Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Sufficiency-principled Transfer Learning via Model Averaging

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read Sufficiency-principled model averaging makes transfer learning work without knowing which sources help.

desk verdict A useful unification of individual and combinatorial transfer similarity, but the main rate theorems may only cover an oracle candidate construction unless the supplement closes the data-dependence gap. read the letter →

arxiv 2507.15416 v1 pith:WDD5LBUJ submitted 2025-07-21 stat.ME math.STstat.TH

classification stat.MEmath.STstat.TH MSC 62J9962B99
keywords transferlearningmodelaveragingsufficiencyprinciplenegativecombinatorialsimilarityindividualasymptoticnormalityq-aggregation
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 establish that transfer learning can be made to work even when the set of transferable sources is completely unknown, by following a sufficiency principle: transfer as much informative knowledge as possible while automatically ignoring harmful sources. It claims this is achievable by model averaging with a penalty that favours larger and better-covering candidate domains, in a unified framework that covers both individual-similarity (each source close to the target) and combinatorial-similarity (a combination of sources close to the target). The theory says the resulting estimators converge faster than target-only OLS and faster than the Trans-Lasso baseline, concentrate their weights on the sufficient informative domain, remain robust to negative transfer, protect privacy through summary-statistic transmission, and admit asymptotic normality for inference.

What carries the argument

The central object is the sufficiency-principled weight criterion $C_{v,\phi}(w)$ of Trans-MAI, built from three terms: the aggregated predictor's squared error, a weighted average of each candidate's prediction error (with tuning $v$), and a sufficiency penalty that charges domains with small sample size or poor coverage of the target design. Candidate domains are nested sets $M_m$ ordered by estimated contrast norms $\|\hat\beta_{(m)}-\hat\beta_{(0)}\|$, which makes the pool of candidate estimators grow naturally and accelerates rates; the criterion is an unbiased risk estimator up to a bias-correction term, and the penalty is what forces weight concentration onto the sufficient informative domain while keeping out non-informative combinations.

What would settle it

Construct a linear regression example with two non-informative sources whose parameters straddle the target so their average equals the target (violating the angular separation condition), run Trans-MAI with $v=0$, and check whether the summed weight on those sources remains non-negligible as sample size grows: the paper's Figure 9 indicates it will not decay, which would falsify the claimed negative-transfer robustness in that regime.

Watch

Extended reading notes

Core claim

The paper claims that a weight-selection criterion mixing prediction loss, candidate-specific losses, and a sufficiency penalty $\phi \hat\sigma^2_{(0)}\sum_m w_m \operatorname{tr}(G_{[m]}^{-1}G_{[0]})$ drives the averaging weights onto the sufficient informative domain and asymptotically to zero on non-informative and insufficient domains. Under the technical conditions, Theorem 1 gives the convergence rate $O_p(\phi^{-1} p^{1/2} n_0^{-1/2}(|A_h|-1)) + O_p(p^{1/2} N_{m_s}^{-1/2}) + O(h) + O_p(\phi n_0^{-1} p^{3/2}\eta^{-2}|A_h^c|)$, which beats the target-only rate when the sufficient informative domain is large and the contrast gap is favourable, and Corollary 1 states it beats Trans-Lasso by preserving sufficiency rather than losing power to q-aggregation. The same principle extends to combinatorial-similarity via Trans-MACs and Trans-MAC, where the weights are selected on the sufficient informative domain rather than directly on the target, yielding the corresponding weight convergence, error bounds, and high-probability optimality.

Load-bearing premise

The load-bearing premise is that no mixture of the non-informative sources can accidentally reconstruct the target: any pair of non-informative contrast directions must have positive covariance-weighted inner product, so no convex combination of bad sources lines up with the target.

Editorial extensions

If this is right

  • If the transferable set is unknown, practitioners no longer need to pre-select sources: the weights automatically shrink non-informative domains toward zero and concentrate on the sufficient informative domain.
  • The convergence rate of the target estimator improves over target-only OLS whenever the sufficient informative domain contains much more data than the target and the contrast gap is favourable.
  • The averaged estimator beats Trans-Lasso in the low-dimensional linear setting because the sufficiency penalty avoids the rate loss caused by q-aggregation when informative sources are abundant.
  • Inference is feasible without knowing the transferable set: Theorem 2 gives asymptotic normality, so confidence intervals and tests can be built from the averaged estimator.
  • Privacy is preserved under individual-similarity because only local Gram matrices and coefficient vectors are transmitted, not raw data.

