Pith. sign in

REVIEW 4 major objections 6 minor 47 references

Deconfounded Warm-Start Thompson Sampling with Applications to Precision Medicine

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Deconfounded warm-start Thompson sampling claims lower regret than standard LinTS by seeding priors with offline causal estimates.

desk verdict A practically motivated warm-start heuristic with an honest limitations section, but the missing naive warm-start baseline and the oracle threshold leave the deconfounding claim unproven. read the letter →

arxiv 2505.17283 v1 pith:SKXUVKPW submitted 2025-05-22 stat.ML cs.LGmath.OCstat.AP

classification stat.MLcs.LGmath.OCstat.AP
keywords Thompsonsamplingcontextualbanditswarm-starthiddenconfoundingdoublydebiasedlassoobservationaldataprecisionmedicinecumulativeregret
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to show that biased observational data, usually set aside, can be safely used to start an online learning algorithm for treatment choice. It proposes Deconfounded Warm-Start Thompson Sampling (DWTS), which first applies a debiasing procedure to the offline data to estimate which measured patient features matter and with what accuracy, then feeds those estimates into the priors of Thompson sampling while leaving hidden features uninformative. Across synthetic and NHANES-based experiments, DWTS reports consistently lower cumulative regret than standard LinTS and a state-of-the-art baseline. If true, this would let adaptive clinical trials reach good treatment policies with fewer enrolled patients.

What carries the argument

The load-bearing object is the prior initialization scheme: Doubly Debiased LASSO (DDL), a high-dimensional debiasing method that adjusts for hidden confounding, provides per-coordinate deconfounded estimates (mean and standard error) for the regression coefficients of the measured features; a threshold converts these into a mask selecting an effective dimension; LinTS then runs on the reduced dimension, with prior means and variances seeded from the offline estimates on the selected coordinates and uninformative priors on the hidden features. This transfers the causal information from the offline structural equation model into the online linear bandit, cutting both bias and dimension.

What would settle it

Construct a semi-synthetic environment from NHANES where one of the offline hidden confounders (e.g., demographic factors) is excluded from the online context, and compare DWTS against standard LinTS; if DWTS no longer shows lower cumulative regret, the claim that it harnesses confounded observational data is undercut.

Watch

Extended reading notes

Core claim

The central claim is that a contextual bandit can be warm-started from confounded observational data without inheriting its bias, provided the offline hidden confounders are measured in the online phase. The paper models the offline data with a linear structural equation where hidden confounders affect both treatment and outcome, and the online phase with a linear response model sharing the same regression coefficients. DWTS uses Doubly Debiased LASSO to estimate deconfounded coefficients and standard errors on measured covariates, selects a sparse support by thresholding, and initializes LinTS priors on those coordinates while leaving priors on the hidden features uninformative. The paper reports that this reduces regret relative to standard LinTS and to an OFUL baseline in synthetic settings and in a virtual environment built from NHANES cardiovascular risk data.

Load-bearing premise

The online phase must measure every confounder that was hidden in the offline data, because the whole warm-start relies on the online linear model being correctly specified; the authors themselves call this potentially unrealistic for real clinical settings.

Editorial extensions

If this is right

  • Adaptive clinical trials can begin with informative priors on measured features, improving early treatment decisions without enrolling additional patients.
  • The effective dimension of the online learning problem is reduced, which should lower sample complexity under sparsity.
  • The approach outperforms OFUL with partially observable confounded data in the paper's experiments, suggesting that full online measurement of confounders combined with offline debiasing is a stronger design.
  • The combined offline-online procedure gives a practical route to incorporate EHR or registry data into sequential decision-making, supporting personalized treatment.

