uncertainty (CU) associated with PDU, the DM makes an optimal decision by minimizing the risk arising from PDU based on the present observation of CU and then assesses the risk of the optimal policy against the CU. A natural question arises as to whether the nested risk minimization/assessment process is equivalent to joint risk minimization/assessment against CU and PDU simultaneously. First, we demonstrate that the equivalence can be established by appropriate choices of the risk measures and give counter examples where such equivalence may fail. One of the interesting findings is that the optimal policies are independent of the choice of the risk measure against the CU under certain conditions. Second, by using the equivalence, we propose computational method for solving the risk-averse contextual optimization problem by solving a one-stage risk minimization problem. The latter is particularly helpful in data-driven environments. We consider a number of risk measures/metrics to characterize the DM's risk preference for PDU and discuss the computational tractability for the resulting risk-averse contextual optimization problem. Third, when the risk-averse contextual optimization problem is defined in the reproducing kernel Hilbert space, we show consistency of the optimal values obtained from solving sample average approximation problems. Some numerical tests, in newsvendor problem and portfolio selection problem, are performed to validate the theoretical results.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The paper studies risk-averse contextual optimization problems with two uncertainty sources: problem data uncertainty (PDU) and contextual uncertainty (CU). It analyzes when the natural nested procedure (minimize PDU risk conditional on a CU observation, then evaluate the resulting policy under CU risk) is equivalent to joint risk minimization over both uncertainties simultaneously. Equivalence is shown to hold for suitable choices of risk measures, with counterexamples when it fails; under stated conditions the optimal policy is independent of the CU risk measure. The equivalence is used to reduce the problem to a single-stage risk minimization, whose tractability is examined for several risk measures/metrics. Consistency of sample-average approximations is proved when the problem is posed in a reproducing kernel Hilbert space, and the results are illustrated numerically on newsvendor and portfolio instances.

Significance. If the equivalence and independence results hold, the work supplies a principled modeling framework and a practical reduction to one-stage problems that is especially useful in data-driven settings. The RKHS consistency result and the explicit numerical validation on standard test problems add concrete value for stochastic optimization under risk aversion.

minor comments (3)
  1. [Abstract / §1] The abstract states that equivalence holds 'by appropriate choices of the risk measures' but does not name the classes (e.g., coherent, convex, distortion) for which the result is proved; a brief enumeration in the introduction would help readers locate the precise statements.
  2. [RKHS consistency section] The consistency theorem for the RKHS formulation should explicitly list the assumptions on the kernel and the risk measure (e.g., continuity, Lipschitz properties) that are used to pass to the limit; these are alluded to but not itemized in the provided description.
  3. [Numerical experiments] In the numerical section, report the number of SAA samples, the kernel bandwidth selection procedure, and the out-of-sample evaluation protocol so that the observed convergence can be reproduced.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The report accurately captures the paper's contributions on equivalence between nested and joint risk measures, policy independence under stated conditions, reduction to one-stage problems, and RKHS consistency. No specific major comments appear in the provided report, so we have no point-by-point responses to address.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper derives equivalence between nested (conditional on CU then risk on PDU) and joint risk minimization for specific choices of risk measures, with explicit counterexamples for failure cases. This is presented as a mathematical result grounded in properties of standard risk measures (e.g., those discussed for PDU). The subsequent reduction to a one-stage problem follows directly from the established equivalence rather than assuming it. Consistency of SAA in RKHS is shown via standard approximation arguments. No self-definitional loops, fitted inputs renamed as predictions, or load-bearing self-citations appear in the derivation chain; the independence of optimal policies under conditions is a derived finding, not an input. The structure is self-contained against external benchmarks of risk measure theory.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract does not introduce free parameters, new entities, or non-standard axioms; relies on established properties of risk measures and optimization theory.

pith-pipeline@v0.9.0 · 5789 in / 1094 out tokens · 54676 ms · 2026-05-23T02:21:12.223721+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

74 extracted references · 74 canonical work pages · 1 internal anchor

  1. [1]

    Spectral measures of risk: A coherent representation of subjective risk aver- sion

    Carlo Acerbi. Spectral measures of risk: A coherent representation of subjective risk aver- sion. Journal of Banking & Finance , 26(7):1505–1518, 2002

    work page 2002

  2. [2]

