Multimodal domain adaptation under label shift and blockwise missing modalities

Molei Liu; Tianxi Cai; Zebin Wang; Ziang Dou

REVIEW 2 major objections 3 minor 60 references

Multimodal domain adaptation under label shift and blockwise missing modalities

T0 review · 2 major / 3 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A reference-anchored, reweight-before-align procedure is claimed to recover a common target-defined multimodal representation and consistent target conditional outcome estimates under label shift and blockwise missing modalities.

desk verdict Novel reference-anchored alignment with an honest but load-bearing assumption gap: the one-rotation result is solid, while the posterior-consistency headline rides on an unproven transport condition. read the letter →

arxiv 2608.01282 v1 pith:2JEAWTLG submitted 2026-08-02 stat.ME

Zebin Wang , Ziang Dou , Molei Liu , Tianxi Cai This is my paper · ORCID

classification stat.ME MSC 62H2062G0562G20

keywords domainadaptationlabelshiftblockwisemissingmodalitiescanonicalcorrelationanalysisridgeregressionsurrogateoutcomesmultimodaldatafusiontargetcalibration

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 aims to show that multimodal domain adaptation remains possible when the unlabeled target population differs from labeled sources in its outcome distribution and each source observes only a subset of modalities. The proposed route is to first use a reference modality present in every domain to estimate the target outcome mixture, reweight source observations to that mixture, and only then align auxiliary modalities to a target-defined canonical correlation analysis anchor using ridge maps. Its central result is that this reweight-before-align construction recovers all fitted aligned coordinates up to one common rotation at rate $\Delta_n$, and that under a separate likelihood-ratio condition the estimated target conditional outcome distribution is consistent at rate $n_0^{-1/2}+\kappa_n+\Delta_n$. If correct, the method yields calibrated target probabilities without requiring sources to share a common input space, and it degrades gracefully when gold-standard labels are sparse by borrowing surrogate outcomes.

What carries the argument

The central object is the target-calibrated, reference-anchored match-up. A reference modality observed in every domain is used to estimate the target outcome distribution via a profile likelihood; each source subject is reweighted by the estimated label-shift ratio $\widehat{\omega}_m(y)$, and the target defines a canonical correlation analysis (CCA, a method that finds linear combinations of two variable sets with maximal correlation) anchor between the reference scores and the concatenated auxiliary scores. Each source auxiliary block is then mapped into this target-anchored coordinate system by weighted ridge regression, so that all aligned coordinates share one common target-defined orientation up to a single orthogonal rotation. The proof's key move is to apply a subspace-perturbation theorem to a symmetric dilation of the CCA operator, which yields one common rotation $O$ for the target CCA pair, all source ridge maps, and every fitted coordinate block; ridge equivariance then propagates this same $O$ through the source normal equations.

What would settle it

Generate two synthetic replications that differ only in whether the class-conditional distribution of the reference score given the outcome is exactly transportable across source and target; the pipeline should show the claimed error rate in the transportable case and a systematic calibration gap in the non-transportable case, so observing equal errors would falsify the transport assumption.

Watch

Extended reading notes

Core claim

The central claim, established as Lemma 2 and Theorem 1, is that a target-defined representation can be learned from heterogeneous source blocks despite label shift and blockwise missingness. Under the label-shift and overlap assumption, the target outcome distribution is identified from the reference modality, and source observations are reweighted so that their marginal cross-modal associations match target associations; the target CCA anchor and every source ridge map are then shown to be recovered up to one common orthogonal rotation $O$ at rate $\Delta_n$, with the same $O$ appearing in all domains and modalities. Under the additional Assumption 3 that the aligned likelihood-ratio estimator is consistent at population coordinates, the target outcome parameter and the target conditional outcome distribution are consistent, at rate $n_0^{-1/2}+\kappa_n+\Delta_n$. The paper is careful to distinguish representation recovery from outcome transport: common orientation alone does not carry source class-conditional densities to the target, so the ratio condition is a separate load-bearing requirement. The same pipeline with a surrogate bridge is claimed to approach the full-label oracle when source gold labels are sparse, supported by simulations and by an application to 12-month recurrence prediction in renal cell carcinoma.