Reading between the lines

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

  • If the online phase cannot measure all offline-hidden confounders, the same warm-start trick should fail; a natural test is to run DWTS with one confounder left unmeasured online and check whether the regret advantage disappears.
  • The threshold is a tuning knob: the paper's theoretical choice requires knowing the smallest true coefficient, so in practice one would need data-driven selection; the sensitivity of DWTS to this choice is not established.
  • The same prior-seeding idea could be transplanted to other Bayesian bandit algorithms, where DDL estimates would have to be replaced by corresponding debiased estimators.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes Deconfounded Warm-Start Thompson Sampling (DWTS), a two-stage algorithm for linear contextual bandits with access to a confounded offline dataset. In the offline stage, a Doubly Debiased LASSO (DDL) procedure estimates per-coordinate coefficients and standard errors for the measured features, and a thresholding rule selects a sparse support. In the online stage, LinTS is warm-started with these estimates on the selected measured coordinates while keeping uninformative priors on the q hidden-confounder coordinates, and the online learner operates in the reduced p_eff+q dimensional space. The authors evaluate DWTS in synthetic experiments and in a virtual environment built from NHANES cardiovascular-risk data, reporting lower cumulative regret than standard LinTS, LinTS restricted to the true support, and an OFUL method for partially observable offline data. The paper positions the contribution as a practical way to transfer causal estimates from biased observational data into an adaptive clinical trial.

Significance. If the reported gains are real and attributable to the debiasing step, DWTS would be a practically useful contribution to precision medicine and sequential decision-making under confounding. The paper is honest about its limitations: Section 6 explicitly acknowledges the assumption that all hidden confounders become observable online, the restriction to linear models, the absence of formal regret guarantees, and the lack of a treatment for non-stationary or multiple offline datasets. These admissions are a strength. However, the empirical evidence does not yet isolate the mechanism behind the regret improvement, and two experiment-design choices (the missing naive warm-start baseline and the oracle threshold used in the synthetic evaluation) weaken the central claim as it stands.

major comments (4)
  1. [§4 and Figure 1] The experimental design lacks a key ablation: DWTS is compared only against LinTS from scratch and LinTS on the true support, but never against LinTS warm-started with non-debiased offline estimates (e.g., OLS, ridge, or plain LASSO on the same offline data, with the same or a comparable selection rule). Without this baseline, the regret reduction in Figure 1 cannot be attributed to the DDL debiasing step; it may reflect the generic benefit of any offline warm-start or the reduced dimension p_eff+q. Adding this comparison is load-bearing for the paper's central claim.
  2. [Remark 3.1 and §4] In the synthetic experiments, the threshold κ_o is chosen using the true minimum nonzero coefficient β_Θ, which is infeasible in practice, as the remark itself acknowledges. This introduces oracle information into the evaluation and makes the reported regret gaps potentially optimistic. The authors should report results with a feasible threshold selection procedure (e.g., cross-validation, a quantile of |θ̂_a,i|, or a sensitivity analysis over κ_o) to support the practical claim that the method can be deployed without knowledge of the true parameters.
  3. [§3 and Algorithm 1, step 7] The prior precision is initialized as B̂_{0,a}^{-1} = diag([σ̂_a^{eff}, 1_q]), but σ̂_a,i is defined in step 3 as the standard error of θ̂_a,i. Precision should be the inverse variance, namely diag([1/(σ̂_a^{eff})^2, 1_q]), if σ̂ denotes a standard error. As written, the algorithm does not implement the claimed warm-start with DDL-estimated variances, and the pseudocode may not be reproducible. This needs to be corrected and the resulting experiments re-run or justified.
  4. [§5.2] The virtual NHANES environment generates rewards as r_i(a) = ρ_a · CVDRisk_i + ε, where CVDRisk_i is the output of nonlinear cardiovascular risk calculators. This reward model does not satisfy the linear response model in Eq. (2) used by all LinTS variants and by DDL. The real-data evaluation therefore tests the algorithms under misspecification without discussing this mismatch. To support the claim that DWTS works under the paper's linear-model assumptions, the authors should either construct a linearized version of the virtual environment, or explicitly analyze and justify the effect of this misspecification on the comparison.
minor comments (6)
  1. [Algorithm 1, line 15] The score is computed as s_{t,a} = μ_{t,a}^⊺ X_t^h, but μ_{t,a} lies in R^{p_eff+q} while X_t^h lies in R^{p+q}; the context must be restricted to the effective and hidden coordinates, e.g., X_t^{h,eff}, or the dimensions should be specified consistently.
  2. [Figures 1 and 2] The second baseline is labeled 'LinTS (Important dimensions)' in the captions but is not defined in the main text; the paper should state explicitly whether this baseline is warm-started or initialized from scratch, and whether it uses the true support of θ*.
  3. [§4] The text says 'In LinTS, we learn all=p+q dimensions from scratch' but does not describe the arm-specific structure of θ*_a (e.g., θ*_a = [(a+1)1_5; 0_5] is only given for one arm); the construction for both arms should be stated completely.
  4. [§5.2] The noise level σ² for the virtual environment is not specified; report the value used in the experiments so the results are reproducible.
  5. [Throughout] There are several typos and grammatical issues (e.g., 'we chose to do not overburden the notations', inconsistent use of Y_t^h(a) vs. Y_t^h); a careful proofread would improve clarity.
  6. [Reproducibility] No code or data availability statement is provided; for an empirical paper, releasing the code and the data-processing pipeline would substantially strengthen the contribution.