    Portfolio Optimization with Spectral Measures of Risk

    Carlo Acerbi and Prospero Simonetti. Portfolio optimization with spectral measures of risk. arXiv preprint cond-mat/0203607, 2002. 29

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  3. [3]

    Spectral risk measures and portfolio selection

    Alexandre Adam, Mohamed Houkari, and Jean-Paul Laurent. Spectral risk measures and portfolio selection. Journal of Banking & Finance , 32(9):1870–1882, 2008

    work page 2008

  4. [4]

    The big data newsvendor: Practical insights from machine learning

    Gah-Yi Ban and Cynthia Rudin. The big data newsvendor: Practical insights from machine learning. Operations Research, 67(1):90–108, 2019

    work page 2019

  5. [5]

    Generalization bounds for regularized portfolio selection with market side information

    Thierry Bazier-Matte and Erick Delage. Generalization bounds for regularized portfolio selection with market side information. INFOR: Information Systems and Operational Research, 58(2):374–401, 2020

    work page 2020

  6. [6]

    Expected utility, penalty functions, and duality in stochastic nonlinear programming

    Aharon Ben-Tal and Marc Teboulle. Expected utility, penalty functions, and duality in stochastic nonlinear programming. Management Science, 32(11):1445–1466, 1986

    work page 1986

  7. [7]

    An old-new concept of convex risk measures: The optimized certainty equivalent

    Aharon Ben-Tal and Marc Teboulle. An old-new concept of convex risk measures: The optimized certainty equivalent. Mathematical Finance, 17(3):449–476, 2007

    work page 2007

  8. [8]

    Topological Spaces: Including a Treatment of Multi-valued Functions, Vector Spaces and Convexity

    Claude Berge. Topological Spaces: Including a Treatment of Multi-valued Functions, Vector Spaces and Convexity. Macmillan, 1963

    work page 1963

  9. [9]

    From predictive to prescriptive analytics

    Dimitris Bertsimas and Nathan Kallus. From predictive to prescriptive analytics. Manage- ment Science, 66(3):1025–1044, 2020

    work page 2020

  10. [10]

    Data-driven optimization: A reproducing kernel hilbert space approach

    Dimitris Bertsimas and Nihal Koduri. Data-driven optimization: A reproducing kernel hilbert space approach. Operations Research, 70(1):454–471, 2022

    work page 2022

  11. [11]

    Convergence of probability measures

    Patrick Billingsley. Convergence of probability measures. John Wiley & Sons, 2013

    work page 2013

  12. [12]

    Theory of classification: A survey of some recent advances

    St´ ephane Boucheron, Olivier Bousquet, and G´ abor Lugosi. Theory of classification: A survey of some recent advances. ESAIM: probability and statistics , 9:323–375, 2005

    work page 2005

  13. [13]

    Elements of Mathematics: General Topology

    Nicolas Bourbaki. Elements of Mathematics: General Topology. Part 2 . Hermann, 1966

    work page 1966

  14. [14]

    End-to-end conditional robust optimization

    Abhilash Chenreddy and Erick Delage. End-to-end conditional robust optimization. arXiv preprint arXiv:2403.04670, 2024

    work page arXiv 2024

  15. [15]

    Data-driven conditional robust optimization

    Abhilash Reddy Chenreddy, Nymisha Bandi, and Erick Delage. Data-driven conditional robust optimization. Advances in Neural Information Processing Systems , 35:9525–9537, 2022

    work page 2022

  16. [16]

    Divergence utilities

    Alexander S Cherny and Michael Kupper. Divergence utilities. Available at SSRN 1023525, 2007

    work page 2007

  17. [17]

    Advancing stability analysis of mean-risk stochastic programs: Bilevel and two-stage models

    Matthias Claus. Advancing stability analysis of mean-risk stochastic programs: Bilevel and two-stage models. PhD thesis, Dissertation, Duisburg, Essen, Universit¨ at Duisburg-Essen, 2016, 2016

    work page 2016

  18. [18]

    Shortfall risk models when information on loss function is incomplete

    Erick Delage, Shaoyan Guo, and Huifu Xu. Shortfall risk models when information on loss function is incomplete. Operations Research, 70(6):3511–3518, 2022

    work page 2022

  19. [19]