Load-bearing premise

The load-bearing premise is that, once the aligned coordinates are fixed, the outcome probabilities given those coordinates can be estimated consistently from reweighted source data at a known rate and the target outcome mixture is uniquely identifiable from that estimate; for the boosted-tree implementation in the paper this remains an assumption rather than a proven rate.

Editorial extensions

If this is right

When this pipeline is deployed, predicted risks in the target remain calibrated even if source prevalence differs sharply from target prevalence, because the outcome mixture is corrected before alignment rather than as a post-hoc recalibration step.
Auxiliary modalities can enter some sources and not others without being pooled into one shared input space; each observed block is projected into the target-anchored coordinate system by its own ridge map.
With sparse gold-standard labels, the surrogate-assisted version recovers most of the ranking performance of a full-label procedure and, with a small fraction of gold labels, matches its probability calibration at larger sample sizes (within 0.3% of oracle AUC at the largest simulated size).
Because the fitted coordinates are identified up to one common rotation, and because that rotation does not alter the conditional outcome distribution, any downstream learner applied to the aligned coordinates inherits the same consistency guarantee.

Reading between the lines

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

The one-common-rotation result is a representation-level guarantee: it implies the aligned coordinates are a well-defined target-defined feature map, so the same match-up could serve downstream tasks other than the specific likelihood-ratio profile used here, such as clustering or classification with a different outcome model.
The reweight-before-align principle may transfer to other anchor-based alignment estimators; as long as the primitive error bounds in Assumption 2 hold, replacing CCA by another anchor estimator should preserve the common-rotation property.
A direct testable extension would compare this method with an align-first, reweight-second variant in the same simulation design; the paper's ablation isolates the CCA/ridge component but does not separately measure the ordering effect, which the simulations do not directly report.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Colleague,

Worth knowing: the alignment core is genuine and new; the headline consistency theorem is not fully proven as advertised. Lemma 2, deriving one common rotation for target CCA and all source ridge maps from standard perturbation theory, is the paper's real contribution and holds up. The "weight before alignment" idea—estimate the target outcome mixture from the reference modality, reweight source observations, then align—is a genuinely useful construction for EHR-style blockwise missingness. The literature survey is careful and positions the work correctly.

The soft spot is exactly where the stress test points. Theorem 1's consistency and rate for the target posterior are conditioned on Assumption 3, which assumes both consistency of the aligned likelihood-ratio learner and unique identification of theta0. More importantly, Assumption 3 embeds the density transport step: that the class-conditional laws of the source ridge coordinates coincide with those of the target CCA coordinates after alignment. Lemma 2 explicitly does not establish this; the paper says so after Eq (18), and the supplement's rate ledger admits that for the boosted-tree implementation "both clauses remain assumed." So the abstract's consistency claim is conditional on an unverified premise. This is not a secret flaw—it is stated honestly—but it should be moved to the front: either present Theorem 1 as a conditional result with Assumption 3 flagged as a separate research problem, or prove transport under structural conditions on loadings and score distributions.

Other concerns are minor. The EM tilt-back safeguard is active in about 95% of simulation replications, so the simulations mostly characterize the stabilized update rather than unrestricted EM; the paper discloses this clearly. Bootstrap standard errors without refitting are common in this literature but do overstate precision. The surrogate pathway is not covered by the rate theorem, and the real-data endpoint is mostly LATTE-derived; the supplement says plainly this is a mixed-endpoint transport analysis, not adjudicated validation.

Bottom line: the method is practical, the alignment theory is worthwhile, and the posterior-consistency theorem needs re-framing or strengthening. I would send this to a serious referee; conditional acceptance with targeted revisions is the right outcome. I'd also bring it to a reading group as an example of unusually honest assumption accounting.

