Pith. sign in

REVIEW 3 major objections 5 minor 54 references

A minimal directional perturbation of the max revenue gap yields a valid p-value for whether a learned assortment optimizer satisfies a structural constraint.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 09:54 UTC pith:OESCRXLV

load-bearing objection Solid, usable theory for a real irregular problem: max-difference directional perturbation after adaptive high-dim assortment learning, with complete proofs and clear power gains over uniform calibration. the 3 major comments →

arxiv 2607.05024 v1 pith:OESCRXLV submitted 2026-07-06 math.ST stat.TH

Post-Learning Inference for Combinatorial Optimizers with High-Dimensional Sparse Contextual Information via Minimal Directional Perturbation

Pengyu Li , Shuting Shen This is my paper
classification math.ST stat.TH MSC 62F0362F0762J1262J0762L10
keywords irregular inferencehigh-dimensional inferenceadaptive data collectiondirectional perturbation testpost-regularization inferencecombinatorial optimizationcontextual multinomial logitassortment optimization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

After adaptive high-dimensional learning of customer preferences, a platform may care less about the exact best product assortment than about whether some optimal assortment still meets a business rule: include core products, keep category balance, or obey an inventory screen. Because the map from preferences to the best assortment jumps at ties, ordinary confidence intervals do not work. This paper reduces the structural question to the sign of a single nonsmooth number—the gap between the best revenue inside the rule class and the best revenue outside it—and builds a p-value by asking how large a random directional nudge of the estimated revenue surface is needed before that gap becomes compatible with the null. The data are gathered by an online sparse likelihood policy that also keeps cumulative revenue loss sublinear. Under a localized signal condition the resulting p-value controls size and has power, while avoiding the conservatism of calibrating uniform error over every feasible assortment.

Core claim

For a high-dimensional contextual multinomial logit model under adaptive assortment selection, the structural hypothesis that the terminal oracle optimizer intersects a prescribed combinatorial class is equivalent to non-negativity of a max-difference revenue functional. A p-value formed from the minimal radius of random unit-sphere perturbations of the debiased terminal revenue surface on the selected support is asymptotically valid under the null and consistent under a localized alternative, without uniform error control over the full candidate class.

What carries the argument

The minimal directional perturbation radius UT: the smallest a ≥ 0 such that, along at least one of m random unit directions on the selected support, the perturbed max-difference of null versus alternative plug-in revenues exceeds -κ; the p-value is the χ^{2} upper tail of UT^{2} plus a directional discretization remainder.

Load-bearing premise

The main rates and validity proofs rest on a primitive condition that revenues are Gaussian and the feasible assortment class is either the full K-subset class with bounded choice-probability ratios or a restricted-intersection subclass; that condition supplies the local Hessian stability and anti-concentration used throughout.

What would settle it

In the reported simulation design (n=20, p=500, K=3, s*=3–5, least-favorable null with revenue gap exactly zero), if the empirical Type-I rate of the proposed p-value stays systematically above 0.05 as T grows to 2000 while the uniform-error baseline remains near zero, the asymptotic size claim fails in the regime the theory targets.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper develops post-learning inference for structural properties of combinatorial optimizers in a high-dimensional sparse contextual MNL assortment model under adaptive data collection. The target is whether the terminal oracle optimizer intersects a prescribed structural class S0 (product inclusion, category proportion, feature screens). The authors reformulate this as a sign test on a nonsmooth max-difference revenue functional and propose a minimal directional perturbation p-value: after ℓ1-penalized online likelihood estimation and debiasing on the selected support, random unit directions capture angular uncertainty while the minimal radius UT needed to reach the null boundary is calibrated via a χ²_bs tail plus a directional residual δm. Theory provides uniform rates (Thm 4.1), effective support recovery (Cor 4.2), a debiased martingale expansion (Cor 4.4), and asymptotic size/power under localization to near-boundary assortments (Thms 4.5–4.6), using anti-concentration for Gaussian maxima differences and martingale Gaussian coupling. Simulations show size control at large T and substantially higher power than a uniform-error-bound baseline across three structural tests, with sublinear regret for the adaptive policy.

