pith. sign in

arxiv: 2403.09532 · v4 · submitted 2024-03-14 · 🧮 math.OC · math.PR· q-fin.MF

Robust SGLD algorithm for solving non-convex distributionally robust optimisation problems

Ariel Neufeld , Matthew Ng Cheng En , Ying Zhang This is my paper

Pith reviewed 2026-05-24 02:37 UTC · model grok-4.3

classification 🧮 math.OC math.PRq-fin.MF
keywords SGLDdistributionally robust optimizationnon-convex optimizationconvergence boundsneural networksexcess riskadversarial samples
0
0 comments X

The pith

A robust SGLD algorithm guarantees non-asymptotic excess risk bounds for non-convex DRO problems.

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

The authors develop a Stochastic Gradient Langevin Dynamics algorithm specifically for a class of non-convex distributionally robust optimization problems. They derive non-asymptotic convergence bounds that allow the algorithm to produce, for any given accuracy level ε greater than zero, an estimator with expected excess risk at most ε. The method is applied to training neural network estimators for regression models when samples are adversarially corrupted, and experiments show it achieves higher test accuracy than standard SGLD by accounting for model uncertainty.

Core claim

By deriving non-asymptotic convergence bounds, the paper builds an SGLD algorithm which for any prescribed accuracy ε>0 outputs an estimator whose expected excess risk is at most ε. As a concrete application, it considers identifying the best non-linear estimator of a given regression model involving a neural network using adversarially corrupted samples, formulated as a DRO problem, and demonstrates both theoretically and numerically the applicability of the proposed robust SGLD algorithm.

What carries the argument

The tailored Stochastic Gradient Langevin Dynamics (SGLD) algorithm for non-convex DRO, which provides non-asymptotic convergence to control excess risk.

If this is right

  • The algorithm outputs estimators meeting any prescribed accuracy ε in terms of expected excess risk.
  • It applies to neural network based regression under adversarial sample corruption.
  • Numerical results indicate superior test accuracy compared to vanilla SGLD when incorporating model uncertainty.
  • The approach handles perturbed samples by optimizing under distributional robustness.

Where Pith is reading between the lines

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

  • This could improve robustness in other machine learning tasks involving data uncertainty beyond the neural network regression example.
  • Future work might explore the computational efficiency of this robust SGLD compared to other DRO solvers.
  • Similar non-asymptotic bounds might be derivable for other sampling-based optimization methods in non-convex settings.

Load-bearing premise

The class of non-convex DRO problems allows an SGLD formulation that supports the derivation of the non-asymptotic excess-risk convergence bound.

What would settle it

Finding a problem in the considered class of non-convex DRO where the algorithm's expected excess risk exceeds the prescribed ε, or where the convergence bound does not hold.

read the original abstract

In this paper we develop a Stochastic Gradient Langevin Dynamics (SGLD) algorithm tailored for solving a certain class of non-convex distributionally robust optimisation (DRO) problems. By deriving non-asymptotic convergence bounds, we build an algorithm which for any prescribed accuracy $\varepsilon>0$ outputs an estimator whose expected excess risk is at most $\varepsilon$. As a concrete application, we consider the problem of identifying the best non-linear estimator of a given regression model involving a neural network using adversarially corrupted samples. We formulate this problem as a DRO problem and demonstrate both theoretically and numerically the applicability of the proposed robust SGLD algorithm. Moreover, numerical experiments show that the robust SGLD estimator outperforms the estimator obtained using vanilla SGLD in terms of test accuracy, which highlights the advantage of incorporating model uncertainty when optimising with perturbed samples.

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

2 major / 1 minor

Summary. The paper develops a robust variant of Stochastic Gradient Langevin Dynamics (SGLD) for a class of non-convex distributionally robust optimization (DRO) problems. It derives non-asymptotic convergence bounds guaranteeing that, for any prescribed ε > 0, the algorithm outputs an estimator whose expected excess risk is at most ε. The method is applied to identifying the best neural-network regressor under adversarially corrupted samples, formulated as a DRO problem; theoretical guarantees and numerical experiments are presented showing improved test accuracy over vanilla SGLD.