Referee Report

2 major / 3 minor

Summary. The paper proposes a four-stage pipeline for multimodal domain adaptation under label shift and blockwise missingness: (1) unsupervised representation learning; (2) reference-based estimation of the target outcome distribution and reweighting of source observations; (3) target-anchored CCA/ridge alignment after reweighting; and (4) likelihood-ratio-based outcome transfer to estimate the target conditional outcome distribution. The main theoretical contributions are Lemma 2, establishing that the target CCA anchor, all source ridge maps, and all fitted aligned coordinates are recovered up to one common rotation at rate Delta_n, and Theorem 1, giving consistency and rates for the target conditional outcome distribution under Assumptions 1–3. The paper also develops a surrogate-label-assisted variant for sparse gold-standard source labels and reports simulations plus a renal cell carcinoma application.

Significance. The 'weight before alignment' idea is well motivated, and the one-common-rotation alignment result is a useful technical step for combining target CCA with source ridge regression under blockwise missingness. The paper is unusually transparent about the limits of its theory, explicitly stating in the supplement that the likelihood-ratio learner's consistency is assumed rather than derived for the boosted-tree implementation. If the outcome-transfer step were justified for a concrete ratio learner, the framework would be a solid contribution. As it stands, the central advertised claim—consistency of the target conditional outcome distribution—is conditional on an assumption that is not verified for the implemented method, which substantially weakens the contribution as stated.

major comments (2)

[§4.2–§4.3, Assumption 3; Supplement S1.1, rate ledger item 4] Theorem 1's consistency claim for the target posterior is assumed, not derived, for the implementation actually used. Assumption 3 (Assumption 6 in the supplement) postulates both that the aligned likelihood-ratio estimator is consistent at rate kappa_n and that the oracle profile criterion identifies theta0. The supplement's own rate ledger states that for boosted trees 'both clauses remain assumed,' and the implementation indeed uses XGBoost-based density-ratio estimation. Therefore Theorem 1 does not establish consistency of the target conditional outcome distribution for the proposed pipeline; it reduces that claim to an unverified high-level condition. The authors should either supply a rate theorem for a concrete ratio-learning procedure under explicit conditions (e.g., a correctly specified parametric ratio model) or restate the contribution as a reduction theorem and confine the exact-posterior consistency claim to the 'working conditional distribution' mentioned only in the final paragraph of Section 4.3.
[§4.1, definition of G^(m); §4.2, Assumption 3; remark after Eq. (18)] The population ridge map G^(m) is defined as the minimizer of a weighted least-squares fit of A^(0)^T r onto z, whereas the target CCA map B^(0) solves a constrained correlation problem. These maps need not coincide, so the class-conditional law of the source aligned coordinates v^(m) need not equal that of the target aligned coordinates v^(0) even when the raw scores are transportable under Assumption 1. Lemma 2 only recovers each fitted coordinate up to a common rotation relative to its own population map, and the paper's own remark after Eq. (18) explicitly disclaims that common orientation implies likelihood-ratio transport. The 'intrinsic transfer rate' kappa_n in Assumption 3 therefore silently absorbs a source-to-target transport bias that the alignment result does not control. The paper should state explicit structural conditions under which the ridge-mapped source coordinates have the same outcome-conditional distribution as the target CCA coordinates, and ideally verify such conditions in the simulation design; otherwise the outcome-transfer step remains an unverified premise.

minor comments (3)

[Abstract and §1.3] The abstract and contribution section state consistency of the target conditional outcome distribution as a headline result without the qualification that appears only at the end of Section 4.3, where the authors note that the composite ratio (8) may define a working conditional distribution rather than the exact target posterior. The claims should be aligned with the theorem's actual scope.
[§3.4, Eq. (8)] The composite ratio is said to correspond to the joint conditional likelihood ratio when the aligned auxiliary blocks are conditionally independent given (U,Y). The paper does not state whether this conditional independence is part of the main theoretical assumptions or only an interpretation; clarifying this would help the reader assess the gap between the composite construction and the exact ratio in Theorem 1.
[Supplement S2.5 and Table S8] The ablation terminology is inconsistent: the main text's 'target-anchor CCA/ridge match-up' is sometimes called the 'CCA switch' and sometimes 'target-anchor CCA,' while Table S8 lists four variants with slightly different names. Aligning the terminology across the main text, supplement, and tables would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the alignment result is derived from external perturbation bounds, and Theorem 1 is explicitly conditional on Assumption 3, whose unverified status for boosted trees is a stated limitation rather than a circular reduction.