Significance. If the results hold, the paper supplies a usable inferential tool for a practically important and statistically irregular problem: testing discrete structural properties of data-dependent combinatorial optimizers after adaptive high-dimensional learning, where the parameter-to-optimizer map is discontinuous and standard Wald/delta-method tools fail. The localization of uncertainty to the max-difference boundary (rather than uniform control over SK) is a clear conceptual and power advantage over confidence-set inversion. The technical apparatus—anti-concentration for differences of maxima under adaptive selection, martingale coupling of the adaptive score, and effective-support debiased expansions—is of independent interest for irregular post-selection and post-adaptive-sampling inference. Explicit non-asymptotic rates, full validity/power proofs, and a regret guarantee that embeds inference in online learning without a pure exploration phase are genuine strengths. The contribution is well positioned at the intersection of high-dimensional inference, combinatorial optimization, and sequential decision-making.

major comments (3)
  1. [Assumption 4.4, Remark 4.1, Claim S.3.1] Assumption 4.4 (Gaussian revenues plus either full K-subsets with ρ≤2 or a restricted-intersection condition on SK) is load-bearing for the local Hessian stability bound that drives the Cn rates in Lemma S.2.1, Claim S.3.1, the debiased expansion (Cor 4.4), and the T-thresholds of Theorems 4.5–4.6. Remark 4.1 correctly flags sufficiency and sketches a Stein-kernel relaxation, but the main theorems are not proved under any weaker primitive. For a paper that advertises a general max-difference perturbation principle for combinatorial optimizers, the manuscript should either (i) state the theorems under an abstract local-Hessian-stability condition with Ass 4.4 as a corollary, or (ii) supply a complete proof under a clearly weaker revenue/tail condition. As written, the scope of the validity claim is narrower than the introduction suggests.
  2. [Theorem 4.5, Remark 4.6, Section 5.1] The theoretically sufficient order κ ≍ σ_{vT,rT} √(s*/(Tλ)) depends on the unknown local gradient scale σ_{vT,rT} and λ_min(Σ*). The implementable choice κ = Cκ √(bs/(Tϵ)) with Cκ = 10^{-4} works in the reported simulations, but Theorem 4.5’s remainder absorption (display (S.2.35), (S.2.40), (S.2.42)) requires κ to dominate several higher-order terms involving Cn, ν, and ηT. The paper needs a clearer finite-sample prescription or a data-driven rule for Cκ (or for estimating σ_{vT,rT} on the localized sets S̄0, S̄1) so that practitioners can verify the conditions under which the o(1) size guarantee is expected to kick in.
  3. [Section 5.1, Remark 2.1] Size is evaluated only at the least-favorable boundary Δ* = 0 obtained by bisection on a single terminal revenue coordinate (Section 5.1). This is informative for Type I control at the knife-edge, but it does not address size under interior nulls (Δ* > 0) or under the stronger null S*_T ⊆ S0 of Remark 2.1. At least one additional size panel under unmodified null contexts with Δ* > 0, and a brief discussion of how the procedure would be modified for the strong null, would make the empirical size claim more complete.
minor comments (5)
  1. [Section 3.2, Theorem 4.5] Notation for the selected support size switches between bs and s* after support recovery; a single convention after Corollary 4.2 would reduce cognitive load.
  2. [Figures 1–3] Figures 1–3 truncate the size axis at 0.10; a short note in the caption that early-horizon size can exceed this range (as already stated in the text) would prevent misreading of the plots.
  3. [Section 5.1, Eq. (16)] The directional residual δm is defined with ϵ in (16), but the simulation protocol fixes δm = 0.002 and backs out m (capped at 50,000). State explicitly whether the cap ever binds for s ∈ {3,4,5} and what is done if it does.
  4. [References] Several references to working manuscripts ([6], [5]) are central to the anti-concentration and comparison arguments; ensure arXiv or published versions are cited if available at revision.
  5. [Title page] Typographical: “COMBINA TORIAL” and “INFORMA TION” in the title header appear to have spurious spaces; “PENGYULI” / “SHUTINGSHEN” spacing in the author line should be cleaned.

