pith. machine review for the scientific record. sign in

arxiv: 2605.04207 · v1 · submitted 2026-05-05 · 📊 stat.ME · econ.GN· q-fin.EC

Recognition: unknown

Optimal Semiparametric Dynamic Pricing with Feature Diversity

Jinhang Chai , Yaqi Duan , Jianqing Fan , Kaizheng Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:30 UTC · model grok-4.3

classification 📊 stat.ME econ.GNq-fin.EC
keywords dynamic pricingsemiparametric modelregret boundslocal polynomial regressionfeature diversitycontextual pricinggreedy algorithmnonparametric estimation
0
0 comments X

The pith

The stagewise greedy pricing algorithm achieves optimal regret rates in semiparametric contextual dynamic pricing by iteratively estimating the unknown noise distribution using local polynomials on reused samples.

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

This paper studies dynamic pricing where buyers purchase with probability 1 minus F of price minus a mean utility m of features and covariates. The authors develop a stagewise greedy algorithm that prices based on current estimates while refining the estimate of the unknown F through local polynomial regression. Feature diversity allows the algorithm to use data collected while exploiting current prices for this estimation, sidestepping the need for separate random exploration phases that prior approaches required. For cases where m is linear, the resulting regret is on the order of T to the max of 1/2 and 3 over 2 beta plus 1, with beta the smoothness of F, and this rate is proven optimal by a matching lower bound. When F is smooth enough that beta is at least 5/2, the regret reaches the ideal parametric rate of square root of T.

Core claim

In the semiparametric model where the purchase probability is 1 - F(p - m(x)), with F an unknown distribution and m a known class of functions, the stagewise greedy pricing algorithm iteratively refines the estimate of F via local polynomial regression while pricing greedily with the current estimate of m and F. By exploiting feature diversity, the algorithm reuses endogenous samples collected during exploitation for the nonparametric estimation, avoiding costly global random exploration. This yields a general regret bound applicable to any estimator of m, with explicit rates for linear, nonparametric additive, and sparse linear classes; for linear m the regret is T to the power max{1/2, 3/(

What carries the argument

The stagewise greedy pricing algorithm that alternates greedy pricing decisions with local polynomial regression updates to the market noise distribution F, made possible by the feature diversity condition that permits reuse of exploitation samples for estimation.

Load-bearing premise

The covariates and features must satisfy a diversity condition that ensures endogenous samples collected during greedy pricing are sufficient to estimate the noise distribution F nonparametrically without bias or extra exploration.

What would settle it

A controlled experiment or simulation in which the observed cumulative regret grows faster than T to the power of 3/(2 beta +1) for a linear utility function and smoothness beta =1 would indicate that the regret upper bound does not hold.

Figures

Figures reproduced from arXiv: 2605.04207 by Jianqing Fan, Jinhang Chai, Kaizheng Wang, Yaqi Duan.