full rationale

The claimed derivation chain has two branches. First, the unified-rotation alignment result (Lemma 2, Section 4.3; Lemmas S1.3–S1.4 in the Supplement) is derived from Assumptions 1–2 using standard external ingredients: covariance concentration (Vershynin 2018), Davis–Kahan/Wedin singular-subspace perturbation (Yu et al. 2015, Wedin 1972), and matrix square-root bounds (Higham 2008). The weights entering the alignment are obtained from reference-score calibration (Eqs. 3–4), which depends only on the reference modality and pooled source labels; target labels are masked throughout. No fitted target outcome distribution is used as an input to the rotation result, and the common-rotation statement is the paper's own deduction from the symmetric-dilation perturbation theorem, not an imported uniqueness claim by the same authors. Second, the target posterior consistency in Theorem 1 is explicitly conditional on Assumption 3 (Eq. 15), which the paper labels a 'separate nuisance condition.' The remark after Eq. (18) is transparent: common orientation alone does not imply source-to-target likelihood-ratio transport, and transport is exactly the separate condition in Assumption 3. Thus the posterior conclusion is a plug-in/argmax consequence of a clearly stated high-level assumption, not a conclusion that is secretly identical to its input. The Supplement's rate ledger (S1.1, item 4) states that for boosted trees 'both clauses remain assumed' and that the surrogate pathway is not covered by the rate theorem; these are explicit limitations of the implementation, not instances of the paper presenting an assumed rate as a derived prediction. The paper does not fit a parameter to target outcomes and then rename the fit as a prediction, nor does it define the aligned coordinates in terms of the target posterior. Self-citations (e.g., Liu et al. 2023, Cai et al. 2016, Zhou et al. 2023, Hou et al. 2026, Wen et al. 2024) are used for baselines, data assets, and feature-extraction pipelines rather than to supply the load-bearing consistency or uniqueness theorems, which cite external sources. Accordingly, no step in the paper's derivation reduces by construction to its own inputs, and the conditional nature of Theorem 1 is a correctness/scope limitation rather than circularity.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim rests on standard label-shift transport, eigenvalue perturbation theory, and stable CCA and ridge assumptions, plus a high-level assumption on the density-ratio learner that is not proven for the actual implementation. Several tuning constants shape the evaluated procedure. No new physical entities are introduced.

free parameters (6)

Target CCA rank r0 = 10 (real data); 5 (simulations)
Rank of the aligned representation and the anchor subspace; fixed by the analyst and the theory assumes a positive singular-value gap at this rank.
CCA ridge lambda_z = 1e-4 (simulations)
Regularization on the auxiliary side of the one-sided CCA; chosen by the authors to stabilize overlapping modality blocks.
Source ridge lambda_m = 1e-4 (simulations); selected from a fixed grid by validation
Ridge penalty for each source ridge alignment map; the theory assumes lambda_m in a fixed positive range.
EM tilt-back safeguard (L1 threshold 0.15, damping schedule) = threshold 0.15, damping 0.4 to 0.05
Ad hoc stabilization that constrains the target update; active in about 95% of simulation replications, so reported results characterize this safeguarded update.
Auxiliary contribution multiplier = 0.5
Any auxiliary block that passes validation contributes with its ratio multiplied by 0.5; an ad hoc shrinkage affecting predicted probabilities.
Surrogate semi-supervised calibration threshold and shrinkage = >=150 gold observations, both classes; shrinkage 10
Activation rules for the monotone intercept correction in the semi-supervised surrogate variant.