    Optimal versus naive diversifi- cation: How inefficient is the 1/n portfolio strategy? The review of Financial studies , 22(5):1915–1953, 2009

    Victor DeMiguel, Lorenzo Garlappi, and Raman Uppal. Optimal versus naive diversifi- cation: How inefficient is the 1/n portfolio strategy? The review of Financial studies , 22(5):1915–1953, 2009. 30

    work page 1915

  20. [20]

    Task-based end-to-end model learning in stochastic optimization

    Priya Donti, Brandon Amos, and J Zico Kolter. Task-based end-to-end model learning in stochastic optimization. Advances in neural information processing systems , 30, 2017

    work page 2017

  21. [21]

    Generaliza- tion bounds in the predict-then-optimize framework

    Othman El Balghiti, Adam N Elmachtoub, Paul Grigas, and Ambuj Tewari. Generaliza- tion bounds in the predict-then-optimize framework. Mathematics of Operations Research, 48(4):2043–2065, 2023

    work page 2043

  22. [22]

    predict, then optimize

    Adam N Elmachtoub and Paul Grigas. Smart “predict, then optimize”. Management Science, 68(1):9–26, 2022

    work page 2022

  23. [23]

    Estimate-then- optimize versus integrated-estimation-optimization versus sample average approximation: A stochastic dominance perspective

    Adam N Elmachtoub, Henry Lam, Haofeng Zhang, and Yunfan Zhao. Estimate-then- optimize versus integrated-estimation-optimization versus sample average approximation: A stochastic dominance perspective. arXiv preprint arXiv:2304.06833 , 2023

    work page arXiv 2023

  24. [24]

    Stochastic finance: an introduction in discrete time

    Hans F¨ ollmer and Alexander Schied. Stochastic finance: an introduction in discrete time . Walter de Gruyter, 2011

    work page 2011

  25. [25]

    Newsvendor solutions via conditional value-at-risk min- imization

    Jun-ya Gotoh and Yuichi Takano. Newsvendor solutions via conditional value-at-risk min- imization. European Journal of Operational Research, 179(1):80–96, 2007

    work page 2007

  26. [26]

    Distributionally robust shortfall risk optimization model and its approximation

    Shaoyan Guo and Huifu Xu. Distributionally robust shortfall risk optimization model and its approximation. Mathematical Programming, 174(1):473–498, 2019

    work page 2019

  27. [27]

    Statistical robustness of empirical risks in machine learning

    Shaoyan Guo, Huifu Xu, and Liwei Zhang. Statistical robustness of empirical risks in machine learning. Journal of Machine Learning Research , 24(125):1–38, 2023

    work page 2023

  28. [28]

    Integrals, conditional expectations, and martingales of multivalued functions

    Fumio Hiai and Hisaharu Umegaki. Integrals, conditional expectations, and martingales of multivalued functions. Journal of Multivariate Analysis , 7(1):149–182, 1977

    work page 1977

  29. [29]

    Risk guarantees for end-to-end prediction and optimization processes

    Nam Ho-Nguyen and Fatma Kılın¸ c-Karzan. Risk guarantees for end-to-end prediction and optimization processes. Management Science, 68(12):8680–8698, 2022

    work page 2022

  30. [30]

    Robust statistics

    Peter J Huber and Elvezio M Ronchetti. Robust statistics. Wiley Series in Probability and Statistics, 2009

    work page 2009

  31. [31]

    Tight approximations of dynamic risk measures

    Dan A Iancu, Marek Petrik, and Dharmashankar Subramanian. Tight approximations of dynamic risk measures. Mathematics of Operations Research, 40(3):655–682, 2015

    work page 2015

  32. [32]

    Optimizer’s information criterion: Dis- secting and correcting bias in data-driven optimization

    Garud Iyengar, Henry Lam, and Tianyu Wang. Optimizer’s information criterion: Dis- secting and correcting bias in data-driven optimization. arXiv preprint arXiv:2306.10081 , 2023

    work page arXiv 2023

  33. [33]

    Stochastic optimization forests

    Nathan Kallus and Xiaojie Mao. Stochastic optimization forests. Management Science , 69(4):1975–1994, 2023

    work page 1975

  34. [34]