Figure 1
Figure 1. Figure 1: Algorithmic diagram, where ϕ(u) = u− 1−F(u) F′(u) is a crutical component in optimal pricing. 4 view at source ↗
Figure 2
Figure 2. Figure 2: Regret exponent as a function of smoothness β ∈ [2, 5]. Lower is better. Our rate saturates at the parametric regime T 1/2 for β ≥ 2.5. Wang & Chen (2025) gives the same rate T 3/5 only at β = 2. Also define the left and right v-shrink points: lv[z, z] := z + v(z − z), rv[z, z] := z + (1 − v)(z − z). To avoid overburdening notation, we do not mark parameters with ⋆ for ground truth. 2 Stagewise Dynamic Pri… view at source ↗
Figure 3
Figure 3. Figure 3: Ground truth functions of the noise CDF, with smoothness β = 2, and 10 knots Evaluation For each horizon T and smoothness parameter β, we report the average cu￾mulative regret over Ntrials = 200 independent runs. We plot the curves on a log–log scale and estimate the slope using linear regression. In addition, we apply the cluster bootstrap method (Cameron et al., 2008), performing 2,000 bootstrap refits t… view at source ↗
Figure 4
Figure 4. Figure 4: Known utility The x-axis stands for total time horizon, and the y-axis stands for total regret. The solid lines represent average regrets, and the shaded areas reflect the standard deviations across trials. Here, blue, orange, blue, red, purple, and brown components correspond to the cases with β = 2, 2.25, 2.5, 2.75, 3, 3.25 respectively. β slope Theory CI [ 2.00 0.595 0.600 [0.540, 0.649] 2.25 0.552 0.54… view at source ↗
Figure 5
Figure 5. Figure 5: Unknown utility. The solid lines represent average regrets, and the shaded areas reflect the standard deviations across trials. β slope Theory CI [ 2.00 0.582 0.600 [0.555, 0.609] 2.25 0.533 0.545 [0.499, 0.568] 2.50 0.512 0.500 [0.488, 0.536] 2.75 0.518 0.500 [0.494, 0.542] 3.00 0.516 0.500 [0.492, 0.540] 3.25 0.509 0.500 [0.486, 0.532] view at source ↗
Figure 6
Figure 6. Figure 6: Unknown utility. Regret comparison of ILPR and kernel-based policy (Fan et al., 2024)in simulation. Our ILPR achieves tremendous benefit. 4.2 Semi-real data We next construct a semi-real online pricing environment using the real data available at the INFORMS 2023 BSS Data Challenge Competition. We define a binary demand indicator that is equal to one if the daily units ordered are non-zero. To ensure relia… view at source ↗
Figure 7
Figure 7. Figure 7: Unknown utility. Regret comparison of ILPR and kernel-based policy (Fan et al., 2024) in a semi-real environment with real data. Our ILPR outperforms the kernel￾based policy and DIP policy. For product-level performance comparison, view at source ↗
Figure 8
Figure 8. Figure 8: Histogram of regret improvement at T = 700. The improvement is huge across products. would be of interest to relax the feature diversity condition or develop robust variants under weaker support assumptions, especially since effective learning primarily relies on regions of the covariate space that are frequently visited. Finally, extending the framework to richer settings—such as more general demand model… view at source ↗
read the original abstract

We study contextual dynamic pricing under a semiparametric demand model in which the purchase probability is $1-F(p-m(\mathbf{x}))$, where $m(\mathbf{x})$ captures mean utility as a function of product features and buyer covariates, and $F$ is an unknown market-noise distribution. Existing methods either incur suboptimal regret or rely on restrictive structural assumptions. We propose a stagewise greedy pricing algorithm that iteratively refines the estimate of $F$ via local polynomial regression while pricing greedily with current estimates. By exploiting feature diversity, the algorithm reuses endogenous samples collected during exploitation for nonparametric estimation, avoiding costly global random exploration used in prior work. We establish a general regret bound that applies to any estimator $\hat m$ of the utility function, and derive explicit rates for linear, nonparametric additive, and sparse linear classes of $m$. For the linear class, our regret scales as $T^{\max\{1/2,\,3/(2\beta+1)\}}$, where $\beta$ is the smoothness of $F$ and $T$ is the time horizon. This improves the best known rates for semiparametric contextual pricing and achieves the parametric $\sqrt{T}$ rate when $\beta \ge 5/2$. We further prove a matching lower bound, showing the optimality of our rate, and present numerical experiments that corroborate the theory and demonstrate the practical advantages of iterative refinement.

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 / 2 minor

Summary. The paper studies contextual dynamic pricing under the semiparametric demand model with purchase probability 1-F(p-m(x)), where m(x) is the mean utility and F is an unknown noise distribution. It proposes a stagewise greedy pricing algorithm that iteratively refines the estimate of F via local polynomial regression while pricing greedily, exploiting feature diversity to reuse endogenous samples and avoid global random exploration. The authors establish a general regret bound applicable to any estimator of m, derive explicit rates including T^{max{1/2, 3/(2β+1)}} for linear m (with β the smoothness of F), prove a matching lower bound establishing optimality, and present numerical experiments.

