Pith. sign in

REVIEW 2 major objections 3 minor 1 cited by

Towards the Efficient Inference by Incorporating Automated Computational Phenotypes under Covariate Shift

T0 review · 2 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper shows that automated computational phenotypes can be incorporated into semi-supervised inference under covariate shift to yield doubly robust, semiparametrically efficient estimators, and that the efficiency gains come entirely…

desk verdict Solid PPI extension under covariate shift with a correctable overstatement about when ACPs yield zero efficiency gain. read the letter →

arxiv 2505.22632 v1 pith:47JT4GBT submitted 2025-05-28 stat.ME

Chao Ying , Jun Jin , Yi Guo , Xiudi Li , Muxuan Liang , Jiwei Zhao This is my paper
classification stat.ME MSC 62G0562G20
keywords automatedcomputationalphenotypessemi-supervisedlearningcovariateshiftsemiparametricefficiencybounddoublyrobustestimationefficientinfluencefunctioncross-fittingprediction-poweredinference
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

This paper addresses a practical bottleneck: gold-standard labels are expensive, while automated computational phenotypes (ACPs) are cheap and available for an entire cohort. It claims that ACPs can be added to semi-supervised estimation under covariate shift without ever reducing asymptotic efficiency, and with strictly positive gains whenever the ACP carries information beyond the recorded features. The proposed cross-fitted estimator is doubly robust and attains the semiparametric efficiency bound, so no regular asymptotically linear estimator can do better in large samples. The paper's closed-form variance comparison attributes the entire gain to ACPs on unlabeled observations; ACPs on labeled observations contribute nothing. If the outcome model transports from labeled to unlabeled data, the practical payoff is shorter confidence intervals and more powerful association tests at no extra labeling cost.

What carries the argument

The argument rides on the semiparametric tangent-space decomposition for the combined labeled/unlabeled likelihood, which yields the efficient influence function. The load-bearing objects are the density ratio $w_0(x)=q(x)/p(x)$, the conditional score $m_0(x)=E\{s(Y,X;\beta_0)\mid x\}$, and its ACP-augmented counterpart $\tilde m_0(x,\tilde y)=E\{s(Y,X;\beta_0)\mid x,\tilde y\}$; the EIF (4) reweights these by the selection indicator $R$. Cross-fitting with $K$ folds and product-rate conditions $a_{1M}a_{2M}=o(M^{-1/2})$, $a_{1M}a_{3M}=o(M^{-1/2})$ lets the estimator tolerate nuisance estimates converging slower than $M^{-1/4}$ and avoids Donsker restrictions. The efficiency-gain formula comes from subtracting the two closed-form variance expressions, which isolates the term depending only on unlabeled ACPs.

What would settle it

Run the paper's synthetic experiment with the same ACP generation but set the unlabeled outcome to $q(y\mid x)\neq p(y\mid x)$, for example by shifting the intercept by a nonzero constant; if the proposed estimator's bias does not vanish as $n,N\to\infty$ and its nominal 95% confidence intervals undercover, that confirms the transportability assumption is load-bearing. A direct check in an applied dataset is to label a random hold-out subset of the unlabeled cohort and test equality of $p(y\mid x)$ and $q(y\mid x)$; rejection would invalidate the identification of $\beta$.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the efficiency bound and the efficient influence function for a parameter $\beta$ defined by $E_q\{s(Y,X;\beta)\}=0$ take an explicit form under covariate shift, and that the bound with ACPs is strictly smaller than the bound without ACPs except in the degenerate case where $\tilde Y$ is a function of $X$. Proposition 3.1 gives the EIF $\Omega\phi_w$ with $\phi_w$ as in equation (4); Proposition 3.3 shows the variance gap equals $\frac{1}{(1-\pi)^2}\Omega E[\{1-\pi_0(X)\}^3/\pi_0(X)\{\tilde m_0(X,\tilde Y)-m_0(X)\}^{\otimes2}]\Omega$, which is positive semidefinite. Theorem 5.5 proves that the cross-fitted estimator solving equation (5) is asymptotically normal with covariance $\Omega V_w\Omega$, exactly the semiparametric efficiency bound, under the product-rate conditions. Theorem 5.3 establishes double robustness, and Proposition 3.5 shows that ACPs available only for labeled data leave the bound unchanged, so the gain is attributable to unlabeled ACPs.

Load-bearing premise

The load-bearing premise is that the outcome distribution transports from the labeled to the unlabeled population, $p(y\mid x)=q(y\mid x)$ (and the same conditional stability holds for the ACP); because $Y$ is unobserved in the unlabeled data, this equality cannot be tested in the semi-supervised setting the paper targets.

Editorial extensions

If this is right

  • ACP-augmented inference never increases asymptotic variance; it strictly reduces it whenever $\tilde Y$ is not a function of $X$ and is predictive given $X$.
  • The entire efficiency gain is driven by ACPs attached to unlabeled observations, so ACP collection effort should be directed to the unlabeled cohort.
  • The estimator remains consistent if either the density-ratio model or the pair of outcome regressions is correctly specified (double robustness).
  • Under the product-rate conditions, nuisance functions can be estimated by flexible machine learning without Donsker-type entropy restrictions and the estimator still reaches the semiparametric bound.
  • Confidence intervals based on equation (8) have nominal asymptotic coverage and their lengths are asymptotically no larger than intervals from any estimator that ignores ACPs.

