arxiv: 2605.09471 · v1 · submitted 2026-05-10 · 🧮 math.ST · stat.TH

Recognition: no theorem link

The Statistical Cost of Adaptation in Multi-Source Transfer Learning

Abhinav Chakraborty, Subha Maity

Pith reviewed 2026-05-12 04:15 UTC · model grok-4.3

classification 🧮 math.ST stat.TH

keywords multi-source transfer learningintrinsic cost of adaptationphase transitionparametric estimationoracle riskbias-agnostic estimatorstatistical limitstransfer learning

0 comments

The pith

Multi-source transfer learning cannot always match oracle performance without knowing biases, even with two sources.

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

The paper defines the intrinsic cost of adaptation as the smallest possible worst-case ratio of risk for any bias-agnostic estimator to the risk of an oracle that knows the source-to-target biases. It establishes that this cost equals one in some regimes but exceeds one in others, showing that bias-agnostic adaptation fails to achieve oracle rates for certain bias configurations even when only two sources are available. For any fixed number of sources the cost undergoes a phase transition that cleanly separates the achievable and non-achievable regimes. The cost further increases as the number of sources grows, yet drops again when extra structure such as ordered biases or clustered parameters is imposed and estimators are tailored to that structure.

Core claim

The central claim is that the intrinsic cost of adaptation, defined as the infimum over all bias-agnostic estimators of the supremum over bias configurations of the ratio of their risk to the oracle risk, is strictly greater than one for some multi-source parametric problems. Even with two sources, the configuration space of fixed unknown biases can place the problem past a phase transition where no estimator achieves the oracle rate; the cost grows with the number of sources, while additional structure on the biases permits specially designed estimators that reduce the cost.

What carries the argument

The intrinsic cost of adaptation, which is the minimal worst-case ratio of the risk of a bias-agnostic estimator to the oracle risk over the space of possible source-to-target biases.

If this is right

For any fixed number of sources there exist bias configurations where oracle performance is attainable by a bias-agnostic estimator and others where it is not.
The adaptation cost grows as the number of sources increases.
When adaptation over the full bias space is impossible, imposing ordered biases, clustered source parameters, or sufficiently separated non-informative sources allows tailored estimators to achieve substantially lower cost.
Theoretical guarantees and empirical results support the existence of these lower-cost estimators under the added structure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The phase transition may be used to design practical tests that decide whether to employ a fully agnostic estimator or one that exploits suspected structure.
In applications with many sources the rising cost suggests that simple pooling strategies will increasingly underperform unless structure is exploited or bias information is collected.
The same worst-case ratio construction could be applied to non-parametric or high-dimensional estimation to obtain analogous limits.

Load-bearing premise

The source-to-target biases are fixed but unknown parameters whose configuration space admits a well-defined worst-case ratio in a correctly specified parametric model.

What would settle it

Compute the minimal worst-case risk ratio for a concrete two-source Gaussian location model whose bias vectors lie past the predicted phase-transition boundary; the ratio should exceed one.

Figures

Figures reproduced from arXiv: 2605.09471 by Abhinav Chakraborty, Subha Maity.

**Figure 2.** Figure 2: Comparison of estimators under the cluster configuration. [PITH_FULL_IMAGE:figures/full_fig_p029_2.png] view at source ↗

**Figure 3.** Figure 3: Comparison of estimators under Separation I (left) and Separation II (right) [PITH_FULL_IMAGE:figures/full_fig_p030_3.png] view at source ↗

read the original abstract

Multi-source transfer learning can improve target-domain estimation by leveraging related source data, but its benefits depend on unknown source-to-target biases. This raises a fundamental question: can a bias-agnostic estimator perform as well as an oracle that knows the true bias configuration? To study this, we introduce the intrinsic cost of adaptation, defined as the smallest worst-case ratio between the risk of any bias-agnostic estimator and the oracle risk. An intrinsic cost of one means oracle performance is achievable without knowing the biases, whereas a larger cost quantifies the unavoidable price of adaptation. Focusing on parametric estimation, we show that multi-source transfer behaves fundamentally differently from the single-source setting: adaptation is not always possible, even with only two sources. For a fixed number of sources, we characterize the intrinsic cost of adaptation and identify a phase transition separating regimes where oracle performance is achievable from those where it is not. As the number of sources grows, we further show that the adaptation cost increases. When adaptation over the full bias configuration space is impossible, additional structure can substantially reduce the cost. We study settings with ordered biases, clustered source parameters, and sufficiently separated non-informative sources, and propose estimators tailored to each regime, with supporting theoretical and empirical results. Overall, our results delineate the statistical limits of multi-source transfer, clarifying when oracle performance is attainable, when structural assumptions help, and when adaptation is fundamentally impossible.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper defines an intrinsic cost of adaptation for multi-source transfer and shows a phase transition where oracle performance fails even with two sources, unlike single-source cases.