Significance. If the central claims hold, the work improves the best known regret rates for semiparametric contextual pricing and achieves the parametric √T rate when β ≥ 5/2. The matching lower bound provides an independent optimality check, and the feature-diversity mechanism for reusing endogenous data offers a practical advance over prior exploration-heavy approaches.

major comments (2)
  1. [§2.3 (Feature Diversity Assumption)] §2.3 (Feature Diversity Assumption): The condition is load-bearing for the explicit rate in the linear case, as it must guarantee that the induced marginal on the argument of F has positive density and sufficient local mass at each stage so that local polynomial regression of F achieves the 3/(2β+1) rate without extra exploration cost. The current formulation does not explicitly quantify the minimal local sample size relative to the price grid or variance of m(x); if this mass is too small, the bias-variance tradeoff degrades and the general regret bound no longer yields the claimed T^{max{1/2, 3/(2β+1)}} rate.
  2. [Theorem 4.1 and §4.2 (Stagewise Analysis)] Theorem 4.1 and §4.2 (Stagewise Analysis): The derivation of the explicit rate for linear m invokes the diversity condition to control the nonparametric estimation error of F across stages. A more precise accounting of how the greedy price updates affect the local density around each evaluation point of F, and how this interacts with the bandwidth choice in local polynomials, is needed to confirm that the max{1/2, 3/(2β+1)} term is attained without hidden logarithmic or constant factors that would alter the optimality claim.
minor comments (2)
  1. [Abstract] Abstract: The description of the stagewise algorithm could briefly note how the iterative refinement of F interacts with the greedy pricing step to make the contribution clearer to readers.
  2. [§6 (Numerical Experiments)] §6 (Numerical Experiments): Additional details on bandwidth selection for the local polynomial estimator and sensitivity to the feature diversity parameter would help reproducibility and illustrate robustness.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on the feature diversity assumption and the stagewise analysis. We address each point below and will incorporate the suggested clarifications into the revised manuscript.

read point-by-point responses

  1. Referee: [§2.3 (Feature Diversity Assumption)] The condition is load-bearing for the explicit rate in the linear case, as it must guarantee that the induced marginal on the argument of F has positive density and sufficient local mass at each stage so that local polynomial regression of F achieves the 3/(2β+1) rate without extra exploration cost. The current formulation does not explicitly quantify the minimal local sample size relative to the price grid or variance of m(x); if this mass is too small, the bias-variance tradeoff degrades and the general regret bound no longer yields the claimed T^{max{1/2, 3/(2β+1)}} rate.

    Authors: We agree that the implications of Assumption 2.3 can be made more explicit. The assumption ensures that the features are diverse enough for the induced distribution on p - m(x) to have density bounded away from zero over the relevant range, which guarantees that the local sample size for estimating F grows linearly with stage length. In the proof of Theorem 4.1 we invoke this lower bound to obtain the standard local-polynomial rate 3/(2β+1) without additional exploration. To address the referee’s concern we will add a remark immediately after Assumption 2.3 that quantifies the minimal local mass in terms of the diversity parameter and the bounded variance of m(x), and we will verify that this mass suffices for the bias-variance tradeoff used in the regret analysis. revision: yes

  2. Referee: [Theorem 4.1 and §4.2 (Stagewise Analysis)] The derivation of the explicit rate for linear m invokes the diversity condition to control the nonparametric estimation error of F across stages. A more precise accounting of how the greedy price updates affect the local density around each evaluation point of F, and how this interacts with the bandwidth choice in local polynomials, is needed to confirm that the max{1/2, 3/(2β+1)} term is attained without hidden logarithmic or constant factors that would alter the optimality claim.