Circularity Check

0 steps flagged

No significant circularity: the p-value is a constructed test statistic whose asymptotic size/power are proved from first-order expansions, martingale coupling, and external anti-concentration, not forced by definition or self-fit.

full rationale

The derivation chain is standard post-regularization inference under adaptive design, not circular. The target is the sign of the max-difference revenue gap (6). The procedure defines a debiased plug-in surface, a minimal directional perturbation radius UT (12), and a p-value pm = (χ²_bs tail of UT² + directional residual δm) ∧ 1 (15). Validity (Theorems 4.5–4.6) is proved by: (i) ℓ1-penalized online MLE rates and effective support recovery (Thm 4.1, Cor 4.2); (ii) a debiased martingale expansion on the selected support (Cor 4.4); (iii) linearization of the revenue surface (Lemma S.2.3); (iv) martingale Gaussian coupling (Lemma S.2.2, via [12]); (v) localization to near-boundary assortments S̄0, S̄1 so that remainders are absorbed by κ; and (vi) spherical-cap covering for random directions (Lemma S.2.4). The χ² calibration and δm are external probability facts, not quantities defined from the target gap. Simulation tuning of (Cλ, ϵ, Cκ, δm) is on held-out replications and does not enter the asymptotic statements. Self-citations ([5] anti-concentration for Gaussian maxima differences; [47]/[6] related assortment inference) supply mathematical tools or related-work context; they do not assume the paper’s size/power conclusion, and the anti-concentration application in Claim S.3.1 is under explicit primitive conditions (Assumption 4.4) rather than a uniqueness theorem that forbids alternatives. No equation reduces the claimed size control to a fitted input or to a definition of the target. Score 0 is appropriate.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 2 invented entities

Central asymptotic claims rest on a standard high-dimensional sparse GLM setup plus several domain-specific regularity conditions needed for adaptive assortment selection and discontinuous argmax maps. Free parameters appear only in the finite-sample implementation and simulation protocol; the asymptotic statements are free of fitted constants once the order conditions on T, ηT, m, κ are met.

free parameters (4)
  • C_λ (penalty multiplier)
    Chosen by three-stage Monte-Carlo grid (final value 0.40) to set λ_{t−1}; enters the practical estimator but only through a sufficiently-large-constant requirement in the theorems.
  • C_κ (threshold multiplier for κ)
    Set to 10^{-4} after held-out calibration; absorbs the unknown local scale σ/√λ that appears in the theoretical order of κ.
  • ϵ (directional accuracy)
    Fixed at 0.50 in simulations; controls the spherical-cap covering probability that determines m and δm.
  • δ_m (directional residual)
    Set to 0.002, which determines the number of unit-sphere directions m (capped at 50 000).
axioms (6)
  • domain assumption Contexts and revenues {v_t,r_t} are i.i.d. across t and mutually independent (Ass 4.1).
    Used throughout for population Hessian Σ* and for conditioning arguments.
  • domain assumption λ_min(Σ*)≥λ>0, features uniformly bounded, and local MNL probability ratios bounded by ρ (Ass 4.2).
    Standard restricted-eigenvalue / bounded-design conditions for high-dim MNL.
  • domain assumption Choice outcomes are conditionally independent of history given current context, revenue and offered set (Ass 4.3).
    Makes the score process a martingale difference sequence.
  • ad hoc to paper Revenues are i.i.d. Gaussian and SK satisfies either the full K-subset + ρ≤2 condition or a restricted-intersection condition (Ass 4.4).
    Primitive sufficient condition for local Hessian stability and anti-concentration of max-revenue differences; authors discuss possible Stein-kernel relaxations but prove under this form.
  • domain assumption Effective support I* obeys a beta-min condition and weak-tail ℓ1 mass η_T→0; modified mutual incoherence holds on I* (Ass 4.5–4.6).
    Standard effective-sparsity and irrepresentability conditions adapted to the online MNL setting.
  • domain assumption Pilot estimator satisfies ∥β̂0−β*∥1≤τ with τ small enough relative to the estimation rate.
    Anchors the local constrained MLE and is assumed known.
invented entities (2)
  • Minimal directional perturbation radius U_T and the associated p-value p_m no independent evidence
    purpose: Converts the nonsmooth max-difference testing problem into a scalar statistic whose null distribution is approximated by a χ² tail plus a directional residual.
    Defined constructively from the debiased revenue surface and random unit directions; no independent physical existence claimed outside the statistical procedure.
  • Localized near-boundary assortment sets S̄0(v_T,r_T) and S̄1(v_T,r_T) no independent evidence
    purpose: Restricts error control and power analysis to assortments that can become maximizers after statistical-scale perturbations.
    Analytic devices introduced for the proofs; not claimed to be observable objects.