Reading between the lines

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

  • A design corollary the paper leaves implicit: an ACP algorithm should be encouraged to consume auxiliary variables beyond the model's feature set; an ACP built only from $X$ is informationally useless for this estimator.
  • The theorem also suggests a resource-allocation rule for EHR studies: spend ACP generation on the unlabeled majority rather than on chart-reviewed patients, since labeled ACPs contribute no asymptotic gain.
  • Because the efficiency bound is conditional on transportability, a natural extension is a sensitivity analysis that reports how much bias in $\hat\beta$ could arise from a given divergence between $p(y\mid x)$ and $q(y\mid x)$; the paper does not provide such a bound.
  • The product-rate double robustness might be exploited in practice by pairing a fast, parametric density-ratio estimator with flexible machine-learned outcome regressions, relaxing the requirement that every nuisance converge quickly.
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

2 major / 3 minor

Summary. The paper studies semi-supervised estimation of parameters defined by estimating equations for the unlabeled population under covariate shift, when automated computational phenotypes (ACPs) are available for both labeled and unlabeled observations. It derives the semiparametric efficient influence function with and without ACPs (Propositions 3.1 and 3.2), gives a closed-form positive-semidefinite efficiency gain from incorporating ACPs (Proposition 3.3), and argues that only unlabeled ACPs, not labeled ACPs, produce the gain (Proposition 3.5). It then proposes a cross-fitted doubly robust estimator for the target parameter, establishes double robustness (Theorem 5.3), asymptotic normality at the semiparametric efficiency bound (Theorem 5.5), and valid confidence intervals (Theorem 5.8), with supporting simulations and real-data analyses.

Significance. The core derivation is self-contained and largely standard semiparametric theory: the EIF is obtained from a tangent-space decomposition and the efficiency comparison is given in closed form. If the results hold, the paper makes a useful contribution to the prediction-powered inference and surrogate-outcome literature by allowing flexible covariate shift, black-box ACPs, and doubly robust efficient estimation. The explicit formula for the efficiency gain and the identification of unlabeled ACPs as the source of the gain are potentially valuable for practitioners. The paper also ships reproducible code and compares with PPI/PPI++/RePPI on several data sets. The main mathematical claims appear sound, but a key interpretive claim about when the gain vanishes is overstated and needs correction, and one headline proposition is stated without proof.

major comments (2)
  1. [§1.2 and Remark 3.4] The claim that 'the only scenario with no efficiency gain occurs when the generation of ACPs depends solely on the available feature X' is not supported by the paper's own Proposition 3.3. The gain matrix is proportional to E[ (1−π0(X))^3/π0(X) {em0(X,bY)−m0(X)}^⊗2 ], so the gain vanishes if and only if em0(X,bY)=m0(X) almost surely, i.e., E[s(Y,X;β0)|X,bY]=E[s(Y,X;β0)|X]. This condition holds whenever bY is conditionally independent of Y given X, including the case where bY=f(X,Z) with Z independent of Y given X, or bY is pure noise independent of (X,Y). These cases are not covered by the 'bY is a function of X' characterization. The paper's own simulation discussion in §6.1 implicitly acknowledges this: ARE equals 1 when the ACP is not predictive of the outcome (α=0), even though bY is generated from an additional covariate Z. The instructions in Remark 3.4 should be revised to state the correct zero-gain condition, and the corresponding sentences in §1.2 should be corrected so that practitioners are not told that any ACP depending on variables beyond X automatically improves efficiency.
  2. [Proposition 3.5 and Remark 3.6] The claim that ACPs in the labeled data contribute no efficiency gain is a headline result of the paper, but Proposition 3.5 is stated without proof, with only 'proof similar to Proposition 3.1 so omitted.' Because this proposition underlies the abstract's assertion that unlabeled ACPs drive the efficiency gain, the appendix should contain the derivation, or at least a detailed proof sketch that specifies the tangent spaces for the intermediate scenario in Table 2 and verifies that the resulting efficiency bound equals that of Proposition 3.2. Without this, a load-bearing claim is left unverified.
minor comments (3)
  1. [Proposition 3.3 and Appendix A] The unsubscripted expectation E in the gain formula is not explicitly defined. In the proof of Proposition 3.3 the notation alternates among E, E_p, and E_q, and the final coefficient (1−π0(X))^3/π0(X) is only correct when E is interpreted as the expectation over the combined covariate distribution with density proportional to π p(x)+(1−π)q(x). Please define this expectation explicitly in the main text.
  2. [Throughout the paper] There are several typographical errors that should be fixed: 'combind' in Appendix B, 'propoesd' in Table 14, and 'identify' where 'identity' is meant in §6.1. These are minor but interrupt reading.
  3. [Section 3.1] The RAL estimator definition is written for a sample of size n, while the data consist of M=n+N observations. The notation should be harmonized so that the total sample size M is used consistently before Remark 5.7 introduces the π→0 scaling.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the efficiency-bound derivation is self-contained; the 'only scenario' wording in Section 1.2 overreaches its own formula but that is a correctness issue, not a circular one.