    Authors: We appreciate the request for a more granular accounting. In §4.2 the proof proceeds by induction over stages: the estimation errors of m̂ and F̂ control the deviation of the greedy price from the myopic optimum, which in turn keeps sampled points inside a neighborhood where the density of the argument to F remains uniformly bounded below by a positive constant derived from Assumption 2.3. The bandwidth is set to h = n^{-1/(2β+1)} with n the cumulative sample size up to the current stage, and standard local-polynomial bounds are applied under this density lower bound. We will expand the appendix with an auxiliary lemma that explicitly tracks the evolution of the minimal local density across stages and shows that the resulting regret terms produce exactly the claimed T^{max{1/2, 3/(2β+1)}} rate with no extra logarithmic factors. This will also confirm that the matching lower bound remains valid. revision: yes

Circularity Check

0 steps flagged

No circularity: general bound + independent lower bound

full rationale

The derivation begins with a general regret bound that holds for arbitrary estimators of m(x) and then specializes the rate under explicit classes of m by invoking standard local-polynomial convergence under the stated feature-diversity assumption. The matching lower bound is established separately and does not rely on the upper-bound construction. No equation reduces a claimed prediction to a fitted parameter by definition, no load-bearing premise collapses to a self-citation, and the diversity condition is an external modeling assumption rather than a quantity defined in terms of the regret itself.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the semiparametric demand model and a feature diversity condition that enables consistent estimation from endogenous data. Smoothness of F is parameterized by beta but treated as given.

axioms (2)
  • domain assumption Purchase probability follows 1 - F(p - m(x)) with F unknown but beta-smooth and m belonging to linear, additive, or sparse linear classes.
    Core modeling assumption stated in the abstract.
  • domain assumption Feature diversity is sufficient to allow nonparametric estimation of F from samples collected under greedy pricing without global random exploration.
    Key technical assumption enabling the regret analysis and avoidance of prior exploration costs.

pith-pipeline@v0.9.0 · 5562 in / 1500 out tokens · 32872 ms · 2026-05-08T17:30:05.135352+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

80 extracted references · 15 canonical work pages

  1. [1]

    2020 , publisher=

    Statistical foundations of data science , author=. 2020 , publisher=

    2020

  2. [2]

    Journal of the American Statistical Association , volume=

    Convexity, classification, and risk bounds , author=. Journal of the American Statistical Association , volume=. 2006 , publisher=

    2006

  3. [3]

    Operations Research , volume=

    Close the gaps: A learning-while-doing algorithm for single-product revenue management problems , author=. Operations Research , volume=. 2014 , publisher=

    2014

  4. [4]

    The review of economics and statistics , volume=

    Bootstrap-based improvements for inference with clustered errors , author=. The review of economics and statistics , volume=. 2008 , publisher=

    2008

  5. [5]

    arXiv preprint arXiv:2502.05776 , year=

    Dynamic Pricing in the Linear Valuation Model using Shape Constraints , author=. arXiv preprint arXiv:2502.05776 , year=

    work page arXiv

  6. [6]

    Mathematics of Operations Research , volume=

    Distribution-free contextual dynamic pricing , author=. Mathematics of Operations Research , volume=. 2024 , publisher=

    2024

  7. [7]

    2020 , publisher=

    Bandit algorithms , author=. 2020 , publisher=

    2020

  8. [8]

    Management Science , volume=

    Mostly exploration-free algorithms for contextual bandits , author=. Management Science , volume=. 2021 , publisher=

    2021

  9. [9]

    Annals of Statistics , year=

    Simultaneous analysis of Lasso and Dantzig selector , author=. Annals of Statistics , year=

  10. [10]

    2012 , journal=

    A unified framework for high-dimensional analysis of M-estimators with decomposable regularizers , author=. 2012 , journal=

    2012

  11. [11]

    Tsybakov , title =

    Alexandre B. Tsybakov , title =. 2009 , publisher =

    2009

  12. [12]

    2019 , publisher=

    High-dimensional statistics: A non-asymptotic viewpoint , author=. 2019 , publisher=

    2019

  13. [13]

    The Journal of Machine Learning Research , volume=

    Minimax-optimal rates for sparse additive models over kernel classes via convex programming , author=. The Journal of Machine Learning Research , volume=. 2012 , publisher=

    2012

  14. [14]

    Foundations and Trends

    An introduction to matrix concentration inequalities , author=. Foundations and Trends. 2015 , publisher=

    2015

  15. [15]

    Econometric theory , volume=

    Nonparametric instrumental regression with errors in variables , author=. Econometric theory , volume=. 2018 , publisher=

    2018

  16. [16]