Reading between the lines

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

  • A practical diagnostic suggested by the geometry is to plot the estimated contrasts $\hat\delta_{(m)}$ in the target-covariance metric; the angular separation condition should be visually checkable, and violations signal that combinatorial-similarity methods (Trans-MAC) are safer than individual-similarity ones.
  • The sufficiency penalty coefficient $\phi$ plays the role of a model-selection temperature: the paper recommends $\log n_0$, but an editor's guess is that data-driven selection of $\phi$ via cross-validation could extend the method's finite-sample performance without breaking the theory.
  • The framework appears extendable to nonlinear, semiparametric, and nonparametric targets if the candidate estimators admit the same kind of unbiased risk estimate, though the paper does not prove those extensions.
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 / 5 minor

Summary. The paper proposes a sufficiency-principled transfer learning framework based on model averaging for multi-source linear regression when the set of transferable sources is unknown. It constructs nested candidate domains by ordering estimated contrast vectors, then selects weights by minimizing a criterion that combines prediction loss, a model-selection term, and a sufficiency penalty (Trans-MAI). The framework is extended to combinatorial similarity via Trans-MACs and Trans-MAC, which average over non-informative domains relative to a selected sufficient informative domain. The main theoretical claims are weight convergence for non-informative and insufficient informative domains, an O_p convergence rate for estimation and prediction that can be faster than target-only OLS and, under extra conditions, faster than Trans-Lasso, asymptotic normality, high-probability optimality, and asymptotic optimality under misspecification. The paper also reports extensive simulations and a Beijing housing rental data analysis.

Significance. If the theoretical results hold, this is a substantial contribution to transfer learning: it explicitly addresses the sufficiency principle, unifies individual and combinatorial similarity within one model-averaging framework, provides feasible asymptotic inference without a known transferable set, and supports the methods with extensive numerical evidence. The geometric interpretation of the key angular-separation condition and the explicit finite-sample simulations are valuable. However, the paper's central rate theorems appear to be stated for an oracle version of the algorithm, because the data-dependent construction of candidate domains is not formally incorporated into the main conditions; this gap is load-bearing and must be fixed before the advertised guarantees apply to the implemented procedure.

major comments (4)
  1. [Section 2.2, Eq. (2.1); Section 3.1.1, Lemma 1 and Theorem 1] The candidate domains M_m in Eq. (2.1) are constructed from estimated contrasts \hat\delta(k), which depend on the same response data used later in the weight criterion, but Lemmas 1 and 2 and Theorem 1 treat \delta[m] and A_h as fixed quasi-true objects associated with these domains. Proposition 2 gives a separation event under conditions (\eta>2h and p^{1/2}n^{-1/2}\eta^{-1}\to 0) that are not listed among the hypotheses of Lemma 1, Lemma 2, or Theorem 1. The paper must either include this event in the hypotheses and show that its probability tends to one, or directly bound the contribution of its complement; otherwise the rates in Theorems 1, 5, and 8 are established only for an oracle version with fixed candidate domains.
  2. [Section 2.3, Eq. (2.2)] The unbiasedness identity E[\hat L_v(w)|X[M]] = R_v(w)+... is derived by conditioning on the design matrices, but I_m in Eq. (2.1) is random and depends on y(0) and the source responses through \hat\delta. Conditioning on X[M] alone does not fix the candidate set, and the selection event is correlated with the target residuals appearing in \hat L_v(w). The derivation must condition on the ordering event from Proposition 2 or use sample splitting; as written, the criterion is not shown to be unbiased for R_v(w).
  3. [Section 2.4; Section 4.1] Trans-MACs and Trans-MAC are defined through the unknown threshold h and the set A_c^h in criteria P(\rho) and Q(\varphi), and Theorems 5-10 assume this partition is available. The paper does not provide a data-driven construction of A_c^h for the implemented algorithm: Section 4.1 selects only m_s via Trans-MAI and leaves the construction of A_c^h unspecified. Unless the numerical implementation uses the true h from the simulation design, the reported empirical results may not correspond to the estimator analyzed in the theorems, and the practical algorithm is underspecified.
  4. [Supplement A] All proofs of the central results are deferred to Supplement A, which is not part of the arXiv text, so the ordering-event and unbiasedness issues above could not be checked against a proof. The authors should make the supplement available to reviewers and explicitly cross-reference the proof of Proposition 2 in the statements of Lemma 1 and Theorem 1.
minor comments (5)
  1. [Throughout] There are numerous typos and repeated misspellings, including "transfering", "Forth" for "Fourth", "convenience rate" for "convergence rate", "repectively", "seceario", and "as meassured"; the paper needs a careful proofreading pass.
  2. [Section 2.3, Eq. (2.3) and Section 2.4, Eq. (2.6)] The symbol \varphi is used both for the Trans-MAI penalty parameter in Eq. (2.3) and for the weight vector in Eq. (2.6) and in Section 3.2.2; this dual use makes the theoretical statements unnecessarily confusing.
  3. [Figures 3-4] The captions state that OLS-Pool and Trans-MACs "behave off-axis"; if the curves exceed the plotting range, the figures should be rescaled or the off-axis values should be reported numerically, otherwise the visual comparisons are incomplete.
  4. [Section 4.2, Figure 9] The fitted dashed curves are described as being of the form c n^{-a}, but the v=0 curve is fit with a positive exponent (0.21), which is inconsistent with the stated functional form; this should be clarified or the fitting form should be stated more carefully.
  5. [Condition 5 and Figure 9] Condition 5 is not verifiable from data, and the paper's own Figure 9 shows that weight convergence for v=0 fails when the condition is violated; since the paper recommends v=0.5 in practice, a short discussion of the practical role of Condition 5 would help readers.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identified: the main rates follow from the stated conditions and the explicit criterion; the data-dependent candidate-domain issue is a proof gap, not a circular reduction.