full rationale

The derivation chain is self-contained. Proposition 3.1 obtains the efficient influence function from the factored likelihood p(y|x,bY)p(bY|x)p(x)^r q(x)^{1-r} and orthogonal tangent-space calculations, and the efficiency bound is computed in closed form. Proposition 3.3 is an algebraic subtraction of V_wo and V_w, and Theorem 5.5 proves that the cross-fitted estimator solving equation (5) is asymptotically normal with variance equal to that bound through the standard sample-splitting, bias, and CLT decomposition under product rate conditions. No fitted parameter is relabeled as a prediction: the target beta is defined by Eq{s(Y,X;beta)}=0, the nuisance functions are estimated but must converge to their true limits, and the reported efficiency gain is a function of the true nuisance functions. Self-citations (e.g., Miao et al. 2023/2024; Deng et al. 2024; Tian et al. 2023; Kim et al. 2024; Lee et al. 2025) appear only as related work and do not supply a load-bearing premise; the semiparametric machinery is cited to Bickel et al. (1993), Tsiatis (2006), and Chernozhukov et al. (2018). I do flag two non-circular concerns. First, Section 1.2 states "The only scenario with no efficiency gain occurs when the generation of ACPs depends solely on the available feature X," but Proposition 3.3's own zero-gain condition is em0(X,bY)=m0(X), which also holds for ACPs that are independent of Y given X even if generated using additional variables; this is an internal-consistency and correctness issue, not circularity. Second, Proposition 3.5's proof is "similar to Proposition 3.1 so omitted," and Remark 2.1 candidly labels p(y|x,bY)=q(y|x,bY) untestable; both are completeness and identification limitations rather than circular reductions. Therefore the circularity score is 0.

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

The paper introduces no new physical or conceptual entities. All assumptions are standard domain assumptions in semi-supervised learning and missing-data theory. The nuisance functions are estimated rather than hand-set. The central burden is transportability of p(y|x) and ignorability of labeling, both untestable in the target application.

free parameters (3)
  • pi(x) propensity model = estimated from data via SuperLearner
    The labeling probability pi(x) is a nuisance function estimated from data; the estimator's properties depend on its consistent estimation, but it is not a free parameter introduced to fit the target result.
  • w(x) density ratio = estimated from data via SuperLearner
    The density ratio w(x) = q(x)/p(x) is a nuisance function estimated from data. In practice it is fit, but the theory treats it as a true unknown and the estimator is doubly robust, so it is not hand-tuned.
  • m(x) and em(x,by) regression functions = estimated from data via SuperLearner
    The outcome regression functions are nuisance parameters estimated from data. They are not chosen to match the efficiency gain; they are standard nuisance estimates.
assumptions (5)
  • domain assumption p(y | x) = q(y | x), the conditional outcome distribution is transportable across labeled and unlabeled populations.
    Assumed in Section 2 and used throughout; the authors themselves note in Remark 2.1 that p(y|x,by) = q(y|x,by) is untestable when Y is missing in unlabeled data. This is the load-bearing identification assumption.
  • domain assumption pr(R=1 | Y, X) = pr(R=1 | X), i.e., labeling is ignorable given X.
    Equation (1), the covariate shift assumption. It implies equation (3) with ACPs. This is not testable in the semi-supervised setting.
  • standard math Nuisance estimators converge at product rates a1M*a2M = o(M^{-1/2}) and a1M*a3M = o(M^{-1/2}).
    Assumption 5.1 and the rate conditions in Theorem 5.5 are standard in double/debiased ML. They are regularity conditions, not free parameters.
  • standard math E_q{ds(Y,X;beta)/dbeta^T} is invertible at beta0.
    Assumed implicitly in the definition of Omega in Section 2, needed for the influence function representation.
  • domain assumption The ACP by generation does not depend on R given (X,Y), i.e., p(by|x,y) = q(by|x,y).
    Assumed in Section 2 before equation (3). This is testable only if Y is available in both populations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards the Efficient Inference by Incorporating Automated Computational Phenotypes under Covariate Shift." pith.science (2026). https://pith.science/paper/47JT4GBT

@misc{pith2026250522632,
  author       = {Pith},
  title        = {Pith review of: Towards the Efficient Inference by Incorporating Automated Computational Phenotypes under Covariate Shift},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/47JT4GBT}},
  note         = {Machine review of arXiv:2505.22632}
}
read the original abstract

Collecting gold-standard phenotype data via manual extraction is typically labor-intensive and slow, whereas automated computational phenotypes (ACPs) offer a systematic and much faster alternative. However, simply replacing the gold-standard with ACPs, without acknowledging their differences, could lead to biased results and misleading conclusions. Motivated by the complexity of incorporating ACPs while maintaining the validity of downstream analyses, in this paper, we consider a semi-supervised learning setting that consists of both labeled data (with gold-standard) and unlabeled data (without gold-standard), under the covariate shift framework. We develop doubly robust and semiparametrically efficient estimators that leverage ACPs for general target parameters in the unlabeled and combined populations. In addition, we carefully analyze the efficiency gains achieved by incorporating ACPs, comparing scenarios with and without their inclusion. Notably, we identify that ACPs for the unlabeled data, instead of for the labeled data, drive the enhanced efficiency gains. To validate our theoretical findings, we conduct comprehensive synthetic experiments and apply our method to multiple real-world datasets, confirming the practical advantages of our approach. \hfill{\texttt{Code}: \href{https://github.com/brucejunjin/ICML2025-ACPCS}{\faGithub}}

Figures

Figures reproduced from arXiv: 2505.22632 by the authors.