assumptions (6)

domain assumption Assumption 1 (label shift and overlap): class-conditional score distributions are stable across domains, target support is contained in source support.
Underlies the reweighting identity E_m{omega_m(Y) f(R,Z)} = E_0{f(...)} in Lemma 1; if the class-conditional features shift beyond label shift, the weights are mis-specified.
domain assumption Assumption 2 (stable regularized alignment): bounded moments, eigenvalues bounded away from zero on retained subspaces, positive CCA singular-value gap delta_CCA, ridge penalties in a fixed positive range, uniform O_p(Delta_n) moment errors.
Needed for the one-common-rotation perturbation argument (Lemma 2); the gap assumption is untestable and can fail in near-degenerate CCA problems.
ad hoc to paper Assumption 3 (aligned ratio consistency and profile identification): the likelihood-ratio estimator is consistent at rate kappa_n and the oracle profile criterion has unique maximizer theta0.
Explicitly assumed, not derived; the supplement states for boosted-tree learners 'both clauses remain assumed'. This assumption is the main channel through which the outcome transfer conclusion enters Theorem 1.
domain assumption Latent factor model (Section 2): conditional on outcome, latent reference and auxiliary factors are independent with additive noise; observed features are linear loadings of latent factors.
Working model justifying the representation and alignment; the method's guarantees are stated only for retained scores under this structure.
domain assumption Conditional independence of aligned auxiliary blocks given (U,Y) for the composite ratio in Eq. (8).
The product-form likelihood ratio is exact only under this independence; otherwise Theorem 1's conclusion is for a 'working conditional distribution' as the paper states after Eq. (10).
domain assumption Surrogate bridge conditions (Section 3.4): surrogate independent of features given outcome, p_S(s|y) invariant across sources, and identifiability of the bridge.
Supports the surrogate-assisted procedure; the paper's rate theorem explicitly does not cover the surrogate bridge (Supplementary S1.1).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Multimodal domain adaptation under label shift and blockwise missing modalities." pith.science (2026). https://pith.science/paper/2JEAWTLG

@misc{pith2026260801282,
  author       = {Pith},
  title        = {Pith review of: Multimodal domain adaptation under label shift and blockwise missing modalities},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2JEAWTLG}},
  note         = {Machine review of arXiv:2608.01282}
}

read the original abstract

Multimodal domain adaptation uses labeled source datasets to predict outcomes in an unlabeled target population. Here, different sources may observe different subsets of modalities and have distribution shifts from the target. In this scenario, the cross-modal patterns used for alignment depend on the outcome distribution, so aligning unadjusted source data can misrepresent the target. Nevertheless, the distribution shift of the outcome cannot be measured directly when labels are unavailable in the target sample and the missing modality blocks prevent simple pooling. We propose a reference-anchored domain adaptation method. A reference modality observed in every source and the target is used to estimate the target outcome distribution and reweight source observations before alignment. The auxiliary modalities are then mapped to a common, target-defined representation obtained by canonical correlation analysis (CCA) in the target and reproduced in each source by ridge-regression maps. Outcome information is transferred through density-ratio models on the aligned representation. When gold-standard source labels are sparse, a surrogate-label-assisted approach is developed to enable robust domain adaptation. We establish a unified-rotation match-up result for the target CCA and source ridge maps and consistency of the target conditional outcome distribution. Simulations and a renal cell carcinoma (RCC) application show improved calibration and stable prediction under distributional shift and blockwise missing modalities.

Figures

Figures reproduced from arXiv: 2608.01282 by the authors.