full rationale

I walked the derivation chain for Trans-MAI, Trans-MACs, and Trans-MAC. Lemma 1, Lemma 2, Theorem 1, Theorem 5, and Theorem 8 are stated under explicit conditions (Conditions 1-6, 12-13, 18, etc.) and are proved from the explicit criterion C_{v,phi} and standard concentration arguments. No parameter is fitted to the reported target outcome and then renamed a prediction. The sufficiency penalty, term (iii) of Eq. (2.3), is a deliberately designed mechanism: with phi going to infinity and Condition 6, weights concentrate on the large-sample informative domain. That concentration is a substantive result about the minimizer of the criterion rather than a definitional identity. The paper's references to the authors' own prior work, such as the remark connecting Lemma 2 to Zhang and Liu (2019, Theorem 5) and the model-averaging context of Zhang and Liu (2023) and Zhang et al. (2024), are contextual or comparative and are not the load-bearing justification of the central rates. The most notable weakness is that the main theorems are stated for candidate domains whose construction in Eq. (2.1) uses the same response data through the estimated contrasts, while the unbiasedness identity (2.2) and subsequent proofs condition on X[M]. Proposition 2 is a separate discussion of ordering separation and is not incorporated into the conditions of Lemma 1 or Theorem 1. This is a possible technical gap in the proof chain, but it is not circularity: the stated rates are not equivalent to the assumptions by construction, and they are not fitted inputs presented as predictions.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The paper tunes v and phi; no parameters are fit from experimental outcomes. The main assumptions are regularity conditions and explicit similarity/separability assumptions. No new physical or model entities are introduced.

free parameters (2)
  • v (weight-balance tuning) = 0.5 in simulations; v in (0,1] in theory
    Balances the loss of the aggregated predictor against the weighted loss of candidates; Lemma 1's required conditions change with v.
  • phi (sufficiency penalty) = log n0 recommended; phi to infinity required
    Controls the trace-based penalty; Lemma 2 and Prop. 1 need phi to infinity while Theorem 3's bound depends on |phi-2|.
assumptions (5)
  • standard math Standard regularity: bounded eigenvalues (Condition 1(i)), sub-Gaussian or bounded moments (Conditions 1(iii), 7, 15), consistent variance estimators (Conditions 3, 12)
    Uniform eigenvalue bounds, moment conditions, and sub-Gaussian tails are invoked throughout Section 3 for concentration and CLT.
  • domain assumption Linear model correct specification for the main convergence and normality theorems; separate misspecified theory used for optimality
    Theorems 1-3 assume candidate models are correctly specified; the misspecified case is only treated for asymptotic optimality (Theorem 4).
  • domain assumption Angular separation of non-informative contrasts (Conditions 4 and 5)
    Lemma 1 needs pairwise positive Sigma0-weighted inner products among non-informative delta[m] so that no convex combination of them mimics the target.
  • domain assumption Sample-size gap for the sufficient informative domain (Condition 6)
    Lemma 2 uses Condition 6 to bound trace differences so weights concentrate on the largest informative domain.
  • domain assumption Separability of estimated contrast orderings (Proposition 2: eta > 2h and p^1/2 n^-1/2 eta^-1 to 0)
    The candidate domain construction is consistent only if estimated contrasts separate informative from non-informative sources.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sufficiency-principled Transfer Learning via Model Averaging." pith.science (2026). https://pith.science/paper/WDD5LBUJ