Figure 1
Figure 1. Variation of ARE for the estimation of Eq(Y ) under different sample sizes, signal strength, and correlation coefficient. α = 0 α = 1 α = 2 α = 3 α = 4 α = 5 α = 0 α = 1 α = 2 α = 3 α = 4 α = 5 n = 300, N changes N = 300, n changes 300 600 900 1200 1500 300 600 900 1200 1500 5 10 15 1 Sample size ARE α α = 0 α = 1 α = 2 α = 3 α = 4 α = 5 ζ = 0 ζ = 0.3 ζ = 0.6 ζ = 0.9 ζ = 1 ζ = 0 ζ = 0.3 ζ = 0.6 ζ = 0.9 ζ = 1 n = 300… view at source ↗
Figure 2
Figure 2. Difference in distributions of the inpatient visit count for labeled (left) and unlabeled (right) dataset. 0 50 100 150 0 1 2 3 4 5 Inpatient Visit Count (Labeled) Frequency 0 500 1000 1500 2000 2500 0 1 2 3 4 5 Inpatient Visit Count (Unlabeled) Frequency (α), and the correlation between ACP and the covariates (ζ). Efficiency gain is quantified by Asymptotic Relative Efficiency (ARE), defined as the ratio of the MSE… view at source ↗
Figure 4
Figure 4. Variation of ARE for the estimation of ξ1 under linear model setting with different sample sizes, signal strength, and correlation coefficient. 20 [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Variation of ARE for the estimation of ξ2 under linear model setting with different sample sizes, signal strength, and correlation coefficient. 21 [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]
Figure 6
Figure 6. Figure 6: Variation of ARE for the estimation of Eq(Y ) under logistic model setting with different sample sizes, signal strength, and correlation coefficient. 22 [PITH_FULL_IMAGE:figures/full_fig_p023_6.png]
Figure 7
Figure 7. Figure 7: Variation of ARE for the estimation of ξ1 under logistic model setting with different sample sizes, signal strength, and correlation coefficient. 23 [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Variation of ARE for the estimation of ξ2 under logistic model setting with different sample sizes, signal strength, and correlation coefficient. 24 [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Towards Efficient Inference under Nonmonotone Missingness with General Imputation

    stat.ME 2025-09 conditional novelty 6.0 of 10

    The RAY decomposition approximates the semiparametrically efficient estimator under blockwise missingness, yielding AI-powered estimators that are unbiased, asymptotically normal, and adaptively no worse than complete...

Reference graph

Works this paper leans on

86 extracted references · 66 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    M., Toni, L., and Rodrigues, M

    Aminian, G., Abroshan, M., Khalili, M. M., Toni, L., and Rodrigues, M. An information-theoretical approach to semi-supervised learning under covariate-shift. In International Conference on Artificial Intelligence and Statistics, pp.\ 7433--7449. PMLR, 2022

    2022

  3. [3]

    N., Bates, S., Fannjiang, C., Jordan, M

    Angelopoulos, A. N., Bates, S., Fannjiang, C., Jordan, M. I., and Zrnic, T. Prediction-powered inference. Science, 382 0 (6671): 0 669--674, 2023 a

    2023

  4. [4]

    N., Duchi, J

    Angelopoulos, A. N., Duchi, J. C., and Zrnic, T. Ppi++: Efficient prediction-powered inference. arXiv preprint arXiv:2311.01453, 2023 b

    arXiv 2023

  5. [5]

    W., and Kang, H

    Athey, S., Chetty, R., Imbens, G. W., and Kang, H. The surrogate index: Combining short-term proxies to estimate long-term treatment effects more rapidly and precisely. Technical report, National Bureau of Economic Research, 2019

    2019

  6. [6]

    Combining experimental and observational data to estimate treatment effects on long term outcomes

    Athey, S., Chetty, R., and Imbens, G. Combining experimental and observational data to estimate treatment effects on long term outcomes. arXiv preprint arXiv:2006.09676, 2020

    arXiv 2006

  7. [7]

    D., Sklar, M., Berk, R., Buja, A., and Zhao, L

    Azriel, D., Brown, L. D., Sklar, M., Berk, R., Buja, A., and Zhao, L. Semi-supervised linear regression. Journal of the American Statistical Association, 117 0 (540): 0 2238--2251, 2022

    2022

  8. [8]

    Assumption lean regression

    Berk, R., Buja, A., Brown, L., George, E., Kuchibhotla, A., Su, W., and Zhao, L. Assumption lean regression. The American Statistician, pp.\ 1--23, 2019. doi:10.1080/00031305.2019.1592781

    arXiv 2019

Show all 86 references
  1. [9]