**Figure 2.** Main simulation. Panels A and B summarize performance across the four perturbation sweeps [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗

**Figure 3.** Surrogate-assisted simulation. Panel A reports target AUC, for which higher values are preferred, [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 31 canonical work pages

[1]

Proceedings of the IEEE , volume=

Multimodal data fusion: an overview of methods, challenges, and prospects , author=. Proceedings of the IEEE , volume=. 2015 , publisher=

work page 2015
[2]

Multimodal Machine Learning: A Survey and Taxonomy , journal =

Baltru. Multimodal Machine Learning: A Survey and Taxonomy , journal =. 2019 , volume =

work page 2019
[3]

Journal of statistical planning and inference , volume=

Improving predictive inference under covariate shift by weighting the log-likelihood function , author=. Journal of statistical planning and inference , volume=. 2000 , publisher=

2000
[4]

Journal of Machine Learning Research , volume=

Augmented transfer regression learning with semi-non-parametric nuisance models , author=. Journal of Machine Learning Research , volume=
[5]

The Annals of Statistics , volume=

Efficient and multiply robust risk estimation under general forms of dataset shift , author=. The Annals of Statistics , volume=. 2024 , publisher=

work page 2024
[6]

Journal of the American Statistical Association , volume=

Doubly flexible estimation under label shift , author=. Journal of the American Statistical Association , volume=. 2025 , publisher=

2025
[7]

Efficient Inference under Label Shift in Unsupervised Domain Adaptation

Lee, Seong-ho and Ma, Yanyuan and Zhao, Jiwei , title =. arXiv preprint arXiv:2508.17780 , year =

work page Pith review arXiv
[8]

Neural computation , volume=

Adjusting the outputs of a classifier to new a priori probabilities: a simple procedure , author=. Neural computation , volume=. 2002 , publisher=

work page 2002

Show all 60 references

[9]

Advances in Neural Information Processing Systems , volume=

A unified view of label shift estimation , author=. Advances in Neural Information Processing Systems , volume=
[10]

International conference on machine learning , pages=

Detecting and correcting for label shift with black box predictors , author=. International conference on machine learning , pages=. 2018 , organization=

2018
[11]

Journal of the American Statistical Association , volume=

Structured matrix completion with applications to genomic data integration , author=. Journal of the American Statistical Association , volume=. 2016 , publisher=

2016
[12]

Biostatistics , volume=

Generalized integrative principal component analysis for multi-type data with block-wise missing structure , author=. Biostatistics , volume=. 2020 , publisher=

2020
[13]

Journal of Machine Learning Research , volume=

Multi-source learning via completion of block-wise overlapping noisy matrices , author=. Journal of Machine Learning Research , volume=
[14]

Journal of the American Statistical Association , volume=

Optimal sparse linear prediction for block-missing multi-modality data without imputation , author=. Journal of the American Statistical Association , volume=. 2020 , publisher=

2020
[15]

Journal of the American Statistical Association , volume=

Integrating multisource block-wise missing data in model selection , author=. Journal of the American Statistical Association , volume=. 2021 , publisher=

2021
[16]

arXiv preprint arXiv:2409.05729 , year=

Efficient estimation with incomplete data via generalised ANOVA decompositions , author=. arXiv preprint arXiv:2409.05729 , year=

arXiv
[17]

arXiv preprint arXiv:2405.18722 , year=

Adaptive and efficient learning with blockwise missing and semi-supervised data , author=. arXiv preprint arXiv:2405.18722 , year=

arXiv
[18]

arXiv preprint arXiv:2503.09494 , year=

Representation retrieval learning for heterogeneous data integration , author=. arXiv preprint arXiv:2503.09494 , year=
[19]

IEEE Transactions on Image Processing , volume=

Incomplete multi-view learning under label shift , author=. IEEE Transactions on Image Processing , volume=. 2023 , publisher=

2023
[20]

Journal of Computational and Graphical Statistics , pages=

Multi-task learning for heterogeneous multi-source block-wise missing data , author=. Journal of Computational and Graphical Statistics , pages=. 2026 , publisher=

2026
[21]

Foundations and Trends

Graphical models, exponential families, and variational inference , author=. Foundations and Trends. 2008 , publisher=

2008
[22]

Philosophical transactions of the royal society A: Mathematical, Physical and Engineering Sciences , volume=

Principal component analysis: a review and recent developments , author=. Philosophical transactions of the royal society A: Mathematical, Physical and Engineering Sciences , volume=. 2016 , publisher=

2016
[23]

2012 , publisher=

Density ratio estimation in machine learning , author=. 2012 , publisher=

2012
[24]