@misc{pith2026250715416,
  author       = {Pith},
  title        = {Pith review of: Sufficiency-principled Transfer Learning via Model Averaging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WDD5LBUJ}},
  note         = {Machine review of arXiv:2507.15416}
}
read the original abstract

When the transferable set is unknowable, transfering informative knowledge as much as possible\textemdash a principle we refer to as \emph{sufficiency}, becomes crucial for enhancing transfer learning effectiveness. However, existing transfer learning methods not only overlook the sufficiency principle, but also rely on restrictive single-similarity assumptions (\eg individual or combinatorial similarity), leading to suboptimal performance. To address these limitations, we propose a sufficiency-principled transfer learning framework via unified model averaging algorithms, accommodating both individual and combinatorial similarities. Theoretically, we establish the asymptotic/high-probability optimality, enhanced convergence rate and asymptotic normality for multi-source linear regression models with a diverging number of parameters, achieving sufficiency, robustness to negative transfer, privacy protection and feasible statistical inference. Extensive simulations and an empirical data analysis of Beijing housing rental data demonstrate the promising superiority of our framework over conventional alternatives.

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. Representation-Aware Distributionally Robust Optimization: A Knowledge Transfer Framework

    stat.ME 2025-09 conditional novelty 6.0 of 10

    A representation-aware Wasserstein DRO framework that shrinks estimators toward an external representation subspace, with asymptotic inference and adaptive robustness tuning.

Reference graph

Works this paper leans on

77 extracted references · 68 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION format.url url empty "" url if FUNCTION article output.bibitem format.authors "author" output.check author format.key output output.year.check new.block format.title "title" output.check new.block crossref missing format.jour.vol output format.article.crossref output.nonnull format.pages output if ne...

  2. [2]

    Li , Ker-Chau K.-C

    barticle [author] Ando , Tomohiro T. Li , Ker-Chau K.-C. ( 2014 ). A model-averaging approach for high-dimensional regression . Journal of the American Statistical Association 109 254--265 . barticle

  3. [3]

    , Cai , T Tony T

    barticle [author] Auddy , Arnab A. , Cai , T Tony T. T. Chakraborty , Abhinav A. ( 2024 ). Minimax and adaptive transfer learning for nonparametric classification under distributed differential privacy constraints . arXiv preprint arXiv:2406.20088 . barticle

  4. [4]

    , Sun , Yiming Y

    barticle [author] Bao , Runxue R. , Sun , Yiming Y. , Gao , Yuhe Y. , Wang , Jindong J. , Yang , Qiang Q. , Mao , Zhi-Hong Z.-H. Ye , Ye Y. ( 2023 ). A recent survey of heterogeneous transfer learning . arXiv preprint arXiv:2310.08459 . barticle

  5. [5]

    ( 2021 )

    barticle [author] Bastani , Hamsa H. ( 2021 ). Predicting with proxies: transfer learning in high dimension . Management Science 67 2964--2984 . barticle

  6. [6]

    barticle [author] Bellec , Pierre C P. C. ( 2018 ). Optimal bounds for aggregation of affine estimators . The Annals of Statistics 46 30--59 . barticle

  7. [7]

    barticle [author] Cai , T Tony T. T. Pu , Hongming H. ( 2024 ). Transfer learning for nonparametric regression: Non-asymptotic minimax analysis and adaptive procedure . arXiv preprint arXiv:2401.12272 . barticle

  8. [8]

    barticle [author] Cai , T Tony T. T. Wei , Hongji H. ( 2021 ). Transfer learning for nonparametric classification . The Annals of Statistics 49 100--128 . barticle

Show all 77 references
  1. [9]

    barticle [author] Campbell , John Y J. Y. , Lo , Andrew W A. W. , MacKinlay , A Craig A. C. Whitelaw , Robert F R. F. ( 1998 ). The econometrics of financial markets . Macroeconomic Dynamics 2 559--562 . barticle

  2. [10]

    barticle [author] Casson , Robert J R. J. Farmer , Lachlan DM L. D. ( 2014 ). Understanding and checking the assumptions of linear regression: A primer for medical researchers . Clinical and Experimental Ophthalmology 42 590--596 . barticle

  3. [11]

    , Dong , Guozhu G

    barticle [author] Chen , Yixin Y. , Dong , Guozhu G. , Han , Jiawei J. , Pei , Jian J. , Wah , Benjamin W B. W. Wang , Jianyong J. ( 2006 ). Regression cubes with lossless compression and aggregation . IEEE Transactions on Knowledge and Data Engineering 18 1585--1599 . barticle