Circularity Check

1 steps flagged · score 4.0 of 10

Synthetic validation uses an oracle threshold κ_o built from the true minimum nonzero coefficient, so the reported warm-start advantage is not fully independent of ground truth.

  1. fitted input called prediction [Remark 3.1 and Section 4 (Synthetic Experiments)]
    "we can choose κ o =min{β Θ−max a,i{z1−α/2 ˆσa,i},0} where β Θ =min a,i{∣θa,i∣∶θ a,i ≠0}. However, note that the theoretical approach require the knowledge of the minimum of the true important coefficients, which is infeasible in real world experimental settings. ... Since, this a synthetic experiment, we compute κ o using the strategy describe in Remark 3.1 for α=0.05."

    In Algorithm 1, the feature mask is m̂_a = {I(|θ̂_{a,i}| ≥ κ_o)}. In the synthetic experiments κ_o is chosen using β_Θ, the true minimum nonzero coefficient, so the mask — and therefore the reduced context, the warm-started means, and the online dimension p_eff+q — is constructed with ground-truth knowledge about the sparse support and coefficient magnitudes. The regret curves in Figure 1 are then read as 'validating the benefits of effective dimension reduction and deconfounded warm-start.' This is a fitted input derived from the true parameters being estimated, so the empirical claim is partly forced by oracle information rather than solely by DDL estimates from the offline data.

full rationale

The core DWTS construction is not circular: it composes an external DDL estimator (Guo et al., 2022), a thresholded mask, and a standard LinTS warm-start, and no equation defines the DWTS output in terms of the regret improvement it claims. The only coauthor self-citation (Keyvanshokooh et al. 2025) appears in a motivating example and is not load-bearing. DDL is cited as an off-the-shelf method, and its assumptions are stated rather than smuggled in via a self-citation chain. The main circularity concern is confined to the synthetic evaluation: Remark 3.1 defines κ_o using the true minimum nonzero coefficient β_Θ, and Section 4 admits this oracle strategy is used in the experiments. That makes the feature-selection step and hence the DWTS regret advantage in Figure 1 partly dependent on ground truth. However, the paper also evaluates DWTS on the NHANES-derived environment with a feasible threshold κ_o=0.01, and it compares against LinTS on the true important dimensions; these provide independent evidence that warm-starting and dimension reduction help. The absence of a non-debiased warm-started LinTS baseline is an experimental isolation gap, not itself a circular step. On balance, the central method has independent content, but the synthetic validation contains a partially oracle-driven prediction, giving an overall score of 4.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No new entities are introduced; the 'reduced context' is a subset of existing features and hidden confounders are standard latent variables. The free parameters and domain assumptions are the main additional burden the central claim rests on.

free parameters (3)
  • κ_o (feature selection threshold) = synthetic: oracle-derived; NHANES: 0.01
    Threshold for masking covariates as important. In synthetic experiments, it is computed using the true β_Θ (Remark 3.1), which requires knowing the minimum nonzero coefficient; in the real-data experiment it is arbitrarily set to 0.01 without sensitivity analysis.
  • Prior variance initialization = σ̂ (standard error used as variance)
    The prior covariance is set to diag(σ̂_eff, 1_q) in Algorithm 1. If σ̂ is a standard error, the variance should be σ̂^2; using σ̂ directly is an unstated modeling choice that can affect exploration.
  • DDL hyperparameters = unspecified
    The DDL procedure has its own tuning parameters (e.g., regularization strength) that are not reported, yet the selected support and standard errors depend on them.
assumptions (5)
  • domain assumption Observational data follows linear SEM (1) with hidden confounders H.
    Section 2, Eq. (1). This is the model DDL is designed for; if the true data is nonlinear or confounders are not linearly related to Z and Y, DDL estimates may be biased.
  • domain assumption Online outcomes follow linear model (2) sharing the same θ* and ϕ* with the observational SEM, and all hidden confounders are observed online.
    Section 2, Eq. (2) and the text around it. This is necessary for the warm-start to transfer; the authors list the observability part as a limitation.
  • domain assumption True θ* is sparse so that feature selection via DDL identifies the support with high probability.
    Invoked in Section 3 and the sparsity discussion; DDL's guarantees require sparsity and other regularity conditions from Guo et al. 2022.
  • standard math Noise terms are independent Gaussian with unit variance.
    Assumed in Eqs. (1) and (2) to justify DDL and the Bayesian update; not verified.
  • domain assumption DDL confidence intervals (Eq. 13 in Guo et al.) are valid under the observed sample sizes.
    Used in Remark 3.1 to propose a theoretical κ_o; requires high-dimensional inference conditions that are not checked.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deconfounded Warm-Start Thompson Sampling with Applications to Precision Medicine." pith.science (2026). https://pith.science/paper/SKXUVKPW