Significance. A non-asymptotic excess-risk guarantee for non-convex DRO would be a useful contribution to robust optimization, particularly for neural-network models. The numerical demonstration of robustness to adversarial corruption is a positive feature. However, the conversion from typical SGLD stationarity bounds to global excess risk is not automatic in the non-convex regime and requires additional structure that is not identified in the abstract.

major comments (2)
  1. [Abstract] Abstract (paragraph 2): The central claim states that the robust SGLD outputs an estimator whose expected excess risk is at most any prescribed ε > 0. Standard non-asymptotic analyses of SGLD for non-convex smooth objectives bound E[‖∇f‖²] (typically O(1/√T) under L-smoothness and bounded variance). Converting a stationarity guarantee into a global excess-risk bound f(θ) − inf f requires either convexity or a condition such as the Polyak-Łojasiewicz inequality; neither is stated for the class of non-convex DRO problems considered. This gap directly undermines the excess-risk claim.
  2. [Abstract] The weakest assumption identified in the reader's report (that the specific class of non-convex DRO problems admits an SGLD formulation whose convergence analysis yields the stated non-asymptotic excess-risk bound) is load-bearing. Without explicit verification that the DRO objective satisfies the necessary structural conditions, the non-asymptotic guarantee does not follow from the usual SGLD analysis.
minor comments (1)
  1. The abstract asserts numerical superiority but supplies no details on error bars, number of runs, data-exclusion rules, or statistical significance tests; these should be added to the experimental section for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and insightful comments on the convergence claims. We agree that the abstract overstated the nature of the non-asymptotic guarantees and will revise accordingly to ensure accuracy.

read point-by-point responses

  1. Referee: [Abstract] Abstract (paragraph 2): The central claim states that the robust SGLD outputs an estimator whose expected excess risk is at most any prescribed ε > 0. Standard non-asymptotic analyses of SGLD for non-convex smooth objectives bound E[‖∇f‖²] (typically O(1/√T) under L-smoothness and bounded variance). Converting a stationarity guarantee into a global excess-risk bound f(θ) − inf f requires either convexity or a condition such as the Polyak-Łojasiewicz inequality; neither is stated for the class of non-convex DRO problems considered. This gap directly undermines the excess-risk claim.

    Authors: We agree that stationarity bounds do not imply excess-risk bounds in the non-convex regime without additional structure such as the Polyak-Łojasiewicz inequality or convexity. Our analysis derives non-asymptotic bounds on the expected squared gradient norm for the robust SGLD applied to the DRO objective under standard smoothness and variance assumptions. The abstract's phrasing regarding excess risk was imprecise and does not follow from the analysis. In revision we will replace the excess-risk claim with a precise statement that the algorithm returns an ε-stationary point (E[‖∇f(θ)‖²] ≤ ε) and will add a remark clarifying that global excess-risk guarantees are not obtained. revision: yes

  2. Referee: [Abstract] The weakest assumption identified in the reader's report (that the specific class of non-convex DRO problems admits an SGLD formulation whose convergence analysis yields the stated non-asymptotic excess-risk bound) is load-bearing. Without explicit verification that the DRO objective satisfies the necessary structural conditions, the non-asymptotic guarantee does not follow from the usual SGLD analysis.

    Authors: We accept that the manuscript does not explicitly verify additional structural conditions (beyond smoothness and bounded variance) that would be needed for an excess-risk bound. The current analysis follows the standard SGLD stationarity framework for non-convex problems. In the revised version we will insert a short discussion after the main convergence theorem that explicitly lists the assumptions used and confirms that no PL or convexity assumption is invoked, thereby aligning the stated guarantees with the actual analysis. revision: yes

Circularity Check

0 steps flagged

No circularity: convergence bounds derived independently from SGLD formulation

full rationale

The paper claims to derive non-asymptotic convergence bounds for a tailored SGLD algorithm on a class of non-convex DRO problems, yielding an estimator with expected excess risk at most ε for any ε>0. No equations or steps in the abstract reduce this bound to a fitted quantity defined by the same paper, a self-citation chain, or an ansatz smuggled from prior work by the authors. The derivation is presented as following from the algorithm's construction and standard analysis techniques under the stated assumptions, without the result being equivalent to its inputs by construction. This is the most common honest finding for papers whose central claims rest on independently verifiable convergence analysis rather than renaming or self-referential fitting.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the existence of non-asymptotic convergence bounds for the proposed robust SGLD on the stated class of non-convex DRO problems; the abstract provides no explicit free parameters, axioms, or invented entities.

pith-pipeline@v0.9.0 · 5682 in / 1042 out tokens · 23355 ms · 2026-05-24T02:37:08.188891+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

145 extracted references · 145 canonical work pages · 3 internal anchors

  1. [1]