  4. [12]

    Hjort , Nils Lid N

    bbook [author] Claeskens , Gerda G. Hjort , Nils Lid N. L. ( 2008 ). Model Selection and Model Averaging . Cambridge University Press , Cambridge, MA . bbook

  5. [13]

    , Rigollet , Philippe P

    barticle [author] Dai , Dong D. , Rigollet , Philippe P. Zhang , Tong T. ( 2012 ). Deviation optimal learning using greedy q-aggregation . The Annals of Statistics 40 1878--1905 . barticle

  6. [14]

    Ameer , PM P

    barticle [author] Deepak , S S. Ameer , PM P. ( 2019 ). Brain tumor classification using deep CNN features via transfer learning . Computers in Biology and Medicine 111 103345 . barticle

  7. [15]

    , Pelger , Markus M

    barticle [author] Duan , Junting J. , Pelger , Markus M. Xiong , Ruoxuan R. ( 2024 ). Target PCA: Transfer learning large dimensional panel data . Journal of Econometrics 244 105521 . barticle

  8. [16]

    Peng , Heng H

    barticle [author] Fan , Jianqing J. Peng , Heng H. ( 2004 ). Nonconcave penalized likelihood with a diverging number of parameters . The Annals of Statistics 32 928--961 . barticle

  9. [17]

    ( 2015 )

    bbook [author] Fox , John J. ( 2015 ). Applied regression analysis and generalized linear models . Sage Publications , Thousand Oaks, CA . bbook

  10. [18]

    Lecu \'e , Guillaume G

    barticle [author] Ga \^ ffas , St \'e phane S. Lecu \'e , Guillaume G. ( 2011 ). Hyper-sparse optimal aggregation . Journal of Machine Learning Research 12 1813--1833 . barticle

  11. [19]

    , Liu , Yang Y

    binproceedings [author] Gao , Dashan D. , Liu , Yang Y. , Huang , Anbu A. , Ju , Ce C. , Yu , Han H. Yang , Qiang Q. ( 2019 ). Privacy-preserving heterogeneous federated transfer learning . In 2019 IEEE International Conference on Big Data 2552--2559 . IEEE , Los Angeles, CA ....

  12. [20]

    Hill , Jennifer J

    bbook [author] Gelman , Andrew A. Hill , Jennifer J. ( 2007 ). Data Analysis Using Regression and Multilevel/Hierarchical Models . Cambridge University Press , Cambridge, MA . bbook

  13. [21]

    , Lederer , Johannes J

    barticle [author] Gold , David D. , Lederer , Johannes J. Tao , Jing J. ( 2020 ). Inference for high-dimensional instrumental variables regression . Journal of Econometrics 217 79--111 . barticle

  14. [22]

    , Khaitan , Siddhartha K S

    barticle [author] Gopalakrishnan , Kasthurirangan K. , Khaitan , Siddhartha K S. K. , Choudhary , Alok A. Agrawal , Ankit A. ( 2017 ). Deep convolutional neural networks with transfer learning for computer vision-based data-driven pavement distress detection . Construction and...

  15. [23]

    , Han , Yi Y

    barticle [author] Gu , Tian T. , Han , Yi Y. Duan , Rui R. ( 2024 ). Robust angle-based transfer learning in high dimensions . Journal of the Royal Statistical Society Series B: Statistical Methodology 00 1--23 . barticle

  16. [24]

    barticle [author] Hansen , Bruce E B. E. ( 2007 ). Least squares model averaging . Econometrica 75 1175--1189 . barticle

  17. [25]

    barticle [author] Hansen , Bruce E B. E. Racine , Jeffrey S J. S. ( 2012 ). Jackknife model averaging . Journal of Econometrics 167 38--46 . barticle

  18. [26]

    , Tibshirani , Robert R

    bbook [author] Hastie , Trevor T. , Tibshirani , Robert R. , Friedman , Jerome H J. H. Friedman , Jerome H J. H. ( 2009 ). The Elements of Statistical Learning: Data mining, Inference, and Prediction . Springer , Berlin . bbook

  19. [27]

    barticle [author] Hjort , Nils Lid N. L. Claeskens , Gerda G. ( 2003 ). Frequentist model average estimators . Journal of the American Statistical Association 98 879--899 . barticle

  20. [28]

    Zhang , Xinyu X

    barticle [author] Hu , Xiaonan X. Zhang , Xinyu X. ( 2023 ). Optimal parameter-transfer learning by semiparametric model averaging . Journal of Machine Learning Research 24 1--53 . barticle