pith-pipeline@v1.1.0-grok45 · 56724 in / 3448 out tokens · 32669 ms · 2026-07-11T09:54:54.321929+00:00 · methodology

0 comments
read the original abstract

We study post-learning inference for structural properties of data-dependent combinatorial optimizers. The target is whether an oracle optimizer, rather than a latent parameter or smooth functional, belongs to a prescribed class, such as a category-mix, inventory, or resource-feasibility class. We focus on a high-dimensional contextual multinomial logit model with sequentially adaptive data collection, where the parameter-to-optimizer map is discontinuous and the policy induces temporal dependence. We propose a novel perturbation test based on a nonsmooth max-difference revenue statistic comparing the best null assortment with the best alternative assortment. The test perturbs the estimated terminal revenue surface on the selected support: random unit directions capture directional uncertainty, while the minimal perturbation radius captures magnitude uncertainty and yields a p-value. This localizes inference near the null--alternative boundary and avoids uniform error control over the full candidate class. The data are collected by an \(\ell_1\)-penalized online likelihood policy that performs variable selection while controlling regret. Using a new anti-concentration argument for Gaussian maxima differences and martingale Gaussian coupling, we establish uniform estimation rates, effective support recovery, and asymptotic validity of the proposed p-value under adaptive assortment selection. We prove asymptotic size control and power consistency under a localized signal condition.

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

54 extracted references · 4 canonical work pages

  1. [1]

    and ZEEVI, A

    AGRAWAL, S., AVADHANULA, V., GOYAL, V. and ZEEVI, A. (2017). Thompson Sampling for the MNL- Bandit. InProceedings of the 2017 Conference on Learning Theory (COLT).Proceedings of Machine Learning Research6576–78. PMLR

    2017

  2. [2]

    and ZEEVI, A

    AGRAWAL, S., AVADHANULA, V., GOYAL, V. and ZEEVI, A. (2019). MNL-Bandit: A Dynamic Learning Approach to Assortment Selection.Operations Research671453–1485

    2019

  3. [3]

    and MCCLOSKEY, A

    ANDREWS, I., KITAGAWA, T. and MCCLOSKEY, A. (2024). Inference on Winners.The Quarterly Journal of Economics139305–358. https://doi.org/10.1093/qje/qjad043

    doi:10.1093/qje/qjad043 2024

  4. [4]

    and SEGEV, D

    AOUAD, A., FARIAS, V., LEVI, R. and SEGEV, D. (2018). The Approximability of Assortment Optimization Under Ranking Preferences.Operations Research661661–1669

    2018

  5. [5]

    BELLONI, A., FANG, E. X. and SHEN, S. (2024). Anti-Concentration Inequalities for the Difference of Maxima of Gaussian Random Vectors

    2024

  6. [6]

    BELLONI, A., HAN, Y., FANG, E. X. and SHEN, S. (2025). Property Test on the Optimal Assortment in the Contextual Multinomial Logit Model with Adaptive Sampling. Working manuscript

    2025

  7. [7]

    and ZHAO, L

    BERK, R., BROWN, L., BUJA, A., ZHANG, K. and ZHAO, L. (2013). Valid Post-Selection Inference.The Annals of Statistics41802–837. https://doi.org/10.1214/12-AOS1077

    doi:10.1214/12-aos1077 2013

  8. [8]

    and ZEEVI, A

    BESBES, O. and ZEEVI, A. (2015). On the (Surprising) Sufficiency of Linear Models for Dynamic Pricing with Demand Learning.Management Science61723–739

    2015

  9. [9]