Relations Between Two Sets of Variates , urldate =

Harold Hotelling , journal =. Relations Between Two Sets of Variates , urldate =
[25]

Neural computation , volume=

Canonical correlation analysis: An overview with application to learning methods , author=. Neural computation , volume=. 2004 , publisher=

2004
[26]

Psychometrika , volume=

Regularized generalized canonical correlation analysis , author=. Psychometrika , volume=. 2011 , publisher=

2011
[27]

Technometrics , volume=

Ridge regularization: An essential concept in data science , author=. Technometrics , volume=. 2020 , publisher=

2020
[28]

Proceedings of the 37th International Conference on Machine Learning , pages =

Maximum Likelihood with Bias-Corrected Calibration is Hard-To-Beat at Label Shift Adaptation , author =. Proceedings of the 37th International Conference on Machine Learning , pages =. 2020 , editor =

2020
[29]

Bernoulli , pages=

Concentration inequalities and moment bounds for sample covariance operators , author=. Bernoulli , pages=. 2017 , publisher=

2017
[30]

Biometrika , volume=

A useful variant of the Davis--Kahan theorem for statisticians , author=. Biometrika , volume=. 2015 , publisher=

2015
[31]

BIT Numerical Mathematics , volume=

Perturbation bounds in connection with singular value decomposition , author=. BIT Numerical Mathematics , volume=. 1972 , publisher=

1972
[32]

2000 , publisher=

Asymptotic statistics , author=. 2000 , publisher=

2000
[33]

Psychometrika , volume=

A generalized solution of the orthogonal procrustes problem , author=. Psychometrika , volume=. 1966 , publisher=

1966
[34]

Journal of the royal statistical society: series B (methodological) , volume=

Maximum likelihood from incomplete data via the EM algorithm , author=. Journal of the royal statistical society: series B (methodological) , volume=. 1977 , publisher=

1977
[35]

Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =

Chen, Tianqi and Guestrin, Carlos , title =. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =. 2016 , publisher =

2016
[36]

ESMO Real World Data and Digital Oncology , volume=

Estimate renal cell carcinoma recurrence rates using electronic health records , author=. ESMO Real World Data and Digital Oncology , volume=. 2026 , publisher=

2026
[37]

Patterns , volume=

LATTE: Label-efficient incident phenotyping from longitudinal electronic health records , author=. Patterns , volume=. 2024 , publisher=

2024
[38]

arXiv preprint arXiv:2509.08553 , year=

A Common Pipeline for Harmonizing Electronic Health Record Data for Translational Research , author=. arXiv preprint arXiv:2509.08553 , year=
[39]

arXiv preprint arXiv:1311.6063 , year=

NILE: fast natural language processing for electronic health records , author=. arXiv preprint arXiv:1311.6063 , year=

arXiv
[40]

Nature methods , volume=

nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation , author=. Nature methods , volume=. 2021 , publisher=

2021
[41]

Cancer research , volume=

Computational radiomics system to decode the radiographic phenotype , author=. Cancer research , volume=. 2017 , publisher=

2017
[42]

International conference on machine learning , pages=

Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

2021
[43]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Learning to exploit temporal structure for biomedical vision-language processing , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=
[44]

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

Regression shrinkage and selection via the lasso , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 1996 , publisher=

1996
[45]

, title =

Brier, Glenn W. , title =. Monthly Weather Review , year =
[46]

climatology

On using “climatology” as a reference strategy in the Brier and ranked probability skill scores , author=. Monthly Weather Review , volume=
[47]

Journal of Machine Learning Research , volume=

Semi-supervised inference for block-wise missing data without imputation , author=. Journal of Machine Learning Research , volume=
[48]

medRxiv preprint 2023.09.29.23296239 , year=

Knowledge-Driven Online Multimodal Automated Phenotyping System , author=. medRxiv preprint 2023.09.29.23296239 , year=

2023
[49]

Vershynin, Roman , title =
[50]