    Adageo: Adaptive geometric learning for optimization and sampling

    Gabriele Abbati, Alessandra Tosi, Michael Osborne, and Seth Flaxman. Adageo: Adaptive geometric learning for optimization and sampling. InInternational conference on artificial intelligence and statistics, pages 226–234. PMLR, 2018

    work page 2018

  2. [2]

    A model-free version of the fundamental theorem of asset pricing and the super-replication theorem.Mathemati- cal Finance, 26(2):233–251, 2016

    Beatrice Acciaio, Mathias Beiglb¨ock, Friedrich Penkner, and Walter Schachermayer. A model-free version of the fundamental theorem of asset pricing and the super-replication theorem.Mathemati- cal Finance, 26(2):233–251, 2016

    work page 2016

  3. [3]

    Bayesian posterior sampling via stochastic gradient fisher scoring

    Sungjin Ahn, Anoop Korattikara, and Max Welling. Bayesian posterior sampling via stochastic gradient fisher scoring. ICML’12, page 1771–1778, Madison, WI, USA, 2012. Omnipress. ISBN 9781450312851

    work page 2012

  4. [4]

    Stochastic gradient mcmc for state space models.SIAM Journal on Mathematics of Data Science, 1(3):555–587, 2019

    Christopher Aicher, Yi-An Ma, Nicholas J Foti, and Emily B Fox. Stochastic gradient mcmc for state space models.SIAM Journal on Mathematics of Data Science, 1(3):555–587, 2019

    work page 2019

  5. [5]

    Physics-informed information field theory for modeling physical systems with uncertainty quantification.Journal of Computational Physics, 486:112100, 2023

    Alex Alberts and Ilias Bilionis. Physics-informed information field theory for modeling physical systems with uncertainty quantification.Journal of Computational Physics, 486:112100, 2023

    work page 2023

  6. [6]

    Efficient optimal transport algorithm by accelerated gradient descent

    Dongsheng An, Na Lei, Xiaoyin Xu, and Xianfeng Gu. Efficient optimal transport algorithm by accelerated gradient descent. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 10119–10128, 2022

    work page 2022

  7. [7]

    Wasserstein distributionally robust estimation in high dimensions: Performance analysis and optimal hyperparameter tuning

    Liviu Aolaritei, Soroosh Shafiee, and Florian D¨orfler. Wasserstein distributionally robust estimation in high dimensions: Performance analysis and optimal hyperparameter tuning

    work page

  8. [8]

    Distributional uncertainty propagation via optimal transport.arXiv preprint arXiv:2205.00343, 2023

    Liviu Aolaritei, Nicolas Lanzetti, Hongruyu Chen, and Florian D¨orfler. Distributional uncertainty propagation via optimal transport.arXiv preprint arXiv:2205.00343, 2023

    work page arXiv 2023

  9. [9]

    Deep learning and optimisation for quality of service modelling

    Krishnakumar Arunachalam, Senthilkumaran Thangamuthu, Vijayanand Shanmugam, Mukesh Raju, and Kamali Premraj. Deep learning and optimisation for quality of service modelling. Journal of King Saud University-Computer and Information Sciences, 34(8):5998–6007, 2022

    work page 2022

  10. [10]

    On stochastic gradient Langevin dynamics with dependent data streams in the logconcave case.Bernoulli, 27(1):1–33, 2021

    Mathias Barkhagen, Ngoc Huy Chau, ´Eric Moulines, Mikl´os R´asonyi, Sotirios Sabanis, and Ying Zhang. On stochastic gradient Langevin dynamics with dependent data streams in the logconcave case.Bernoulli, 27(1):1–33, 2021

    work page 2021

  11. [11]

    Sensitivity of multiperiod optimization problems with respect to the adapted Wasserstein distance.SIAM J

    Daniel Bartl and Johannes Wiesel. Sensitivity of multiperiod optimization problems with respect to the adapted Wasserstein distance.SIAM J. Financial Math., 14(2):704–720, 2023

    work page 2023

  12. [12]

    Computational aspects of robust optimized certainty equivalents.Preprint

    Daniel Bartl, Samuel Drapeau, and Ludovic Tangpi. Computational aspects of robust optimized certainty equivalents.Preprint. Available at https://arxiv. org/abs/1706.10186 v1, 2017

    work page arXiv 2017

  13. [13]

    Sensitivity analysis of Wasserstein distributionally robust optimization problems.Proc