    J., Klaassen, J., Ritov, Y., and Wellner, J

    Bickel, P. J., Klaassen, J., Ritov, Y., and Wellner, J. A. Efficient and Adaptive Estimation for Semiparametric Models. Johns Hopkins University Press Baltimore, 1993

    1993

  2. [10]

    Models as approximations i: consequences illustrated with linear regression

    Buja, A., Brown, L., Berk, R., George, E., Pitkin, E., Traskin, M., Zhang, K., and Zhao, L. Models as approximations i: consequences illustrated with linear regression. Statistical Science, 34 0 (4): 0 523--544, 2019

    2019

  3. [11]

    Buonaccorsi, J. P. Measurement Error: Models, Methods, and Applications. Chapman and Hall/CRC, 2010

    2010

  4. [12]

    and Guo, Z

    Cai, T. and Guo, Z. Semisupervised inference for explained variance in high dimensional linear regression and its applications. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 82 0 (2): 0 391--419, 2020

    2020

  5. [13]

    J., Ruppert, D., Stefanski, L

    Carroll, R. J., Ruppert, D., Stefanski, L. A., and Crainiceanu, C. M. Measurement Error in Nonlinear Models: A Modern Perspective. Chapman and Hall/CRC, 2006

    2006

  6. [14]

    and Cai, T

    Chakrabortty, A. and Cai, T. Efficient and adaptive linear regression in semi-supervised settings. Ann. Statist., 46 0 (4): 0 1541--1572, 2018. doi:10.1214/17-AOS1594

    2018 doi

  7. [15]

    Semi-supervised learning

    Chapelle, O., Scholkopf, B., and Zien, A. Semi-supervised learning. IEEE Transactions on Neural Networks, 20 0 (3): 0 542--542, 2009

    2009

  8. [16]

    and White, H

    Chen, X. and White, H. Improved rates and asymptotic normality for nonparametric neural network estimators. IEEE Transactions on Information Theory, 45 0 (2): 0 682--691, 1999

    1999

  9. [17]

    N., and Cai, T

    Cheng, D., Ananthakrishnan, A. N., and Cai, T. Robust and efficient semi-supervised estimation of average treatment effects with application to electronic health records data. Biometrics, 77 0 (2): 0 413--423, 2021

    2021

  10. [18]

    Double/debiased machine learning for treatment and structural parameters: Double/debiased machine learning

    Chernozhukov, V., Chetverikov, D., Demirer, M., Duflo, E., Hansen, C., Newey, W., and Robins, J. Double/debiased machine learning for treatment and structural parameters: Double/debiased machine learning. The Econometrics Journal, 21 0 (1), 2018

    2018

  11. [19]

    Applied causal inference powered by ml and ai

    Chernozhukov, V., Hansen, C., Kallus, N., Spindler, M., and Syrgkanis, V. Applied causal inference powered by ml and ai. arXiv preprint arXiv:2403.02467, 2024

    2024 arXiv

  12. [20]

    A computational approach to politeness with application to social factors

    Danescu-Niculescu-Mizil, C., Sudhof, M., Jurafsky, D., Leskovec, J., and Potts, C. A computational approach to politeness with application to social factors. arXiv preprint arXiv:1306.6078, 2013

    2013 arXiv

  13. [21]

    Optimal and safe estimation for high-dimensional semi-supervised learning

    Deng, S., Ning, Y., Zhao, J., and Zhang, H. Optimal and safe estimation for high-dimensional semi-supervised learning. Journal of the American Statistical Association, 119 0 (548): 0 2748--2759, 2024

    2024

  14. [22]

    Du Plessis, M. C. and Sugiyama, M. Semi-supervised learning of class balance under class-prior change by distribution matching. Neural Networks, 50: 0 110--119, 2014

    2014

  15. [23]

    Elliott, M. R. Surrogate endpoints in clinical trials. Annual Review of Statistics and its Application, 10 0 (1): 0 75--96, 2023

    2023

  16. [24]

    and Kennedy, E

    Fisher, A. and Kennedy, E. H. Visually communicating and teaching intuition for influence functions. The American Statistician, 75 0 (2): 0 162--172, 2021

    2021

  17. [25]

    B., and Han, L

    Gao, C., Gilbert, P. B., and Han, L. On the role of surrogates in conformal inference of individual causal effects. arXiv preprint arXiv:2412.12365, 2024

    2024 arXiv

  18. [26]

    A unified view of label shift estimation

    Garg, S., Wu, Y., Balakrishnan, S., and Lipton, Z. A unified view of label shift estimation. Advances in Neural Information Processing Systems, 33: 0 3290--3300, 2020

    2020

  19. [27]

    J., and Jurafsky, D

    Gligori \'c , K., Zrnic, T., Lee, C., Cand \`e s, E. J., and Jurafsky, D. Can unconfident llm annotations be used for confident conclusions? arXiv preprint arXiv:2408.15204, 2024

    2024 arXiv

  20. [28]