    Heteroscedasticity-aware residuals- based contextual stochastic optimization

    Rohit Kannan, G¨ uzin Bayraksan, and James Luedtke. Heteroscedasticity-aware residuals- based contextual stochastic optimization. arXiv preprint arXiv:2101.03139 , 2021

    work page arXiv 2021

  35. [35]

    Residuals-based distributionally robust optimization with covariate information

    Rohit Kannan, G¨ uzin Bayraksan, and James R Luedtke. Residuals-based distributionally robust optimization with covariate information. Mathematical Programming, 207(1):369– 425, 2024. 31

    work page 2024

  36. [36]

    Data-driven sample average ap- proximation with covariate information

    Rohit Kannan, G¨ uzin Bayraksan, and James R Luedtke. Data-driven sample average ap- proximation with covariate information. Operations Research, 2025

    work page 2025

  37. [37]

    Representation results for law invariant time consistent functions

    Michael Kupper and Walter Schachermayer. Representation results for law invariant time consistent functions. Mathematics and Financial Economics , 2(3):189–210, 2009

    work page 2009

  38. [38]

    On law invariant coherent risk measures

    Shigeo Kusuoka. On law invariant coherent risk measures. Advances in mathematical economics, pages 83–95, 2001

    work page 2001

  39. [39]

    Data-driven newsvendor problems regularized by a profit risk constraint

    Shaochong Lin, Youhua Chen, Yanzhi Li, and Zuo-Jun Max Shen. Data-driven newsvendor problems regularized by a profit risk constraint. Production and Operations Management , 31(4):1630–1644, 2022

    work page 2022

  40. [40]

    Newsvendor conditional value-at-risk minimisation: A feature-based approach under adaptive data selection

    Congzheng Liu and Wenqi Zhu. Newsvendor conditional value-at-risk minimisation: A feature-based approach under adaptive data selection. European Journal of Operational Research, 313(2):548–564, 2024

    work page 2024

  41. [41]

    Risk bounds and calibration for a smart predict-then-optimize method

    Heyuan Liu and Paul Grigas. Risk bounds and calibration for a smart predict-then-optimize method. Advances in Neural Information Processing Systems , 34:22083–22094, 2021

    work page 2021

  42. [42]

    Multistage utility preference robust optimization

    Jia Liu, Zhiping Chen, and Huifu Xu. Multistage utility preference robust optimization. arXiv preprint arXiv:2109.04789 , 2021

    work page arXiv 2021

  43. [43]

    Deep reinforcement learning for option pricing and hedging under dynamic expectile risk measures

    Saeed Marzban, Erick Delage, and Jonathan Yu-Meng Li. Deep reinforcement learning for option pricing and hedging under dynamic expectile risk measures. Quantitative Finance, 23(10):1411–1430, 2023

    work page 2023

  44. [44]

    Universal kernels

    Charles A Micchelli, Yuesheng Xu, and Haizhang Zhang. Universal kernels. Journal of Machine Learning Research, 7(12), 2006

    work page 2006

  45. [45]

    Robustifying conditional portfolio decisions via optimal transport

    Viet Anh Nguyen, Fan Zhang, Shanshan Wang, Jos´ e Blanchet, Erick Delage, and Yinyu Ye. Robustifying conditional portfolio decisions via optimal transport. Operations Research, 2024

    work page 2024

  46. [46]

    On convergence of kernel learning estimators

    Vladimir I Norkin and Michiel A Keyzer. On convergence of kernel learning estimators. SIAM Journal on Optimization , 20(3):1205–1223, 2010

    work page 2010

  47. [47]

    Dual stochastic dominance and related mean-risk models

    WLodzimierz Ogryczak and Andrzej Ruszczynski. Dual stochastic dominance and related mean-risk models. SIAM Journal on Optimization , 13(1):60–78, 2002

    work page 2002

  48. [48]

    Risk aversion in the small and in the large

    John W Pratt. Risk aversion in the small and in the large. In Uncertainty in economics, pages 59–79. Elsevier, 1978

    work page 1978

  49. [49]

    Integrated condi tional estimation-optimization

    Meng Qi, Paul Grigas, and Zuo-Jun Max Shen. Integrated conditional estimation- optimization. arXiv preprint arXiv:2110.12351 , 2021

    work page arXiv 2021

  50. [50]