    Daniel Bartl, Samuel Drapeau, Jan Obł´oj, and Johannes Wiesel. Sensitivity analysis of Wasserstein distributionally robust optimization problems.Proc. R. Soc. A, 477(2256):20210176, 2021

    work page 2021

  14. [14]

    Duality theory for robust utility maximisation

    Daniel Bartl, Michael Kupper, and Ariel Neufeld. Duality theory for robust utility maximisation. Finance and Stochastics, 25(3):469–503, 2021

    work page 2021

  15. [15]

    Bartl, A

    Daniel Bartl, Ariel Neufeld, and Kyunghyun Park. Sensitivity of robust optimization problems under drift and volatility uncertainty.arXiv preprint arXiv:2311.11248, 2023

    work page arXiv 2023

  16. [16]

    Numerical method for nonlinear Kolmogorov PDEs via sensitivity analysis

    Daniel Bartl, Ariel Neufeld, and Kyunghyun Park. Numerical method for nonlinear Kolmogorov PDEs via sensitivity analysis.arXiv preprint arXiv:2403.11910, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  17. [17]

    Data-driven non-parametric robust control under dependence uncertainty

    Erhan Bayraktar and Tao Chen. Data-driven non-parametric robust control under dependence uncertainty. InPeter Carr Gedenkschrift: Research Advances in Mathematical Finance, pages 141–178. World Scientific, 2024

    work page 2024

  18. [18]

    Equilibria under knightian price uncertainty.Econometrica, 87 (1):37–64, 2019

    Patrick Beissner and Frank Riedel. Equilibria under knightian price uncertainty.Econometrica, 87 (1):37–64, 2019

    work page 2019

  19. [19]

    Estimation and uncertainty quantification for the output from quantum simulators.Foundations of Data Science, 1(2):157–176, 2019

    Ryan Bennink, Ajay Jasra, Kody JH Law, and Pavel Lougovski. Estimation and uncertainty quantification for the output from quantum simulators.Foundations of Data Science, 1(2):157–176, 2019

    work page 2019

  20. [20]

    Robust distortion risk measures

    Carole Bernard, Silvana M Pesenti, and Steven Vanduffel. Robust distortion risk measures. Mathematical Finance, 2023

    work page 2023

  21. [21]

    Models for minimax stochastic linear optimization problems with risk aversion.Mathematics of Operations Research, 35(3):580–602, 2010

    Dimitris Bertsimas, Xuan Vinh Doan, Karthik Natarajan, and Chung-Piaw Teo. Models for minimax stochastic linear optimization problems with risk aversion.Mathematics of Operations Research, 35(3):580–602, 2010

    work page 2010

  22. [22]

    Scaling up dynamic topic models

    Arnab Bhadury, Jianfei Chen, Jun Zhu, and Shixia Liu. Scaling up dynamic topic models. In Proceedings of the 25th International Conference on World Wide Web, pages 381–390, 2016. 35

    work page 2016

  23. [23]

    Multiple-priors optimal investment in discrete time for unbounded utility function.The Annals of Applied Probability, 28(3):1856–1892, 2018

    Romain Blanchard and Laurence Carassus. Multiple-priors optimal investment in discrete time for unbounded utility function.The Annals of Applied Probability, 28(3):1856–1892, 2018

    work page 2018

  24. [24]

    Quantifying distributional model risk via optimal transport

    Jose Blanchet and Karthyek Murthy. Quantifying distributional model risk via optimal transport. Mathematics of Operations Research, 44(2):565–600, 2019

    work page 2019

  25. [25]

    On distributionally robust extreme value analysis

    Jose Blanchet, Fei He, and Karthyek Murthy. On distributionally robust extreme value analysis. Extremes, 23:317–347, 2020

    work page 2020

  26. [26]

    Arbitrage and duality in nondominated discrete-time models

    Bruno Bouchard and Marcel Nutz. Arbitrage and duality in nondominated discrete-time models. The Annals of Applied Probability, 25(2):823 – 859, 2015. doi: 10.1214/14-AAP1011. URL https://doi.org/10.1214/14-AAP1011

    work page doi:10.1214/14-aap1011 2015

  27. [27]

    Langevin algorithms for very deep neural networks with application to image classification.Procedia Computer Science, 222:303–310, 2023

    Pierre Bras. Langevin algorithms for very deep neural networks with application to image classification.Procedia Computer Science, 222:303–310, 2023

    work page 2023

  28. [28]