    Covariate shift by kernel mean matching

    Gretton, A., Smola, A., Huang, J., Schmittfull, M., Borgwardt, K., and Sch \"o lkopf, B. Covariate shift by kernel mean matching. Dataset Shift in Machine Learning, 3 0 (4): 0 5, 2009

    2009

  21. [29]

    R., and Cheng, D

    Gronsbell, J., Gao, J., Shi, Y., McCaw, Z. R., and Cheng, D. Another look at inference after prediction. arXiv preprint arXiv:2411.19908, 2024

    2024

  22. [30]

    Label correction of crowdsourced noisy annotations with an instance-dependent noise transition model

    Guo, H., Wang, B., and Yi, G. Label correction of crowdsourced noisy annotations with an instance-dependent noise transition model. In Advances in Neural Information Processing Systems, 2024

    2024

  23. [31]

    Top challenges from the first practical online controlled experiments summit

    Gupta, S., Kohavi, R., Tang, D., Xu, Y., Andersen, R., Bakshy, E., Cardin, N., Chandran, S., Chen, N., Coey, D., et al. Top challenges from the first practical online controlled experiments summit. ACM SIGKDD Explorations Newsletter, 21 0 (1): 0 20--35, 2019

    2019

  24. [32]

    Demystifying statistical learning based on efficient influence functions

    Hines, O., Dukes, O., Diaz-Ordaz, K., and Vansteelandt, S. Demystifying statistical learning based on efficient influence functions. The American Statistician, 76 0 (3): 0 292--304, 2022

    2022

  25. [33]

    Surrogate assisted semi-supervised inference for high dimensional risk prediction

    Hou, J., Guo, Z., and Cai, T. Surrogate assisted semi-supervised inference for high dimensional risk prediction. Journal of Machine Learning Research, 24 0 (265): 0 1--58, 2023

    2023

  26. [34]

    Correcting sample selection bias by unlabeled data

    Huang, J., Gretton, A., Borgwardt, K., Sch \"o lkopf, B., and Smola, A. Correcting sample selection bias by unlabeled data. Advances in Neural Information Processing Systems, 19, 2006

    2006

  27. [35]

    and Newey, W

    Ichimura, H. and Newey, W. K. The influence function of semiparametric estimators. Quantitative Economics, 13 0 (1): 0 29--61, 2022

    2022

  28. [36]

    Long-term causal inference under persistent confounding via data combination

    Imbens, G., Kallus, N., Mao, X., and Wang, Y. Long-term causal inference under persistent confounding via data combination. Journal of the Royal Statistical Society Series B: Statistical Methodology, pp.\ qkae095, 2024

    2024

  29. [37]

    Maximum mean discrepancy for class ratio estimation: Convergence bounds and kernel selection

    Iyer, A., Nath, S., and Sarawagi, S. Maximum mean discrepancy for class ratio estimation: Convergence bounds and kernel selection. In International Conference on Machine Learning, pp.\ 530--538. PMLR, 2014

    2014

  30. [38]

    Predictions as surrogates: Revisiting surrogate outcomes in the age of ai

    Ji, W., Lei, L., and Zrnic, T. Predictions as surrogates: Revisiting surrogate outcomes in the age of ai. arXiv preprint arXiv:2501.09731, 2025

    2025 arXiv

  31. [39]

    and Mao, X

    Kallus, N. and Mao, X. On the role of surrogates in the efficient estimation of treatment effects with limited outcome data. Journal of the Royal Statistical Society Series B: Statistical Methodology, pp.\ qkae099, 2024

    2024

  32. [40]

    Kelly, S. J. and Ismail, M. Stress and type 2 diabetes: a review of how stress contributes to the development of type 2 diabetes. Annual Review of Public Health, 36 0 (1): 0 441--462, 2015

    2015

  33. [41]

    Kennedy, E. H. Semiparametric doubly robust targeted double machine learning: a review. Handbook of Statistical Methods for Precision Medicine, pp.\ 207--236, 2024

    2024

  34. [42]

    Retasa: A nonparametric functional estimation approach for addressing continuous target shift

    Kim, H., Zhang, X., Zhao, J., and Tian, Q. Retasa: A nonparametric functional estimation approach for addressing continuous target shift. In The Twelfth International Conference on Learning Representations, 2024

    2024

  35. [43]

    Semi-supervised regression: A recent review

    Kostopoulos, G., Karlos, S., Kotsiantis, S., and Ragos, O. Semi-supervised regression: A recent review. Journal of Intelligent & Fuzzy Systems, 35 0 (2): 0 1483--1500, 2018

    2018

  36. [44]

    Kouw, W. M. and Loog, M. A review of domain adaptation without target labels. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43 0 (3): 0 766--785, 2021. doi:10.1109/TPAMI.2019.2945942

    2021

  37. [45]

    and Martinet, G

    Kpotufe, S. and Martinet, G. Marginal singularity and the benefits of labels in covariate-shift. The Annals of Statistics, 49 0 (6): 0 3299--3323, 2021

    2021

  38. [46]

    and Sch \"o lkopf, B

    Lawrence, N. and Sch \"o lkopf, B. Estimating a kernel Fisher discriminant in the presence of label noise. In International Conference on Machine Learning, 2001

    2001

  39. [47]

    Doubly flexible estimation under label shift

    Lee, S.-h., Ma, Y., and Zhao, J. Doubly flexible estimation under label shift. Journal of the American Statistical Association, 120 0 (549): 0 278--290, 2025

    2025

  40. [48]

    T., Bost, S., Lyu, T., Wu, Y., Hogan, W., Prosperi, M., et al

    Li, P., Spector, E., Alkhuzam, K., Patel, R., Donahoo, W. T., Bost, S., Lyu, T., Wu, Y., Hogan, W., Prosperi, M., et al. Developing an automated algorithm for identification of children and adolescents with diabetes using electronic health records from the oneflorida+ clinical...