    H., GALLEGO, G

    BLANCHET, J. H., GALLEGO, G. and GOYAL, V. (2016). A Markov Chain Approximation to Choice Modeling.Operations Research64886–905

    2016

  10. [10]

    P., TERWIESCH, C

    CACHON, G. P., TERWIESCH, C. and XU, Y. (2005). Retail Assortment Planning in the Presence of Consumer Search.Manufacturing & Service Operations Management7330–346

    2005

  11. [11]

    and GALLIEN, J

    CARO, F. and GALLIEN, J. (2007). Dynamic Assortment with Demand Learning for Seasonal Consumer Goods.Management Science53276–292

    2007

  12. [12]

    D., MASINI, R

    CATTANEO, M. D., MASINI, R. P. and UNDERWOOD, W. G. (2025). Yurinskii’s coupling for martingales. The Annals of Statistics532179–2203. https://doi.org/10.1214/25-AOS2538

    work page doi:10.1214/25-aos2538 2025

  13. [13]

    and ZHANG, Y

    CHANG, X., CHEN, X., LAI, Z., LI, H., LIU, Z. and ZHANG, Y. (2026). Online Statistical Inference for Contextual Bandits via Stochastic Gradient Descent.Journal of the American Statistical Association 1–14. https://doi.org/10.1080/01621459.2026.2621503

    doi:10.1080/01621459.2026.2621503 2026

  14. [14]

    and WANG, Y

    CHEN, X., KRISHNAMURTHY, A. and WANG, Y. (2024). Robust Dynamic Assortment Optimization in the Presence of Outlier Customers.Operations Research72999–1015. https://doi.org/10.1287/opre.2020. 0281

    doi:10.1287/opre.2020 2024

  15. [15]

    and XIN, L

    CHEN, X., MA, W., SIMCHI-LEVI, D. and XIN, L. (2024). Assortment Planning for Recommendations at Checkout Under Inventory Constraints.Mathematics of Operations Research49297–325. POST-LEARNING INFERENCE FOR COMBINATORIAL OPTIMIZERS57

    2024

  16. [16]

    and SIMCHI-LEVI, D

    CHEN, X., OWEN, Z., PIXTON, C. and SIMCHI-LEVI, D. (2022). A Statistical Learning Approach to Personalization in Revenue Management.Management Science681923–1937

    2022

  17. [17]

    and WANG, Y

    CHEN, X. and WANG, Y. (2018). A Note on a Tight Lower Bound for Capacitated MNL-Bandit Assortment Selection Models.Operations Research Letters46534–537

    2018

  18. [18]

    and ZHOU, Y

    CHEN, X., WANG, Y. and ZHOU, Y. (2020). Dynamic Assortment Optimization with Changing Contextual Information.Journal of Machine Learning Research211–44

    2020

  19. [19]

    and WANG, K

    CHEN, Y., FAN, J., MA, C. and WANG, K. (2019). Spectral Method and Regularized MLE Are Both Optimal for Top-K Ranking.The Annals of Statistics472204–2235

    2019

  20. [20]

    and KATO, K

    CHERNOZHUKOV, V., CHETVERIKOV, D. and KATO, K. (2013). Gaussian approximations and multiplier bootstrap for maxima of sums of high-dimensional random vectors.The Annals of Statistics412786–

    2013

  21. [21]

    https://doi.org/10.1214/13-AOS1161

    doi:10.1214/13-aos1161

  22. [22]

    and KATO, K

    CHERNOZHUKOV, V., CHETVERIKOV, D. and KATO, K. (2014). Anti-concentration and honest, adaptive confidence bands.The Annals of Statistics421787–1818. https://doi.org/10.1214/14-AOS1235

    work page doi:10.1214/14-aos1235 2014

  23. [23]

    and KATO, K

    CHERNOZHUKOV, V., CHETVERIKOV, D. and KATO, K. (2016). Empirical and multiplier bootstraps for suprema of empirical processes of increasing complexity, and related Gaussian couplings.Stochastic Processes and their Applications1263632–3651. In Memoriam: Evarist Giné. https://doi.org/10.1016/ j.spa.2016.04.009