, title =

Higham, Nicholas J. , title =
[51]

and Steyerberg, Ewout W

Van Calster, Ben and Nieboer, Daan and Vergouwe, Yvonne and De Cock, Bavo and Pencina, Michael J. and Steyerberg, Ewout W. , title =. Journal of Clinical Epidemiology , year =
[52]

and Ogundimu, Emmanuel O

Collins, Gary S. and Ogundimu, Emmanuel O. and Altman, Douglas G. , title =. Statistics in Medicine , year =
[53]

BMC Bioinformatics , year =

Robin, Xavier and Turck, Natacha and Hainard, Alexandre and Tiberti, Natalia and Lisacek, Fr. BMC Bioinformatics , year =
[54]

Optimal Transport for Multi-source Domain Adaptation under Target Shift , booktitle =

Redko, Ievgen and Courty, Nicolas and Flamary, R. Optimal Transport for Multi-source Domain Adaptation under Target Shift , booktitle =. 2019 , publisher =

2019
[55]

Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics , series =

Li, Yitong and Murias, Michael and Major, Samantha and Dawson, Geraldine and Carlson, David , title =. Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics , series =. 2019 , publisher =

2019
[56]

Aggregating From Multiple Target-Shifted Sources , booktitle =

Shui, Changjian and Li, Zijian and Li, Jiaqi and Gagn. Aggregating From Multiple Target-Shifted Sources , booktitle =. 2021 , publisher =

2021
[57]

arXiv preprint arXiv:2503.02506 , year =

Xu, Congbin and Qian, Chengde and Wang, Zhaojun and Zou, Changliang , title =. arXiv preprint arXiv:2503.02506 , year =

arXiv
[58]

Multi-source Multi-modal Domain Adaptation , journal =

Zhao, Sicheng and Jiang, Jing and Tang, Wenbo and Zhu, Jiankun and Chen, Hui and Xu, Pengfei and Schuller, Bj. Multi-source Multi-modal Domain Adaptation , journal =
[59]

arXiv preprint arXiv:2602.11511 , year =

Liu, Ziqi and Tian, Ye and Tang, Weijing , title =. arXiv preprint arXiv:2602.11511 , year =
[60]

arXiv preprint arXiv:2607.01821 , year =

Yu, Junhan and Zhang, Kejian and Zhou, Doudou and Zhu, Guojun , title =. arXiv preprint arXiv:2607.01821 , year =

arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Proceedings of the IEEE , volume=

Multimodal data fusion: an overview of methods, challenges, and prospects , author=. Proceedings of the IEEE , volume=. 2015 , publisher=

work page 2015

[2] [2]

Multimodal Machine Learning: A Survey and Taxonomy , journal =

Baltru. Multimodal Machine Learning: A Survey and Taxonomy , journal =. 2019 , volume =

work page 2019

[3] [3]

Journal of statistical planning and inference , volume=

Improving predictive inference under covariate shift by weighting the log-likelihood function , author=. Journal of statistical planning and inference , volume=. 2000 , publisher=

2000

[4] [4]

Journal of Machine Learning Research , volume=

Augmented transfer regression learning with semi-non-parametric nuisance models , author=. Journal of Machine Learning Research , volume=

[5] [5]

The Annals of Statistics , volume=

Efficient and multiply robust risk estimation under general forms of dataset shift , author=. The Annals of Statistics , volume=. 2024 , publisher=

work page 2024

[6] [6]

Journal of the American Statistical Association , volume=

Doubly flexible estimation under label shift , author=. Journal of the American Statistical Association , volume=. 2025 , publisher=

2025

[7] [7]

Efficient Inference under Label Shift in Unsupervised Domain Adaptation

Lee, Seong-ho and Ma, Yanyuan and Zhao, Jiwei , title =. arXiv preprint arXiv:2508.17780 , year =

work page Pith review arXiv

[8] [8]

Neural computation , volume=

Adjusting the outputs of a classifier to new a priori probabilities: a simple procedure , author=. Neural computation , volume=. 2002 , publisher=

work page 2002