    2025

  41. [49]

    Provably end-to-end label-noise learning without anchor points

    Li, X., Liu, T., Han, B., Niu, G., and Sugiyama, M. Provably end-to-end label-noise learning without anchor points. In International Conference on Machine Learning, 2021

    2021

  42. [50]

    Detecting and correcting for label shift with black box predictors

    Lipton, Z., Wang, Y.-X., and Smola, A. Detecting and correcting for label shift with black box predictors. In International Conference on Machine Learning, pp.\ 3122--3130. PMLR, 2018

    2018

  43. [51]

    Identifiability of label noise transition matrix

    Liu, Y., Cheng, H., and Zhang, K. Identifiability of label noise transition matrix. In International Conference on Machine Learning, 2023

    2023

  44. [52]

    Improved estimators for semi-supervised high-dimensional regression model

    Livne, I., Azriel, D., and Goldberg, Y. Improved estimators for semi-supervised high-dimensional regression model. Electronic Journal of Statistics, 16 0 (2): 0 5437--5487, 2022

    2022

  45. [53]

    Assumption-lean and data-adaptive post-prediction inference

    Miao, J., Miao, X., Wu, Y., Zhao, J., and Lu, Q. Assumption-lean and data-adaptive post-prediction inference. arXiv preprint arXiv:2311.14220, 2023

    2023 arXiv

  46. [54]

    Valid inference for machine learning-assisted genome-wide association studies

    Miao, J., Wu, Y., Sun, Z., Miao, X., Lu, T., Zhao, J., and Lu, Q. Valid inference for machine learning-assisted genome-wide association studies. Nature Genetics, 56 0 (11): 0 2361--2369, 2024

    2024

  47. [55]

    and Witten, D

    Motwani, K. and Witten, D. Revisiting inference after prediction. Journal of Machine Learning Research, 24 0 (394): 0 1--18, 2023

    2023

  48. [56]

    D., Christoffel, M., and Sugiyama, M

    Nguyen, T. D., Christoffel, M., and Sugiyama, M. Continuous target shift adaptation in supervised learning. In Asian Conference on Machine Learning, pp.\ 285--300. PMLR, 2016

    2016

  49. [57]

    Prentice, R. L. Surrogate endpoints in clinical trials: definition and operational criteria. Statistics in Medicine, 8 0 (4): 0 431--440, 1989

    1989

  50. [58]

    Quinonero-Candela, J., Sugiyama, M., Schwaighofer, A., and Lawrence, N. D. Dataset Shift in Machine Learning. MIT Press, 2008

    2008

  51. [59]

    Generalization error bounds in semi-supervised classification under the cluster assumption

    Rigollet, P. Generalization error bounds in semi-supervised classification under the cluster assumption. Journal of Machine Learning Research, 8, 05 2006

    2006

  52. [60]

    In all likelihoods: Robust selection of pseudo-labeled data

    Rodemann, J., Jansen, C., Schollmeyer, G., and Augustin, T. In all likelihoods: Robust selection of pseudo-labeled data. In International Symposium on Imprecise Probability: Theories and Applications, pp.\ 412--425. PMLR, 2023

    2023

  53. [61]

    On causal and anticausal learning

    Sch \"o lkopf, B., Janzing, D., Peters, J., Sgouritsa, E., Zhang, K., and Mooij, J. On causal and anticausal learning. In 29th International Conference on Machine Learning (ICML 2012), pp.\ 1255--1262. Omnipress, 2012

    2012

  54. [62]

    A rate of convergence for mixture proportion estimation, with application to learning from noisy labels

    Scott, C. A rate of convergence for mixture proportion estimation, with application to learning from noisy labels . In International Conference on Artificial Intelligence and Statistics, 2015

    2015

  55. [63]

    Improving predictive inference under covariate shift by weighting the log-likelihood function

    Shimodaira, H. Improving predictive inference under covariate shift by weighting the log-likelihood function. Journal of Statistical Planning and Inference, 90 0 (2): 0 227--244, 2000

    2000

  56. [64]

    A general m-estimation theory in semi-supervised framework

    Song, S., Lin, Y., and Zhou, Y. A general m-estimation theory in semi-supervised framework. Journal of the American Statistical Association, pp.\ 1--11, 2023

    2023

  57. [65]

    and Cowie, C

    Stark Casagrande, S. and Cowie, C. C. Health insurance coverage among people with and without diabetes in the us adult population. Diabetes Care, 35 0 (11): 0 2243--2249, 2012