    Data-driven approximation of contextual chance-constrained stochastic programs

    Hamed Rahimian and Bernardo Pagnoncelli. Data-driven approximation of contextual chance-constrained stochastic programs. SIAM Journal on Optimization , 33(3):2248–2274, 2023. 32

    work page 2023

  51. [51]

    Contextual stochastic programs with expected-value constraints

    Hamed Rahimian and Bernardo Pagnoncelli. Contextual stochastic programs with expected-value constraints. Optimization Online, 2024

    work page 2024

  52. [52]

    Generalized deviations in risk analysis

    R Tyrrell Rockafellar, Stan Uryasev, and Michael Zabarankin. Generalized deviations in risk analysis. Finance and Stochastics, 10:51–74, 2006

    work page 2006

  53. [53]

    Optimization of conditional value-at-risk

    R Tyrrell Rockafellar, Stanislav Uryasev, et al. Optimization of conditional value-at-risk. Journal of risk , 2:21–42, 2000

    work page 2000

  54. [54]

    Functional Analysis: Second Edition

    Walter Rudin. Functional Analysis: Second Edition . McGraw-Hill, 1991

    work page 1991

  55. [55]

    Optimization of convex risk functions.Math- ematics of operations research, 31(3):433–452, 2006

    Andrzej Ruszczy´ nski and Alexander Shapiro. Optimization of convex risk functions.Math- ematics of operations research, 31(3):433–452, 2006

    work page 2006

  56. [56]

    A survey of contextual optimization methods for decision-making under uncertainty

    Utsav Sadana, Abhilash Chenreddy, Erick Delage, Alexandre Forel, Emma Frejinger, and Thibaut Vidal. A survey of contextual optimization methods for decision-making under uncertainty. European Journal of Operational Research, 2024

    work page 2024

  57. [57]

    A generalized representer theorem

    Bernhard Sch¨ olkopf, Ralf Herbrich, and Alex J Smola. A generalized representer theorem. In International conference on computational learning theory , pages 416–426. Springer, 2001

    work page 2001

  58. [58]

    Regulariza- tion via mass transportation

    Soroosh Shafieezadeh-Abadeh, Daniel Kuhn, and Peyman Mohajerin Esfahani. Regulariza- tion via mass transportation. Journal of Machine Learning Research , 20(103):1–68, 2019

    work page 2019

  59. [59]

    Interchangeability principle and dynamic equations in risk averse stochastic programming

    Alexander Shapiro. Interchangeability principle and dynamic equations in risk averse stochastic programming. Operations Research Letters, 45(4):377–381, 2017

    work page 2017

  60. [60]

    Lectures on stochastic programming: Modeling and theory, 2014

    Alexander Shapiro, Darinka Dentcheva, and Andrzej Ruszczy´ nski. Lectures on stochastic programming: Modeling and theory, 2014

    work page 2014

  61. [61]

    Stochastic mathematical programs with equilibrium constraints, modelling and sample average approximation

    Alexander Shapiro and Huifu Xu. Stochastic mathematical programs with equilibrium constraints, modelling and sample average approximation. Optimization, 57(3):395–418, 2008

    work page 2008

  62. [62]

    Universality, charac- teristic kernels and rkhs embedding of measures

    Bharath K Sriperumbudur, Kenji Fukumizu, and Gert RG Lanckriet. Universality, charac- teristic kernels and rkhs embedding of measures. Journal of Machine Learning Research , 12(7), 2011

    work page 2011

  63. [63]

    On data- driven prescriptive analytics with side information: A regularized nadaraya-watson ap- proach

    Prateek R Srivastava, Yijie Wang, Grani A Hanasusanto, and Chin Pang Ho. On data- driven prescriptive analytics with side information: A regularized nadaraya-watson ap- proach. arXiv preprint arXiv:2110.04855 , 2021

    work page arXiv 2021

  64. [64]

    Support Vector Machine

    Ingo Steinwart and Christmann Andreas. Support Vector Machine . Springer, New York, 2008

    work page 2008

  65. [65]

    Distortion riskmetrics on general spaces

    Qiuqi Wang, Ruodu Wang, and Yunran Wei. Distortion riskmetrics on general spaces. ASTIN Bulletin: The Journal of the IAA , 50(3):827–851, 2020

    work page 2020

  66. [66]