    2016

  24. [24]

    and KATO, K

    CHERNOZHUKOV, V., CHETVERIKOV, D. and KATO, K. (2017). Central limit theorems and bootstrap in high dimensions.The Annals of Probability452309–2352. https://doi.org/10.1214/16-AOP1113

    doi:10.1214/16-aop1113 2017

  25. [25]

    and KATO, K

    CHERNOZHUKOV, V., CHETVERIKOV, D. and KATO, K. (2019). Inference on causal and structural parame- ters using many moment inequalities.The Review of Economic Studies861867–1900

    2019

  26. [26]

    and KOIKE, Y

    CHERNOZHUKOV, V., CHETVERIKOV, D., KATO, K. and KOIKE, Y. (2022). Improved central limit theorem and bootstrap approximations in high dimensions.The Annals of Statistics502562–2586. https: //doi.org/10.1214/22-AOS2193

    doi:10.1214/22-aos2193 2022

  27. [27]

    CHEUNG, W. C. and SIMCHI-LEVI, D. (2017). Thompson Sampling for Online Personalized Assortment Optimization Problems with Multinomial Logit Choice Models. Working paper, available at SSRN 3075658

    2017

  28. [28]

    and LIU, Q

    FAN, X., GRAMA, I. and LIU, Q. (2015). Exponential inequalities for martingales with applications. Electronic Journal of Probability201–22. https://doi.org/10.1214/EJP.v20-3496

    doi:10.1214/ejp.v20-3496 2015

  29. [29]

    and DUBEY, A

    GALLEGO, G., IYENGAR, G., PHILLIPS, R. and DUBEY, A. (2004). Managing Flexible Products on a Network. Working paper, Columbia University

    2004

  30. [30]

    and ZHANG, A

    GAO, C., SHEN, Y. and ZHANG, A. Y. (2023). Uncertainty Quantification in the Bradley–Terry–Luce Model. Information and Inference: A Journal of the IMA121073–1140. https://doi.org/10.1093/imaiai/iaac032

    doi:10.1093/imaiai/iaac032 2023

  31. [31]

    J., MONTERO, S., MOON, H

    GILLEN, B. J., MONTERO, S., MOON, H. R. and SHUM, M. (2019). BLP-2LASSO for Aggregate Discrete Choice Models with Rich Covariates.The Econometrics Journal22262–281. https://doi.org/10.1093/ ectj/utz010

    2019

  32. [32]

    and RUSMEVICHIENTONG, P

    GOLREZAEI, N., NAZERZADEH, H. and RUSMEVICHIENTONG, P. (2014). Real-Time Optimization of Personalized Assortments.Management Science601532–1551

    2014

  33. [33]

    GUADAGNI, P. M. and LITTLE, J. D. C. (1983). A Logit Model of Brand Choice Calibrated on Scanner Data.Marketing Science2203–238. https://doi.org/10.1287/mksc.2.3.203

    work page doi:10.1287/mksc.2.3.203 1983

  34. [34]

    and MONTANARI, A

    JAVANMARD, A. and MONTANARI, A. (2014). Confidence Intervals and Hypothesis Testing for High- Dimensional Regression.Journal of Machine Learning Research152869–2909

    2014

  35. [35]

    and ZHANG, D

    JIANG, Z., LI, J. and ZHANG, D. (2025). A High-Dimensional Choice Model for Online Retailing.Manage- ment Science713320–3339. https://doi.org/10.1287/mnsc.2020.02715

    work page doi:10.1287/mnsc.2020.02715 2025

  36. [36]

    G., FISHER, M

    KÖK, A. G., FISHER, M. L. and VAIDYANATHAN, R. (2015). Assortment Planning: Review of Literature and Industry Practice. InRetail Supply Chain Management175–236. Springer

    2015

  37. [37]

    D., SUN, D

    LEE, J. D., SUN, D. L., SUN, Y. and TAYLOR, J. E. (2016). Exact Post-Selection Inference, with Application to the Lasso.The Annals of Statistics44907–927. https://doi.org/10.1214/15-AOS1371

    doi:10.1214/15-aos1371 2016

  38. [38]

    LI, S. (2011). Concise Formulas for the Area and V olume of a Hyperspherical Cap.Asian Journal of Mathematics & Statistics466–70