    2012

  58. [66]

    When training and test sets are different: characterizing learning transfer

    Storkey, A. When training and test sets are different: characterizing learning transfer. Dataset Shift in Machine Learning, 30: 0 3--28, 2009

    2009

  59. [67]

    and Kawanabe, M

    Sugiyama, M. and Kawanabe, M. Machine Learning in Non-Stationary Environments: Introduction to Covariate Shift Adaptation. MIT press, 2012

    2012

  60. [68]

    Direct importance estimation for covariate shift adaptation

    Sugiyama, M., Suzuki, T., Nakajima, S., Kashima, H., von B \"u nau, P., and Kawanabe, M. Direct importance estimation for covariate shift adaptation. Annals of the Institute of Statistical Mathematics, 60 0 (4): 0 699--746, 2008

    2008

  61. [69]

    Fisher consistency for prior probability shift

    Tasche, D. Fisher consistency for prior probability shift. The Journal of Machine Learning Research, 18 0 (1): 0 3338--3369, 2017

    2017

  62. [70]

    Elsa: Efficient label shift adaptation through the lens of semiparametric models

    Tian, Q., Zhang, X., and Zhao, J. Elsa: Efficient label shift adaptation through the lens of semiparametric models. In International Conference on Machine Learning, pp.\ 34120--34142. PMLR, 2023

    2023

  63. [71]

    Inferring the long-term causal effects of long-term treatments from short-term experiments

    Tran, A., Bibaut, A., and Kallus, N. Inferring the long-term causal effects of long-term treatments from short-term experiments. arXiv preprint arXiv:2311.08527, 2023

    2023 arXiv

  64. [72]

    Tsiatis, A. A. Semiparametric Theory and Missing Data. New York: Springer, 2006

    2006

  65. [73]

    Van der Laan, M. J. and Rose, S. Targeted Learning: Causal Inference for Observational and Experimental Data, volume 4. Springer, 2011

    2011

  66. [74]

    J., Polley, E

    Van der Laan, M. J., Polley, E. C., and Hubbard, A. E. Super learner. Statistical Applications in Genetics and Molecular Biology, 6 0 (1), 2007

    2007

  67. [75]

    van der Vaart, A. W. Asymptotic Statistics. Cambridge University Press, 1998

    1998

  68. [76]

    and Walther, G

    Wager, S. and Walther, G. Adaptive concentration of regression trees, with application to random forests. arXiv preprint arXiv:1503.06388, 2015

    2015 arXiv

  69. [77]

    H., and Leek, J

    Wang, S., McCormick, T. H., and Leek, J. T. Methods for correcting inference based on outcomes predicted by machine learning. Proceedings of the National Academy of Sciences, 117 0 (48): 0 30266--30275, 2020

    2020

  70. [78]

    Usb: A unified semi-supervised learning benchmark for classification

    Wang, Y., Chen, H., Fan, Y., Sun, W., Tao, R., Hou, W., Wang, R., Yang, L., Zhou, Z., Guo, L.-Z., et al. Usb: A unified semi-supervised learning benchmark for classification. Advances in Neural Information Processing Systems, 35: 0 3938--3961, 2022

    2022

  71. [79]

    and Lafferty, J

    Wasserman, L. and Lafferty, J. D. Statistical analysis of semi-supervised regression. In Platt, J. C., Koller, D., Singer, Y., and Roweis, S. T. (eds.), Advances in Neural Information Processing Systems 20, pp.\ 801--808. Curran Associates, Inc., 2008

    2008

  72. [80]

    Yi, G. Y. Statistical Analysis with Measurement Error or Misclassification: Strategy, Method and Application. Springer, 2017

    2017

  73. [81]

    Yi, G. Y. Likelihood methods with measurement error and misclassification. In Handbook of measurement error models, pp.\ 99--126. Chapman and Hall/CRC, 2021

    2021

  74. [82]

    D., and Cai, T

    Zhang, A., Brown, L. D., and Cai, T. T. Semi-supervised inference: General theory and estimation of means. Ann. Statist., 47 0 (5): 0 2538--2566, 10 2019. doi:10.1214/18-AOS1756

    2019 doi

  75. [83]

    Domain adaptation under target and conditional shift

    Zhang, K., Sch \"o lkopf, B., Muandet, K., and Wang, Z. Domain adaptation under target and conditional shift. In International Conference on Machine Learning, pp.\ 819--827. PMLR, 2013

    2013

  76. [84]

    Evaluating the surrogate index as a decision-making tool using 200 a/b tests at netflix

    Zhang, V., Zhao, M., Le, A., and Kallus, N. Evaluating the surrogate index as a decision-making tool using 200 a/b tests at netflix. arXiv preprint arXiv:2311.11922, 2023

    2023 arXiv

  77. [85]

    and Bradic, J

    Zhang, Y. and Bradic, J. High-dimensional semi-supervised learning: in search of optimal inference of the mean. Biometrika, 109 0 (2): 0 387--403, 2022

    2022

  78. [86]

    Zhu, X. J. Semi-supervised learning literature survey. Technical report, University of Wisconsin-Madison Department of Computer Sciences, 2005

    2005

Pith tools

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