    The Annals of Statistics , pages=

    Nonparametric regression with errors in variables , author=. The Annals of Statistics , pages=. 1993 , publisher=

    1993

  17. [17]

    Available at SSRN 5133677 , year=

    Tight Regret Bounds in Contextual Pricing with Semi-parametric Demand Learning , author=. Available at SSRN 5133677 , year=

  18. [18]

    Fan, Jianqing and Gijbels, Irene , title =

  19. [19]

    arXiv preprint arXiv:2110.01602 , year=

    Clustering a mixture of gaussians with unknown covariance , author=. arXiv preprint arXiv:2110.01602 , year=

    work page arXiv

  20. [20]

    Introduction to the non-asymptotic analysis of random matrices

    Introduction to the non-asymptotic analysis of random matrices , author=. arXiv preprint arXiv:1011.3027 , year=

    work page Pith review arXiv

  21. [21]

    arXiv preprint arXiv:2302.10160 , year=

    Pseudo-labeling for Kernel Ridge Regression under Covariate Shift , author=. arXiv preprint arXiv:2302.10160 , year=

    work page arXiv

  22. [22]

    arXiv preprint arXiv:2111.08221 , year=

    Fairness-aware online price discrimination with nonparametric demand models , author=. arXiv preprint arXiv:2111.08221 , year=

    work page arXiv

  23. [23]

    International Conference on Artificial Intelligence and Statistics , pages=

    Doubly Fair Dynamic Pricing , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2023 , organization=

    2023

  24. [24]

    Production and Operations Management (Forthcoming) , year=

    Network revenue management with demand learning and fair resource-consumption balancing , author=. Production and Operations Management (Forthcoming) , year=

  25. [25]

    International Conference on Machine Learning , pages=

    Regularized online allocation problems: Fairness and beyond , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  26. [26]

    Operations research , volume=

    Dynamic pricing with an unknown demand model: Asymptotically optimal semi-myopic policies , author=. Operations research , volume=. 2014 , publisher=

    2014

  27. [27]

    Management Science , volume=

    Personalized dynamic pricing with machine learning: High-dimensional features and heterogeneous elasticity , author=. Management Science , volume=. 2021 , publisher=

    2021

  28. [28]

    Management Science , volume=

    Multimodal dynamic pricing , author=. Management Science , volume=. 2021 , publisher=

    2021

  29. [29]

    Management Science , volume=

    Stochastic optimization forests , author=. Management Science , volume=. 2023 , publisher=

    2023

  30. [30]

    Available at SSRN 3930622 , year=

    Dynamic pricing with fairness constraints , author=. Available at SSRN 3930622 , year=

  31. [31]

    Conference on learning theory , pages=

    Non-stationary reinforcement learning without prior knowledge: An optimal black-box approach , author=. Conference on learning theory , pages=. 2021 , organization=

    2021

  32. [32]

    Temporal Fairness in Learning and Earning: Price Protection Guarantee and Phase Transitions: Feng, Qing| uZhu, Ruihao| uJasin, Stefanus , year=

    Temporal Fairness in Learning and Earning: Price Protection Guarantee and Phase Transitions , author=. Temporal Fairness in Learning and Earning: Price Protection Guarantee and Phase Transitions: Feng, Qing| uZhu, Ruihao| uJasin, Stefanus , year=

  33. [33]

    arXiv preprint arXiv:2010.02521 , year=

    Doubly robust covariate shift regression with semi-nonparametric nuisance models , author=. arXiv preprint arXiv:2010.02521 , year=

    work page arXiv 2010

  34. [34]

    arXiv preprint arXiv:2208.05134 , year=

    Doubly robust augmented model accuracy transfer inference with high dimensional features , author=. arXiv preprint arXiv:2208.05134 , year=

    work page arXiv

  35. [35]

    arXiv preprint arXiv:2209.04977 , year=

    Semi-supervised Triply Robust Inductive Transfer Learning , author=. arXiv preprint arXiv:2209.04977 , year=

    work page arXiv

  36. [36]

    Journal of the American Statistical Association , volume=

    Embracing the blessing of dimensionality in factor models , author=. Journal of the American Statistical Association , volume=. 2018 , publisher=