@misc{pith2026250517283,
  author       = {Pith},
  title        = {Pith review of: Deconfounded Warm-Start Thompson Sampling with Applications to Precision Medicine},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SKXUVKPW}},
  note         = {Machine review of arXiv:2505.17283}
}
read the original abstract

Randomized clinical trials often require large patient cohorts before drawing definitive conclusions, yet abundant observational data from parallel studies remains underutilized due to confounding and hidden biases. To bridge this gap, we propose Deconfounded Warm-Start Thompson Sampling (DWTS), a practical approach that leverages a Doubly Debiased LASSO (DDL) procedure to identify a sparse set of reliable measured covariates and combines them with key hidden covariates to form a reduced context. By initializing Thompson Sampling (LinTS) priors with DDL-estimated means and variances on these measured features -- while keeping uninformative priors on hidden features -- DWTS effectively harnesses confounded observational data to kick-start adaptive clinical trials. Evaluated on both a purely synthetic environment and a virtual environment created using real cardiovascular risk dataset, DWTS consistently achieves lower cumulative regret than standard LinTS, showing how offline causal insights from observational data can improve trial efficiency and support more personalized treatment decisions.

Figures

Figures reproduced from arXiv: 2505.17283 by the authors.

Figure 1
Figure 1. Median cumulative regret (solid) with 10%–90% quantile bands (shaded) over [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Median cumulative regret (solid) with 10%–90% quantile bands (shaded) over [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 33 canonical work pages

  1. [1]

    Agrawal and N

    S. Agrawal and N. Goyal. Thompson sampling for contextual bandits with linear payoffs. In International conference on machine learning, pages 127--135. PMLR, 2013

  2. [2]

    Banerjee, S

    S. Banerjee, S. R. Sinclair, M. Tambe, L. Xu, and C. L. Yu. Artificial replay: a meta-algorithm for harnessing historical data in bandits. arXiv preprint arXiv:2210.00025, 2022

  3. [3]

    S. Basu, J. B. Sussman, S. A. Berkowitz, R. A. Hayward, and J. S. Yudkin. Development and validation of risk equations for complications of type 2 diabetes (recode) using individual participant data from randomised trials. The lancet Diabetes & endocrinology, 5 0 (10): 0 788--798, 2017

  4. [4]

    Belloni, V

    A. Belloni, V. Chernozhukov, and C. Hansen. Inference for high-dimensional sparse econometric models. In Advances in economics and econometrics: Tenth world congress, volume 3, pages 245--95. Cambridge University Press Cambridge, 2013

  5. [5]

    M. A. Brookhart, T. St \"u rmer, R. J. Glynn, J. Rassen, and S. Schneeweiss. Confounding control in healthcare database research: challenges and potential approaches. Medical care, 48 0 (6): 0 S114--S120, 2010

  6. [6]

    Buehlmann

    P. Buehlmann. Boosting for high-dimensional linear models. 2006

  7. [7]

    Cheng and T

    D. Cheng and T. Cai. Adaptive combination of randomized and observational data. arXiv preprint arXiv:2111.15012, 2021

  8. [8]

    Curth and M

    A. Curth and M. Van der Schaar. Nonparametric estimation of heterogeneous treatment effects: From theory to learning algorithms. In International Conference on Artificial Intelligence and Statistics, pages 1810--1818. PMLR, 2021

Show all 47 references
  1. [9]