    Langevin algorithms for markovian neural networks and deep stochastic control

    Pierre Bras and Gilles Pag `es. Langevin algorithms for markovian neural networks and deep stochastic control. In2023 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE, 2023

    work page 2023

  29. [29]

    The promises and pitfalls of stochastic gradient Langevin dynamics

    Nicolas Brosse, Alain Durmus, and Eric Moulines. The promises and pitfalls of stochastic gradient Langevin dynamics. InAdvances in Neural Information Processing Systems, pages 8268–8278, 2018

    work page 2018

  30. [30]

    Viability and arbitrage under knightian uncertainty.Econometrica, 89(3):1207–1234, 2021

    Matteo Burzoni, Frank Riedel, and H Mete Soner. Viability and arbitrage under knightian uncertainty.Econometrica, 89(3):1207–1234, 2021

    work page 2021

  31. [31]

    The robust superreplication problem: a dynamic approach.SIAM Journal on Financial Mathematics, 10(4):907–941, 2019

    Laurence Carassus, Jan Obł´oj, and Johannes Wiesel. The robust superreplication problem: a dynamic approach.SIAM Journal on Financial Mathematics, 10(4):907–941, 2019

    work page 2019

  32. [32]

    On stochastic gradient langevin dynamics with dependent data streams: The fully nonconvex case.SIAM Journal on Mathematics of Data Science, 3(3):959–986, 2021

    Ngoc Huy Chau, ´Eric Moulines, Miklos R´asonyi, Sotirios Sabanis, and Ying Zhang. On stochastic gradient langevin dynamics with dependent data streams: The fully nonconvex case.SIAM Journal on Mathematics of Data Science, 3(3):959–986, 2021

    work page 2021

  33. [33]

    On the convergence of stochastic gradient mcmc algorithms with high-order integrators.Advances in neural information processing systems, 28, 2015

    Changyou Chen, Nan Ding, and Lawrence Carin. On the convergence of stochastic gradient mcmc algorithms with high-order integrators.Advances in neural information processing systems, 28, 2015

    work page 2015

  34. [34]

    Distributionally robust linear and discrete optimization with marginals.Operations Research, 70(3):1822–1834, 2022

    Louis Chen, Will Ma, Karthik Natarajan, David Simchi-Levi, and Zhenzhen Yan. Distributionally robust linear and discrete optimization with marginals.Operations Research, 70(3):1822–1834, 2022

    work page 2022

  35. [35]

    A robust learning approach for regression models based on distributionally robust optimization.Journal of Machine Learning Research, 19(13):1–48, 2018

    Ruidi Chen and Ioannis Ch Paschalidis. A robust learning approach for regression models based on distributionally robust optimization.Journal of Machine Learning Research, 19(13):1–48, 2018

    work page 2018

  36. [36]

    On stationary-point hitting time and ergodicity of stochastic gradient Langevin dynamics.Journal of Machine Learning Research, 2020

    Xi Chen, Simon S Du, and Xin T Tong. On stationary-point hitting time and ergodicity of stochastic gradient Langevin dynamics.Journal of Machine Learning Research, 2020

    work page 2020

  37. [37]

    Duality formulas for robust pricing and hedging in discrete time.SIAM Journal on Financial Mathematics, 8(1):738–765, 2017

    Patrick Cheridito, Michael Kupper, and Ludovic Tangpi. Duality formulas for robust pricing and hedging in discrete time.SIAM Journal on Financial Mathematics, 8(1):738–765, 2017

    work page 2017

  38. [38]

    Martingale optimal transport duality.Mathematische Annalen, 379:1685–1712, 2021

    Patrick Cheridito, Matti Kiiski, David J Pr¨omel, and H Mete Soner. Martingale optimal transport duality.Mathematische Annalen, 379:1685–1712, 2021

    work page 2021

  39. [39]

    Non-asymptotic estimation of risk measures using stochastic gradient Langevin dynamics.arXiv preprint arXiv:2111.12248, 2021

    Jiarui Chu and Ludovic Tangpi. Non-asymptotic estimation of risk measures using stochastic gradient Langevin dynamics.arXiv preprint arXiv:2111.12248, 2021

    work page arXiv 2021

  40. [40]

    Further and stronger analogy between sampling and optimization: Langevin Monte Carlo and gradient descent

    Arnak S Dalalyan. Further and stronger analogy between sampling and optimization: Langevin Monte Carlo and gradient descent. InProceedings of the 2017 Conference on Learning Theory, volume 65 ofProceedings of Machine Learning Research, pages 678–689. PMLR, 07–10 Jul 2017

    work page 2017

  41. [41]