  21. [29]

    , Wang , Mingqiu M

    barticle [author] Huang , Jiayu J. , Wang , Mingqiu M. Wu , Yuanshan Y. ( 2022 ). Transfer learning with high-dimensional quantile regression . arXiv preprint arXiv:2211.14578 . barticle

  22. [30]

    Rigollet , Philippe P

    barticle [author] Lecu \'e , Guillaume G. Rigollet , Philippe P. ( 2014 ). Optimal learning with q-aggregation . The Annals of Statistics 42 211--224 . barticle

  23. [31]

    , Cai , T Tony T

    barticle [author] Li , Sai S. , Cai , T Tony T. T. Li , Hongzhe H. ( 2022 ). Transfer learning for high-dimensional linear regression: Prediction, estimation and minimax optimality . Journal of the Royal Statistical Society Series B: Statistical Methodology 84 149--173 . barticle

  24. [32]

    , Cai , T Tony T

    barticle [author] Li , Sai S. , Cai , T Tony T. T. Li , Hongzhe H. ( 2023 ). Transfer learning in large-scale gaussian graphical models with false discovery rate control . Journal of the American Statistical Association 118 2171--2183 . barticle

  25. [33]

    , Zhang , Linjun L

    barticle [author] Li , Sai S. , Zhang , Linjun L. , Cai , T Tony T. T. Li , Hongzhe H. ( 2024 ). Estimation and inference for high-dimensional generalized linear models with knowledge transfer . Journal of the American Statistical Association 119 1274--1285 . barticle

  26. [34]

    Reimherr , Matthew M

    barticle [author] Lin , Haotian H. Reimherr , Matthew M. ( 2022 ). On transfer learning in functional linear regression . arXiv preprint arXiv:2206.04277 . barticle

  27. [35]

    Xi , Ruibin R

    barticle [author] Lin , Nan N. Xi , Ruibin R. ( 2011 ). Aggregated estimating equation estimation . Statistics and its Interface 4 73--83 . barticle

  28. [36]

    , Zhao , Junlong J

    barticle [author] Lin , Ziqian Z. , Zhao , Junlong J. , Wang , Fang F. Wang , Hansheng H. ( 2024 ). Profiled transfer learning for high dimensional linear model . arXiv preprint arXiv:2406.00701 . barticle

  29. [37]

    ( 2015 )

    barticle [author] Liu , Chu-An C.-A. ( 2015 ). Distribution theory of the least squares averaging estimator . Journal of Econometrics 186 142--159 . barticle

  30. [38]

    , Yao , Qingsong Q

    barticle [author] Liu , Qingfeng Q. , Yao , Qingsong Q. Zhao , Guoqing G. ( 2020 ). Model averaging estimation for conditional volatility models with an application to stock market volatility forecast . Journal of Forecasting 39 841--863 . barticle

  31. [39]

    Abdulazeez , Adnan M A

    barticle [author] Maulud , Dastan D. Abdulazeez , Adnan M A. M. ( 2020 ). A review on linear regression comprehensive in machine learning . Journal of Applied Science and Technology Trends 1 140--147 . barticle

  32. [40]

    ( 2015 )

    barticle [author] Moral-Benito , Enrique E. ( 2015 ). Model averaging in economics: An overview . Journal of Economic Surveys 29 46--75 . barticle

  33. [41]

    bbook [author] Murphy , Kevin P K. P. ( 2022 ). Probabilistic machine learning: An introduction . MIT Press , Cambridge, MA . bbook

  34. [42]

    bbook [author] Olivas , Emilio Soria E. S. , Guerrero , Jos \'e David Mart \' n J. D. M. , Martinez-Sober , Marcelino M. , Magdalena-Benedito , Jose Rafael J. R. Serrano L \'o pez , Antonio Jos \'e A. J. ( 2009 ). Handbook of Research on Machine Learning Applications and Trend...