    I. J. Dahabreh and K. Bibbins-Domingo. Causal inference about the effects of interventions from observational studies in medical journals. Jama, 331 0 (21): 0 1845--1853, 2024

  2. [10]

    B. T. Denton. Optimization of sequential decision making for chronic diseases: From data to decisions. In Recent Advances in Optimization and Modeling of Contemporary Problems, pages 316--348. INFORMS, 2018

  3. [11]

    Efron, T

    B. Efron, T. Hastie, I. Johnstone, and R. Tibshirani. Least angle regression. 2004

  4. [12]

    D. C. Goff, D. M. Lloyd-Jones, G. Bennett, S. Coady, R. B. D’agostino, R. Gibbons, P. Greenland, D. T. Lackland, D. Levy, C. J. O’donnell, et al. 2013 acc/aha guideline on the assessment of cardiovascular risk: a report of the american college of cardiology/american heart asso...

  5. [13]

    Z. Guo, D. \'C evid, and P. B \"u hlmann. Doubly debiased lasso: High-dimensional inference under hidden confounding. Annals of statistics, 50 0 (3): 0 1320, 2022

  6. [14]

    B. Hao, R. Jain, T. Lattimore, B. Van Roy, and Z. Wen. Leveraging demonstrations to improve online learning: Quality matters. In International Conference on Machine Learning, pages 12527--12545. PMLR, 2023

  7. [15]

    Hatt and S

    T. Hatt and S. Feuerriegel. Estimating average treatment effects via orthogonal regularization. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pages 680--689, 2021

  8. [16]

    T. Hatt, J. Berrevoets, A. Curth, S. Feuerriegel, and M. van der Schaar. Combining observational and randomized data for estimating heterogeneous treatment effects. arXiv preprint arXiv:2202.12891, 2022

  9. [17]

    L. Hu, J. Ji, and F. Li. Estimating heterogeneous survival treatment effect in observational data using machine learning. Statistics in medicine, 40 0 (21): 0 4691--4713, 2021

  10. [18]

    M. Ilse, P. Forr \'e , M. Welling, and J. M. Mooij. Combining interventional and observational data using causal reductions. arXiv preprint arXiv:2103.04786, 2021

  11. [19]

    Kallus, A

    N. Kallus, A. M. Puli, and U. Shalit. Removing hidden confounding by experimental grounding. Advances in neural information processing systems, 31, 2018

  12. [20]

    Keyvanshokooh, M

    E. Keyvanshokooh, M. Zhalechian, C. Shi, M. P. Van Oyen, and P. Kazemian. Contextual learning with online convex optimization: Theory and application to medical decision-making. Management Science, 2025

  13. [21]

    W. Kim, S. Park, G. Iyengar, A. Zeevi, and M.-h. Oh. Linear bandits with partially observable features. arXiv preprint arXiv:2502.06142, 2025

  14. [22]

    Y. Kim, J. Hao, T. Mallavarapu, J. Park, and M. Kang. Hi-lasso: High-dimensional lasso. IEEE Access, 7: 0 44562--44573, 2019

  15. [23]

    S. R. K \"u nzel, J. S. Sekhon, P. J. Bickel, and B. Yu. Metalearners for estimating heterogeneous treatment effects using machine learning. Proceedings of the national academy of sciences, 116 0 (10): 0 4156--4165, 2019

  16. [24]

    Kuzmanovic, T

    M. Kuzmanovic, T. Hatt, and S. Feuerriegel. Deconfounding temporal autoencoder: estimating treatment effects over time using noisy proxies. In Machine Learning for Health, pages 143--155. PMLR, 2021

  17. [25]

    Lattimore and C

    T. Lattimore and C. Szepesv \'a ri. Bandit algorithms. Cambridge University Press, 2020

  18. [26]

    Louizos, U

    C. Louizos, U. Shalit, J. M. Mooij, D. Sontag, R. Zemel, and M. Welling. Causal effect inference with deep latent-variable models. Advances in neural information processing systems, 30, 2017

  19. [27]

    Luo and Z

    S. Luo and Z. Chen. Sequential lasso cum ebic for feature selection with ultra-high dimensional feature space. Journal of the American Statistical Association, 109 0 (507): 0 1229--1240, 2014

  20. [28]

    Mei and Z

    Z. Mei and Z. Shi. On lasso for high dimensional predictive regression. Journal of Econometrics, 242 0 (2): 0 105809, 2024