    2011

  39. [39]

    LIU, Y., FANG, E. X. and LU, J. (2023). Lagrangian inference for ranking problems.Operations Research 71202–223

    2023

  40. [40]

    K., LEVY, M., KAHN, B

    MANTRALA, M. K., LEVY, M., KAHN, B. E., FOX, E. J., GAIDAREV, P., DANKWORTH, B. and SHAH, D. (2009). Why Is Assortment Planning So Difficult for Retailers? A Framework and Research Agenda. Journal of Retailing8571–83

    2009

  41. [41]

    MCFADDEN, D. (1974). Conditional Logit Analysis of Qualitative Choice Behavior. InFrontiers in Econo- metrics(P. Zarembka, ed.) 105–142. Academic Press, New York

    1974

  42. [42]

    and SHAH, D

    NEGAHBAN, S., OH, S. and SHAH, D. (2017). Rank Centrality: Ranking from Pairwise Comparisons. Operations Research65266–287. 58P. LI AND S. SHEN

    2017

  43. [43]

    NEGAHBAN, S., OH, S., THEKUMPARAMPIL, K. K. and XU, J. (2018). Learning from Comparisons and Choices.Journal of Machine Learning Research191–95

    2018

  44. [44]

    RUSMEVICHIENTONG, P., SHEN, Z.-J. M. and SHMOYS, D. B. (2010). Dynamic Assortment Optimization with a Multinomial Logit Choice Model and Capacity Constraint.Operations Research581666–1680

    2010

  45. [45]

    and ZEEVI, A

    SAURÉ, D. and ZEEVI, A. (2013). Optimal Dynamic Assortment Planning with Demand Learning.Manufac- turing & Service Operations Management15387–404

    2013

  46. [46]

    SCHMITT, B. A. (1992). Perturbation bounds for matrix square roots and Pythagorean sums.Linear Algebra and its Applications174215–227

    1992

  47. [47]

    B., BALAKRISHNAN, S., BRADLEY, J., PAREKH, A., RAMCHANDRAN, K

    SHAH, N. B., BALAKRISHNAN, S., BRADLEY, J., PAREKH, A., RAMCHANDRAN, K. and WAIN- WRIGHT, M. J. (2016). Estimation from Pairwise Comparisons: Sharp Minimax Bounds with Topology Dependence.Journal of Machine Learning Research171–47

    2016

  48. [48]

    SHEN, S., CHEN, X., FANG, E. X. and LU, J. (2023). Combinatorial Inference on the Optimal Assortment in the Multinomial Logit Model. InProceedings of the 24th ACM Conference on Economics and Computation.EC ’231080. Association for Computing Machinery. https://doi.org/10.1145/3580507. 3597753

    doi:10.1145/3580507 2023

  49. [49]

    TALLURI, K. T. and VANRYZIN, G. J. (2004).The Theory and Practice of Revenue Management. Springer

    2004

  50. [50]

    VAN DEGEER, S. A. (2008). High-dimensional Generalized Linear Models and the Lasso.The Annals of Statistics36614–645. https://doi.org/10.1214/009053607000000929

    doi:10.1214/009053607000000929 2008

  51. [51]

    A., BÜHLMANN, P., RITOV, Y

    VAN DEGEER, S. A., BÜHLMANN, P., RITOV, Y. and DEZEURE, R. (2014). On Asymptotically Optimal Confidence Regions and Tests for High-Dimensional Models.The Annals of Statistics421166–1202

    2014

  52. [52]

    and ZHANG, S

    ZHANG, C.-H. and ZHANG, S. S. (2014). Confidence Intervals for Low Dimensional Parameters in High Dimensional Linear Models.Journal of the Royal Statistical Society: Series B76217–242

    2014

  53. [53]

    W., JANSON, L

    ZHANG, K. W., JANSON, L. and MURPHY, S. A. (2022). Statistical Inference After Adaptive Sampling for Longitudinal Data

    2022

  54. [54]

    and LEI, J

    ZHANG, T., LEE, H. and LEI, J. (2025). Winners with Confidence: Discrete Argmin Inference with an Application to Model Selection

    2025