  35. [43]

    barticle [author] Pan , Sinno Jialin S. J. Yang , Qiang Q. ( 2009 ). A survey on transfer learning . IEEE Transactions on Knowledge and Data Engineering 22 1345--1359 . barticle

  36. [44]

    Yang , Yuhong Y

    barticle [author] Peng , Jingfu J. Yang , Yuhong Y. ( 2022 ). On improvability of model selection by model averaging . Journal of Econometrics 229 246--262 . barticle

  37. [45]

    ( 1984 )

    barticle [author] Portnoy , Stephen S. ( 1984 ). Asymptotic behavior of M-estimators of p regression parameters when p^ 2 /n is large. I. consistency . The Annals of Statistics 12 1298--1309 . barticle

  38. [46]

    barticle [author] Raftery , Adrian E A. E. , Madigan , David D. Hoeting , Jennifer A J. A. ( 1997 ). Bayesian model averaging for linear regression models . Journal of the American Statistical Association 92 179--191 . barticle

  39. [47]

    ( 2012 )

    barticle [author] Rigollet , Philippe P. ( 2012 ). Kullback--Leibler aggregation and misspecified generalized linear models . The Annals of Statistics 40 639--665 . barticle

  40. [48]

    , Peters , Matthew E M

    binproceedings [author] Ruder , Sebastian S. , Peters , Matthew E M. E. , Swayamdipta , Swabha S. Wolf , Thomas T. ( 2019 ). Transfer learning in natural language processing . In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computatio...

  41. [49]

    ( 1970 )

    barticle [author] Ruhe , Axel A. ( 1970 ). Perturbation bounds for means of eigenvalues and invariant subspaces . BIT Numerical Mathematics 10 343--354 . barticle

  42. [50]

    barticle [author] Schifano , Elizabeth D E. D. , Wu , Jing J. , Wang , Chun C. , Yan , Jun J. Chen , Ming-Hui M.-H. ( 2016 ). Online updating of statistical inference in the big data setting . Technometrics 58 393--403 . barticle

  43. [51]

    Vetter , Thomas R T

    barticle [author] Schober , Patrick P. Vetter , Thomas R T. R. ( 2021 ). Linear regression in medical research . Anesthesia & Analgesia 132 108--109 . barticle

  44. [52]

    barticle [author] Steel , Mark FJ M. F. ( 2020 ). Model averaging and its use in economics . Journal of Economic Literature 58 644--719 . barticle

  45. [53]

    barticle [author] Theodoris , Christina V C. V. , Xiao , Ling L. , Chopra , Anant A. , Chaffin , Mark D M. D. , Al Sayed , Zeina R Z. R. , Hill , Matthew C M. C. , Mantineo , Helene H. , Brydon , Elizabeth M E. M. , Zeng , Zexian Z. , Liu , X Shirley X. S. Ellinor , Patrick T ...

  46. [54]

    Feng , Yang Y

    barticle [author] Tian , Ye Y. Feng , Yang Y. ( 2023 ). Transfer learning under high-dimensional generalized linear models . Journal of the American Statistical Association 118 2684--2697 . barticle

  47. [55]

    , Jin , Chi C

    binproceedings [author] Tripuraneni , Nilesh N. , Jin , Chi C. Jordan , Michael M. ( 2021 ). Provable meta-learning of linear representations . In Proceedings of the 38th International Conference on Machine Learning 10434--10443 . PMLR , Virtual Event . binproceedings

  48. [56]

    , Yang , Zhongqi Z

    binproceedings [author] Wang , Ziyu Z. , Yang , Zhongqi Z. , Azimi , Iman I. Rahmani , Amir M A. M. ( 2024 ). Differential private federated transfer learning for mental health monitoring in everyday settings: A case study on stress detection . In 2024 46th Annual Internationa...

  49. [57]

    , Khoshgoftaar , Taghi M T

    barticle [author] Weiss , Karl K. , Khoshgoftaar , Taghi M T. M. Wang , DingDing D. ( 2016 ). A survey of transfer learning . Journal of Big Data 3 1--40 . barticle

  50. [58]

    bbook [author] Wooldridge , Jeffrey M J. M. ( 2016 ). Introductory Econometrics: A Modern Approach . Cengage Learning , Boston, MA . bbook

  51. [59]

    , Guo , Zijian Z

    barticle [author] Xiong , Xin X. , Guo , Zijian Z. Cai , Tianxi T. ( 2023 ). Distributionally robust transfer learning . arXiv preprint arXiv:2309.06534 . barticle

  52. [60]

    Zhang , Xinyu X

    barticle [author] Xu , Wenchao W. Zhang , Xinyu X. ( 2022 ). From model selection to model averaging: A comparison for nested linear models . Econometric Theory 0 1--33 . barticle

  53. [61]

    ( 2003 )

    barticle [author] Yang , Yuhong Y. ( 2003 ). Regression with multiple candidate models: Selecting or mixing? Statistica Sinica 13 783--809 . barticle

  54. [62]

    Yang , Yuhong Y

    barticle [author] Yuan , Zheng Z. Yang , Yuhong Y. ( 2005 ). Combining linear regression models: when and how? Journal of the American Statistical Association 100 1202--1214 . barticle