  21. [29]

    N rgaard, V

    M. N rgaard, V. Ehrenstein, and J. P. Vandenbroucke. Confounding in observational studies based on large health care databases: problems and potential solutions--a primer for the clinician. Clinical epidemiology, pages 185--193, 2017

  22. [30]

    J. Pearl. Causality. Cambridge university press, 2009

  23. [31]

    Powell, C

    M. Powell, C. Clark, A. Alyakin, J. T. Vogelstein, and B. Hart. Exploration of residual confounding in analyses of associations of metformin use and outcomes in adults with type 2 diabetes. JAMA Network Open, 5 0 (11): 0 e2241505--e2241505, 2022

  24. [32]

    E. T. Rosenman, G. Basse, A. B. Owen, and M. Baiocchi. Combining observational and experimental datasets using shrinkage estimators. Biometrics, 79 0 (4): 0 2961--2973, 2023

  25. [33]

    Schneeweiss, W

    S. Schneeweiss, W. Eddings, R. J. Glynn, E. Patorno, J. Rassen, and J. M. Franklin. Variable selection for confounding adjustment in high-dimensional covariate spaces when analyzing healthcare databases. Epidemiology, 28 0 (2): 0 237--248, 2017

  26. [34]

    Sharma, S

    N. Sharma, S. Basu, K. Shanmugam, and S. Shakkottai. On under-exploration in bandits with mean bounds from confounded data. arXiv preprint arXiv:2002.08405, 2020

  27. [35]

    C. Shi, D. Blei, and V. Veitch. Adapting neural networks for the estimation of treatment effects. Advances in neural information processing systems, 32, 2019

  28. [36]

    E. A. Stuart, S. R. Cole, C. P. Bradshaw, and P. J. Leaf. The use of propensity scores to assess the generalizability of results from randomized trials. Journal of the Royal Statistical Society Series A: Statistics in Society, 174 0 (2): 0 369--386, 2011

  29. [37]

    T. A. Stukel, E. S. Fisher, D. E. Wennberg, D. A. Alter, D. J. Gottlieb, and M. J. Vermeulen. Analysis of observational studies in the presence of treatment selection bias: effects of invasive cardiac management on ami survival using propensity score and instrumental variable ...

  30. [38]

    D. Tang, R. Jain, B. Hao, and Z. Wen. Efficient online learning with offline datasets for infinite horizon mdps: A bayesian approach. arXiv preprint arXiv:2310.11531, 2023

  31. [39]

    Tennenholtz, U

    G. Tennenholtz, U. Shalit, S. Mannor, and Y. Efroni. Bandits with partially observable confounded data. In Uncertainty in Artificial Intelligence, pages 430--439. PMLR, 2021

  32. [40]

    Tibshirani

    R. Tibshirani. Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society Series B: Statistical Methodology, 58 0 (1): 0 267--288, 1996

  33. [41]

    Wagenmaker and A

    A. Wagenmaker and A. Pacchiano. Leveraging offline data in online reinforcement learning. In International Conference on Machine Learning, pages 35300--35338. PMLR, 2023

  34. [42]

    Wager and S

    S. Wager and S. Athey. Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 113 0 (523): 0 1228--1242, 2018

  35. [43]

    Wang and D

    Y. Wang and D. M. Blei. The blessings of multiple causes. Journal of the American Statistical Association, 114 0 (528): 0 1574--1596, 2019

  36. [44]

    Yamada, W

    M. Yamada, W. Jitkrittum, L. Sigal, E. P. Xing, and M. Sugiyama. High-dimensional feature selection by feature-wise kernelized lasso. Neural computation, 26 0 (1): 0 185--207, 2014

  37. [45]

    Yang and P

    S. Yang and P. Ding. Combining multiple observational data sources to estimate causal effects. Journal of the American Statistical Association, 2019

  38. [46]

    L. Yao, S. Li, Y. Li, M. Huai, J. Gao, and A. Zhang. Representation learning for treatment effect estimation from observational data. Advances in neural information processing systems, 31, 2018

  39. [47]

    Zhang, A

    C. Zhang, A. Agarwal, H. Daum \'e III, J. Langford, and S. N. Negahban. Warm-starting contextual bandits: Robustly combining supervised and bandit feedback. arXiv preprint arXiv:1901.00301, 2019

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.