    User-friendly guarantees for the Langevin monte carlo with inaccurate gradient.Stochastic Processes and their Applications, 129(12):5278–5311, 2019

    Arnak S Dalalyan and Avetik Karagulyan. User-friendly guarantees for the Langevin monte carlo with inaccurate gradient.Stochastic Processes and their Applications, 129(12):5278–5311, 2019

    work page 2019

  42. [42]

    Distributionally robust optimization under moment uncertainty with application to data-driven problems.Operations research, 58(3):595–612, 2010

    Erick Delage and Yinyu Ye. Distributionally robust optimization under moment uncertainty with application to data-driven problems.Operations research, 58(3):595–612, 2010

    work page 2010

  43. [43]

    An adaptively weighted stochastic gradient mcmc algorithm for monte carlo simulation and global optimization.Statistics and Computing, 32(4):58, 2022

    Wei Deng, Guang Lin, and Faming Liang. An adaptively weighted stochastic gradient mcmc algorithm for monte carlo simulation and global optimization.Statistics and Computing, 32(4):58, 2022

    work page 2022

  44. [44]

    A theoretical framework for the pricing of contingent claims in the presence of model uncertainty.The Annals of Applied Probability, 16(2):827 – 852, 2006

    Laurent Denis and Claude Martini. A theoretical framework for the pricing of contingent claims in the presence of model uncertainty.The Annals of Applied Probability, 16(2):827 – 852, 2006. doi: 10.1214/105051606000000169. URL https://doi.org/10.1214/ 105051606000000169. 36 A. NEUFELD, M. NG, AND Y . ZHANG

    work page doi:10.1214/105051606000000169 2006

  45. [45]

    Martingale optimal transport and robust hedging in continuous time.Probability Theory and Related Fields, 160(1-2):391–427, 2014

    Yan Dolinsky and H Mete Soner. Martingale optimal transport and robust hedging in continuous time.Probability Theory and Related Fields, 160(1-2):391–427, 2014

    work page 2014

  46. [46]

    Robust hedging with proportional transaction costs.Finance and Stochastics, 18:327–347, 2014

    Yan Dolinsky and H Mete Soner. Robust hedging with proportional transaction costs.Finance and Stochastics, 18:327–347, 2014

    work page 2014

  47. [47]

    Analysis of Langevin Monte Carlo via convex optimization.The Journal of Machine Learning Research, 20(1):2666–2711, 2019

    Alain Durmus, Szymon Majewski, and Bła˙zej Miasojedow. Analysis of Langevin Monte Carlo via convex optimization.The Journal of Machine Learning Research, 20(1):2666–2711, 2019

    work page 2019

  48. [48]

    Robust risk aggregation with neural networks.Mathematical finance, 30(4):1229–1272, 2020

    Stephan Eckstein, Michael Kupper, and Mathias Pohl. Robust risk aggregation with neural networks.Mathematical finance, 30(4):1229–1272, 2020

    work page 2020

  49. [49]

    Risk, ambiguity, and the savage axioms.The quarterly journal of economics, 75 (4):643–669, 1961

    Daniel Ellsberg. Risk, ambiguity, and the savage axioms.The quarterly journal of economics, 75 (4):643–669, 1961

    work page 1961

  50. [50]

    Intertemporal asset pricing under knightian uncertainty

    Larry G Epstein and Tan Wang. Intertemporal asset pricing under knightian uncertainty. In Uncertainty in Economic Theory, pages 445–487. Routledge, 2004

    work page 2004

  51. [51]

    Time-independent generalization bounds for sgld in non- convex settings.Advances in Neural Information Processing Systems, 34:19836–19846, 2021

    Tyler Farghly and Patrick Rebeschini. Time-independent generalization bounds for sgld in non- convex settings.Advances in Neural Information Processing Systems, 34:19836–19846, 2021

    work page 2021

  52. [52]

    Portfolio optimization with ambiguous correlation and stochastic volatilities.SIAM Journal on Control and Optimization, 54(5):2309– 2338, 2016

    Jean-Pierre Fouque, Chi Seng Pun, and Hoi Ying Wong. Portfolio optimization with ambiguous correlation and stochastic volatilities.SIAM Journal on Control and Optimization, 54(5):2309– 2338, 2016

    work page 2016

  53. [53]