    2018

  37. [37]

    2023 , eprint=

    Utility Fairness in Contextual Dynamic Pricing with Demand Learning , author=. 2023 , eprint=

    2023

  38. [38]

    International Conference on Machine Learning , pages=

    Online pricing with offline data: Phase transition and inverse square law , author=. International Conference on Machine Learning , pages=. 2020 , organization=

    2020

  39. [39]

    The Annals of Statistics , volume=

    Transfer learning for contextual multi-armed bandits , author=. The Annals of Statistics , volume=. 2024 , publisher=

    2024

  40. [40]

    Advances in neural information processing systems , volume=

    Improved algorithms for linear stochastic bandits , author=. Advances in neural information processing systems , volume=

  41. [41]

    Applications of the van Trees inequality: a Bayesian Cram

    Gill, Richard D and Levit, Boris Y , journal=. Applications of the van Trees inequality: a Bayesian Cram. 1995 , publisher=

    1995

  42. [42]

    Surveys in operations research and management science , volume=

    Dynamic pricing and learning: historical origins, current research, and new directions , author=. Surveys in operations research and management science , volume=. 2015 , publisher=

    2015

  43. [43]

    Management Science , volume=

    A statistical learning approach to personalization in revenue management , author=. Management Science , volume=. 2022 , publisher=

    2022

  44. [44]

    Available at SSRN 4140550 , year=

    Context-based dynamic pricing with separable demand models , author=. Available at SSRN 4140550 , year=

  45. [45]

    Management Science , volume=

    On the (surprising) sufficiency of linear models for dynamic pricing with demand learning , author=. Management Science , volume=. 2015 , publisher=

    2015

  46. [46]

    International Conference on Artificial Intelligence and Statistics , pages=

    Towards agnostic feature-based dynamic pricing: Linear policies vs linear valuation with unknown noise , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2022 , organization=

    2022

  47. [47]

    Operations Research , volume=

    Nonparametric pricing analytics with customer covariates , author=. Operations Research , volume=. 2021 , publisher=

    2021

  48. [48]

    arXiv preprint arXiv:1604.07463 , year=

    Dynamic pricing with demand covariates , author=. arXiv preprint arXiv:1604.07463 , year=

    work page arXiv

  49. [49]

    Journal of the American Statistical Association , pages=

    Policy optimization using semiparametric models for dynamic pricing , author=. Journal of the American Statistical Association , pages=. 2024 , publisher=

    2024

  50. [50]

    Operations Research , volume=

    Dynamic pricing under a general parametric choice model , author=. Operations Research , volume=. 2012 , publisher=

    2012

  51. [51]

    Advances in neural information processing systems , volume=

    A smoothed analysis of the greedy algorithm for the linear contextual bandit problem , author=. Advances in neural information processing systems , volume=

  52. [52]

    CoRR , volume=

    Practical Evaluation and Optimization of Contextual Bandit Algorithms , author=. CoRR , volume=

  53. [53]

    International Conference on Artificial Intelligence and Statistics , pages=

    Stochastic linear contextual bandits with diverse contexts , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2020 , organization=

    2020

  54. [54]

    1990 , institution=

    Efficient memory-based learning for robot control , author=. 1990 , institution=

    1990

  55. [55]

    arXiv preprint arXiv:2109.00157 , year=

    A survey of exploration methods in reinforcement learning , author=. arXiv preprint arXiv:2109.00157 , year=

    work page arXiv

  56. [56]

    Journal of Machine Learning Research , volume=

    Dynamic pricing in high-dimensions , author=. Journal of Machine Learning Research , volume=

  57. [57]

    Management Science , volume=

    Dynamic pricing with demand learning and reference effects , author=. Management Science , volume=. 2022 , publisher=

    2022

  58. [58]

    Management Science , volume=

    Meta dynamic pricing: Transfer learning across experiments , author=. Management Science , volume=. 2022 , publisher=

    2022

  59. [59]

    2018 , publisher=

    High-dimensional probability: An introduction with applications in data science , author=. 2018 , publisher=

    2018

  60. [60]