    Robust contextual portfolio opti- mization with gaussian mixture models

    Yijie Wang, Grani A Hanasusanto, and Chin Pang Ho. Robust contextual portfolio opti- mization with gaussian mixture models. Optimization Online, 2022. 33

    work page 2022

  67. [67]

    Generalization bounds for contextual stochastic optimization using kernel regression

    Yijie Wang, Grani A Hanasusanto, and Chin Pang Ho. Generalization bounds for contextual stochastic optimization using kernel regression. arXiv preprint arXiv:2407.10764 , 2024

    work page arXiv 2024

  68. [68]

    Optimal robust policy for feature-based newsvendor

    Luhao Zhang, Jincheng Yang, and Rui Gao. Optimal robust policy for feature-based newsvendor. Management Science, 70(4):2315–2329, 2024

    work page 2024

  69. [69]

    Statistical robustness of kernel learning estimator with respect to data perturbation

    Sainan Zhang, Huifu Xu, and Hailin Sun. Statistical robustness of kernel learning estimator with respect to data perturbation. arXiv preprint arXiv:2406.10555 , 2024

    work page arXiv 2024

  70. [70]

    Data-driven piecewise affine decision rules for stochastic programming with covariate information

    Yiyang Zhang, Junyi Liu, and Xiaobo Zhao. Data-driven piecewise affine decision rules for stochastic programming with covariate information. arXiv preprint arXiv:2304.13646 , 2023

    work page arXiv 2023

  71. [71]

    Joint estimation and robustness optimization

    Taozeng Zhu, Jingui Xie, and Melvyn Sim. Joint estimation and robustness optimization. Management Science, 68(3):1659–1677, 2022. 34 A Interchangeability Principle We first restate an interchangeability principle in [28, Theorem 2.2]. Lemma A.1 Consider a real separable Banach space Z with Borel field B(Z) and a complete sample space Ω associated with sig...

    work page 2022

  72. [72]

    Lemma A.2 (Lemma 1 in [42]) Consider a Polish space Z with Borel field B(Z) and a prob- ability space (Ω, F , P)

    extends the interchangeability principle to a Polish space. Lemma A.2 (Lemma 1 in [42]) Consider a Polish space Z with Borel field B(Z) and a prob- ability space (Ω, F , P). Let Z : Ω ⇒ Z be a F-measurable set-valued mapping with closed values. Let L be a linear space of measurable functions g : Ω → Z and LZ := {g ∈ L : g(ω) ∈ Z (ω) ⊆ Z, for a.e. ω ∈ Ω}. ...

    work page

  73. [73]

    Consider, for another example, a newsvendor problem with entropic risk measure and h(z, s, y) = eγ(a(z−y)++b(y−z)+), (D.55) where z, y ∈ R and a, b, γ are positive constants

    + a + b 1 − α(1 + 2β2 1 + 2|y|2) + ∥x∥2 ≥ h(z(x), s(x), y), and R X ×Y ϕ(x, y)P (dxdy) < ∞ holds for P ∈ M2 1. Consider, for another example, a newsvendor problem with entropic risk measure and h(z, s, y) = eγ(a(z−y)++b(y−z)+), (D.55) where z, y ∈ R and a, b, γ are positive constants. Then |h(z(x), s(x), y)| = eγ(a(z(x)−y)++b(y−z(x))+) ≤ eγ(a+b)|z(x)−y|. ...

    work page

  74. [74]

    for N ≥ N0 and R(z, s) =Pdz t=1 ∥z(t)∥2 H+ ∥s∥2 H ≤ β2 1 + β2 2, and the second inequality comes from (E.61). On the other hand, by setting 47 ϵ < δ 6, we have Prob sup k∈{1,...,K} R1(zk, sk, ϵ, P N) ≥ δ − 6ϵ ! ≤ KX k=1 Prob R1(zk, sk, ϵ, P N) ≥ δ − 6ϵ ≤ KX k=1 Prob R1(zk, sk, ϵ, P N) ≥ δ − 6ϵ + Prob R1(zk, sk, ϵ, P N) ≤ −(δ − 6ϵ) ≤ 2 KX k=1 e−N[I(δ−6ϵ,zk...

    work page 2035