    Variationally inferred sampling through a refined bound

    V´ıctor Gallego and David Rios Insua. Variationally inferred sampling through a refined bound. Entropy, 23(1):123, 2021

    work page 2021

  54. [54]

    Distributionally robust stochastic optimization with Wasserstein distance.Mathematics of Operations Research, 48(2):603–655, 2023

    Rui Gao and Anton Kleywegt. Distributionally robust stochastic optimization with Wasserstein distance.Mathematics of Operations Research, 48(2):603–655, 2023

    work page 2023

  55. [55]

    Wasserstein distributionally robust optimization and variation regularization.Oper

    Rui Gao, Xi Chen, and Anton J Kleywegt. Wasserstein distributionally robust optimization and variation regularization.Oper. Res., 2022

    work page 2022

  56. [56]

    Maxmin expected utility with non-unique prior.Journal of mathematical economics, 18(2):141–153, 1989

    Itzhak Gilboa and David Schmeidler. Maxmin expected utility with non-unique prior.Journal of mathematical economics, 18(2):141–153, 1989

    work page 1989

  57. [57]

    A stochastic subgradient method for distributionally robust non-convex and non-smooth learning.Journal of Optimization Theory and Applications, 194(3):1014–1041, 2022

    Mert G¨urb¨uzbalaban, Andrzej Ruszczy´nski, and Landi Zhu. A stochastic subgradient method for distributionally robust non-convex and non-smooth learning.Journal of Optimization Theory and Applications, 194(3):1014–1041, 2022

    work page 2022

  58. [58]

    Robust control and model uncertainty.American Economic Review, 91(2):60–66, 2001

    Lars Peter Hansen and Thomas J Sargent. Robust control and model uncertainty.American Economic Review, 91(2):60–66, 2001

    work page 2001

  59. [59]

    Uncertainty quantification for plant disease detection using Bayesian deep learning.Applied Soft Computing, 96:106597, 2020

    S Hern ´andez and Juan L L ´opez. Uncertainty quantification for plant disease detection using Bayesian deep learning.Applied Soft Computing, 96:106597, 2020

    work page 2020

  60. [60]

    Model uncertainty, recalibration, and the emer- gence of delta–vega hedging.Finance Stoch., 21:873–930, 2017

    Sebastian Herrmann and Johannes Muhle-Karbe. Model uncertainty, recalibration, and the emer- gence of delta–vega hedging.Finance Stoch., 21:873–930, 2017

    work page 2017

  61. [61]

    Hedging with small uncertainty aversion.Finance Stoch., 21:1–64, 2017

    Sebastian Herrmann, Johannes Muhle-Karbe, and Frank Thomas Seifried. Hedging with small uncertainty aversion.Finance Stoch., 21:1–64, 2017

    work page 2017

  62. [62]

    Zhengyang Hu, Goutham Ramaraj, and Guiping Hu. Production planning with a two-stage stochas- tic programming model in a kitting facility under demand and yield uncertainties.International Journal of Management Science and Engineering Management, 15(3):237–246, 2020

    work page 2020

  63. [63]

    Robust risk-aware reinforcement learning.SIAM Journal on Financial Mathematics, 13(1):213–226, 2022

    Sebastian Jaimungal, Silvana M Pesenti, Ye Sheng Wang, and Hariom Tatsat. Robust risk-aware reinforcement learning.SIAM Journal on Financial Mathematics, 13(1):213–226, 2022

    work page 2022

  64. [64]

    Poisoning attacks on data- driven utility learning in games

    Ruoxi Jia, Ioannis C Konstantakopoulos, Bo Li, and Costas Spanos. Poisoning attacks on data- driven utility learning in games. In2018 annual American control conference (ACC), pages 5774–5780. IEEE, 2018

    work page 2018

  65. [65]

    Improvements on scalable stochastic Bayesian inference methods for multivariate hawkes process.Statistics and Computing, 34(2):85, 2024

    Alex Ziyu Jiang and Abel Rodriguez. Improvements on scalable stochastic Bayesian inference methods for multivariate hawkes process.Statistics and Computing, 34(2):85, 2024

    work page 2024

  66. [66]

    Sensitivity of causal distributionally robust optimization.arXiv preprint arXiv:2408.17109, 2024

    Yifan Jiang and Jan Obloj. Sensitivity of causal distributionally robust optimization.arXiv preprint arXiv:2408.17109, 2024

    work page arXiv 2024

  67. [67]

    Robust reinforcement learning via adversarial training with langevin dynamics.Advances in Neural Information Processing Systems, 33:8127–8138, 2020