    Randomized sketches for kernels: Fast and optimal nonparametric regression , author=

  61. [61]

    Wainwright , TITLE =

    Yaqi Duan and Mengdi Wang and Martin J. Wainwright , TITLE =. Ann. Statist. , FJOURNAL =. 2024 , VOLUME =. doi:10.1214/24-AOS2399 , SICI =

    work page doi:10.1214/24-aos2399 2024

  62. [62]

    arXiv preprint arXiv:2211.03899 , year=

    Policy evaluation from a single path: Multi-step methods, mixing and mis-specification , author=. arXiv preprint arXiv:2211.03899 , year=

    work page arXiv

  63. [63]

    Annals of Statistics , pages=

    Local Rademacher Complexities , author=. Annals of Statistics , pages=. 2005 , publisher=

    2005

  64. [64]

    Journal of Machine Learning Research , volume=

    Using confidence bounds for exploitation-exploration trade-offs , author=. Journal of Machine Learning Research , volume=

  65. [65]

    Advances in neural information processing systems , volume=

    Bounded regret for finite-armed structured bandits , author=. Advances in neural information processing systems , volume=

  66. [66]

    Mobile health: sensors, analytic methods, and applications , pages=

    From ads to interventions: Contextual bandits in mobile health , author=. Mobile health: sensors, analytic methods, and applications , pages=. 2017 , publisher=

    2017

  67. [67]

    Management Science , volume=

    Feature-based dynamic pricing , author=. Management Science , volume=. 2020 , publisher=

    2020

  68. [68]

    Operations Research , volume=

    Nonstationary bandits with habituation and recovery dynamics , author=. Operations Research , volume=. 2020 , publisher=

    2020

  69. [69]

    Manufacturing & Service Operations Management , volume=

    Learning personalized product recommendations with customer disengagement , author=. Manufacturing & Service Operations Management , volume=. 2022 , publisher=

    2022

  70. [70]

    Operations Research , volume=

    MNL-bandit: A dynamic learning approach to assortment selection , author=. Operations Research , volume=. 2019 , publisher=

    2019

  71. [71]

    Operations Research , volume=

    Online decision making with high-dimensional covariates , author=. Operations Research , volume=. 2020 , publisher=

    2020

  72. [72]

    Management Science , volume=

    Dynamic learning and pricing with model misspecification , author=. Management Science , volume=. 2019 , publisher=

    2019

  73. [73]

    Mathematics of Operations Research , volume=

    A primal--dual learning algorithm for personalized dynamic pricing with an inventory constraint , author=. Mathematics of Operations Research , volume=. 2022 , publisher=

    2022

  74. [74]

    arXiv preprint arXiv:2303.07570 , year=

    High-Dimensional Dynamic Pricing under Non-Stationarity: Learning and Earning with Change-Point Detection , author=. arXiv preprint arXiv:2303.07570 , year=

    work page arXiv

  75. [75]

    Proceedings of IEEE 36th annual foundations of computer science , pages=

    Gambling in a rigged casino: The adversarial multi-armed bandit problem , author=. Proceedings of IEEE 36th annual foundations of computer science , pages=. 1995 , organization=

    1995

  76. [76]

    arXiv preprint arXiv:1803.06971 , year=

    What doubling tricks can and can't do for multi-armed bandits , author=. arXiv preprint arXiv:1803.06971 , year=

    work page arXiv

  77. [77]

    Journal of the American Statistical Association , volume=

    Are latent factor regression and sparse regression adequate? , author=. Journal of the American Statistical Association , volume=. 2024 , publisher=

    2024

  78. [78]

    Journal of Econometrics , volume=

    Factor-adjusted regularized model selection , author=. Journal of Econometrics , volume=. 2020 , publisher=

    2020

  79. [79]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Large covariance estimation by thresholding principal orthogonal complements , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2013 , publisher=

    2013

  80. [80]

    arXiv preprint arXiv:2412.19252 , year=

    Localized exploration in contextual dynamic pricing achieves dimension-free regret , author=. arXiv preprint arXiv:2412.19252 , year=

    work page arXiv