  55. [63]

    , Zhong , Wei W

    barticle [author] Zeng , Hao H. , Zhong , Wei W. Xu , Xingbai X. ( 2024 ). Transfer learning for spatial autoregressive models with application to US presidential election prediction . arXiv preprint arXiv:2405.15600 . barticle

  56. [64]

    Li , Xiang X

    barticle [author] Zhang , Wei W. Li , Xiang X. ( 2022 ). Data privacy preserving federated transfer learning in machinery fault diagnostics using prior distributions . Structural Health Monitoring 21 1329--1344 . barticle

  57. [65]

    Liang , Hua H

    barticle [author] Zhang , Xinyu X. Liang , Hua H. ( 2011 ). Focused information criterion and model averaging for generalized additive partial linear models . The Annals of Statistics 39 174--200 . barticle

  58. [66]

    Liu , Chu-An C.-A

    barticle [author] Zhang , Xinyu X. Liu , Chu-An C.-A. ( 2019 ). Inference after model averaging in linear regression models . Econometric Theory 35 816--841 . barticle

  59. [67]

    Liu , Chu-An C.-A

    barticle [author] Zhang , Xinyu X. Liu , Chu-An C.-A. ( 2023 ). Model averaging prediction by K-fold cross-validation . Journal of Econometrics 235 280--301 . barticle

  60. [68]

    , Xue , Songkai S

    binproceedings [author] Zhang , Xuefei X. , Xue , Songkai S. Zhu , Ji J. ( 2020 ). A flexible latent space model for multilayer networks . In Proceedings of the 37th International Conference on Machine Learning 11288--11297 . PMLR , Virtual Event . binproceedings

  61. [69]

    Zhu , Zhongyi Z

    barticle [author] Zhang , Yijiao Y. Zhu , Zhongyi Z. ( 2025 ). Transfer learning for high-dimensional quantile regression via convolution smoothing . Statistica Sinica 35 1--20 . barticle

  62. [70]

    , Zou , Guohua G

    barticle [author] Zhang , Xinyu X. , Zou , Guohua G. , Liang , Hua H. Carroll , Raymond J R. J. ( 2020 ). Parsimonious model averaging with a diverging number of parameters . Journal of the American Statistical Association 115 972--984 . barticle

  63. [71]

    , Liu , Huihang H

    barticle [author] Zhang , Xinyu X. , Liu , Huihang H. , Wei , Yizheng Y. Ma , Yanyuan Y. ( 2024 ). Prediction using many samples with models possibly containing partially shared parameters . Journal of Business & Economic Statistics 42 187--196 . barticle

  64. [72]

    , Pan , Sinno Jialin S

    barticle [author] Zhao , Lili L. , Pan , Sinno Jialin S. J. Yang , Qiang Q. ( 2017 ). A unified framework of active transfer learning for cross-system recommendation . Artificial Intelligence 245 38--55 . barticle

  65. [73]

    , Zheng , Shengbin S

    barticle [author] Zhao , Junlong J. , Zheng , Shengbin S. Leng , Chenlei C. ( 2023 ). Residual importance weighted transfer learning for high-dimensional linear regression . arXiv preprint arXiv:2311.07972 . barticle

  66. [74]

    ( 2018 )

    barticle [author] Zhu , Ying Y. ( 2018 ). Sparse linear models and _1 -regularized 2SLS with high-dimensional endogenous regressors and instruments . Journal of Econometrics 202 196--213 . barticle

  67. [75]

    , Zhang , Xinyu X

    barticle [author] Zhu , Rong R. , Zhang , Xinyu X. , Wan , Alan TK A. T. Zou , Guohua G. ( 2022 ). Frequentist model averaging under inequality constraints . Journal of Statistical Planning and Inference 221 100--113 . barticle

  68. [76]

    , Qi , Zhiyuan Z

    barticle [author] Zhuang , Fuzhen F. , Qi , Zhiyuan Z. , Duan , Keyu K. , Xi , Dongbo D. , Zhu , Yongchun Y. , Zhu , Hengshu H. , Xiong , Hui H. He , Qing Q. ( 2020 ). A comprehensive survey on transfer learning . Proceedings of the IEEE 109 43--76 . barticle

  69. [77]

    Zhang , Hao Helen H

    barticle [author] Zou , Hui H. Zhang , Hao Helen H. H. ( 2009 ). On the adaptive elastic-net with a diverging number of parameters . The Annals of Statistics 37 1733--1751 . barticle

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.