    Parameswaran Kamalaruban, Yu-Ting Huang, Ya-Ping Hsieh, Paul Rolland, Cheng Shi, and V olkan Cevher. Robust reinforcement learning via adversarial training with langevin dynamics.Advances in Neural Information Processing Systems, 33:8127–8138, 2020

    work page 2020

  68. [68]

    Yuri Kinoshita and Taiji Suzuki. Improved convergence rate of stochastic gradient Langevin dy- namics with variance reduction and its application to optimization.Advances in Neural Information Processing Systems, 35:19022–19034, 2022. 37

    work page 2022

  69. [69]

    A smooth model of decision making under ambiguity.Econometrica, 73(6):1849–1892, 2005

    Peter Klibanoff, Massimo Marinacci, and Sujoy Mukerji. A smooth model of decision making under ambiguity.Econometrica, 73(6):1849–1892, 2005

    work page 2005

  70. [70]

    Houghton Mifflin, 1921

    Frank Hyneman Knight.Risk, uncertainty and profit, volume 31. Houghton Mifflin, 1921

    work page 1921

  71. [71]

    Appointment scheduling under time-dependent patient no-show behavior.Management Science, 66(8):3480–3500, 2020

    Qingxia Kong, Shan Li, Nan Liu, Chung-Piaw Teo, and Zhenzhen Yan. Appointment scheduling under time-dependent patient no-show behavior.Management Science, 66(8):3480–3500, 2020

    work page 2020

  72. [72]

    Risk measures based on weak optimal transport.arXiv preprint arXiv:2312.05973, 2023

    Michael Kupper, Max Nendel, and Alessandro Sgarabottolo. Risk measures based on weak optimal transport.arXiv preprint arXiv:2312.05973, 2023

    work page arXiv 2023

  73. [73]

    Principled learn- ing method for Wasserstein distributionally robust optimization with local perturbations

    Yongchan Kwon, Wonyoung Kim, Joong-Ho Won, and Myunghee Cho Paik. Principled learn- ing method for Wasserstein distributionally robust optimization with local perturbations. In International Conference on Machine Learning, pages 5567–5576. PMLR, 2020

    work page 2020

  74. [74]

    On the adversarial robustness of robust estimators.IEEE Transactions on Information Theory, 66(8):5097–5109, 2020

    Lifeng Lai and Erhan Bayraktar. On the adversarial robustness of robust estimators.IEEE Transactions on Information Theory, 66(8):5097–5109, 2020

    work page 2020

  75. [75]

    Bipolar Theorems for Sets of Non-negative Random Variables

    Johannes Langner and Gregor Svindland. Bipolar theorems for sets of non-negative random variables.arXiv preprint arXiv:2212.14259, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  76. [76]

    Swing contract pricing: A parametric approach with adjoint automatic differentiation and neural networks.arXiv preprint arXiv:2306.03822, 2023

    Vincent Lemaire, Christian Yeo, et al. Swing contract pricing: A parametric approach with adjoint automatic differentiation and neural networks.arXiv preprint arXiv:2306.03822, 2023

    work page arXiv 2023

  77. [77]

    Preconditioned stochastic gradient Langevin dynamics for deep neural networks

    Chunyuan Li, Changyou Chen, David Carlson, and Lawrence Carin. Preconditioned stochastic gradient Langevin dynamics for deep neural networks. InProceedings of the AAAI conference on artificial intelligence, volume 30, 2016

    work page 2016

  78. [78]

    High-order stochastic gradient thermostats for Bayesian learning of deep models

    Chunyuan Li, Changyou Chen, Kai Fan, and Lawrence Carin. High-order stochastic gradient thermostats for Bayesian learning of deep models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 30, 2016

    work page 2016

  79. [79]

    Policy gradient algorithms for robust mdps with non-rectangular uncertainty sets.arXiv preprint arXiv:2305.19004, 2023

    Mengmeng Li, Tobias Sutter, and Daniel Kuhn. Policy gradient algorithms for robust mdps with non-rectangular uncertainty sets.arXiv preprint arXiv:2305.19004, 2023

    work page arXiv 2023

  80. [80]

    Scalable mcmc for mixed membership stochastic blockmodels

    Wenzhe Li, Sungjin Ahn, and Max Welling. Scalable mcmc for mixed membership stochastic blockmodels. InArtificial Intelligence and Statistics, pages 723–731. PMLR, 2016

    work page 2016

Showing first 80 references.