read the letter

The core contribution is a clean definition of the intrinsic cost of adaptation: the smallest worst-case ratio between the risk of any bias-agnostic estimator and the oracle risk that knows the source-to-target biases. In parametric estimation they show this cost exceeds one even for two sources, and they map out the phase transition that separates regimes where adaptation is possible from those where it is not. They also prove that the cost grows with the number of sources under the worst-case bias configuration. That distinction from single-source transfer is the clearest new point. They handle the cases where full adaptation fails by adding structure—ordered biases, clustered parameters, or separated non-informative sources—and give tailored estimators with supporting bounds and simulations. The math stays inside the parametric fixed-bias model, which keeps the derivations tractable. The main limitation is that the worst-case ratio can feel pessimistic once you move outside the exact parametric setup or allow biases to vary with the data. The empirical checks are limited to the structured regimes they analyze, so they do not test how often the phase transition appears in typical data. Readers working on theoretical transfer learning or on designing estimators that must remain bias-agnostic will find the phase-transition characterization useful. The work is grounded enough and the claims are stated precisely enough that it should go to referees rather than a desk reject.

Referee Report

0 major / 3 minor

Summary. The manuscript introduces the intrinsic cost of adaptation in multi-source transfer learning, defined as the smallest worst-case ratio between the risk of any bias-agnostic estimator and the oracle risk that knows the source-to-target biases. Focusing on parametric estimation, it shows that multi-source transfer differs fundamentally from single-source: adaptation is not always possible even with two sources. For a fixed number of sources, the paper characterizes this cost and identifies a phase transition separating regimes where oracle performance is achievable from those where it is not. It further shows that the adaptation cost increases with the number of sources. When full adaptation over the bias space is impossible, the work examines structured regimes (ordered biases, clustered source parameters, sufficiently separated non-informative sources), proposes tailored estimators, and provides supporting theoretical and empirical results.

Significance. If the phase-transition characterization and cost bounds hold, this work makes a valuable contribution by delineating the statistical limits of bias-agnostic multi-source transfer learning. The explicit distinction from the single-source case, the identification of regimes where oracle performance is attainable without bias knowledge, and the analysis of how cost scales with the number of sources clarify fundamental trade-offs. The additional results on structured bias settings and corresponding estimators further strengthen the practical relevance by showing how modest assumptions can reduce the adaptation cost.

minor comments (3)

The definition of the intrinsic cost (as the infimum over estimators of the supremum risk ratio) is central; ensure the main text explicitly states the precise function classes and risk measures used in the parametric model to avoid any ambiguity in the worst-case quantification.
In the sections presenting the phase-transition results, include a brief discussion of how the transition thresholds depend on the dimension or other model parameters, as this would help readers assess the practical scope of the 'oracle achievable' regime.
For the empirical results supporting the structured-bias estimators, add a short paragraph on the simulation design (e.g., how bias configurations are sampled and how many Monte Carlo repetitions are used) to facilitate reproducibility.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work on the intrinsic cost of adaptation in multi-source transfer learning. The summary accurately captures the key distinctions from the single-source setting, the phase-transition characterization, and the analysis of structured bias regimes. We appreciate the recommendation for minor revision and will incorporate any editorial improvements in the revised manuscript.

Circularity Check

0 steps flagged

No significant circularity; definition and characterizations are self-contained

full rationale

The paper defines the intrinsic cost of adaptation explicitly as the smallest worst-case ratio of risks between any bias-agnostic estimator and the oracle risk over the bias configuration space. It then analyzes this quantity in parametric models to derive phase transitions and cost increases with the number of sources. These results follow from direct minimax analysis over the fixed but unknown biases rather than from any self-referential fitting, renaming of known patterns, or load-bearing self-citations that reduce the central claims to tautologies. The distinction from single-source transfer and the identification of regimes where adaptation is impossible emerge from the multi-source setup and worst-case ratio without circular reduction to the inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claims rest on the newly introduced definition of intrinsic cost and on the assumption of a parametric model with fixed unknown biases; no free parameters are explicitly fitted in the abstract, but the worst-case ratio implicitly depends on the bias configuration space.

axioms (1)

domain assumption Parametric estimation model with fixed unknown source-to-target biases
The analysis is restricted to parametric settings as stated; the intrinsic cost is defined only when biases are fixed but unknown.

invented entities (1)

Intrinsic cost of adaptation no independent evidence
purpose: Quantifies the unavoidable statistical price of bias-agnostic estimation versus oracle
Newly defined quantity whose value is characterized via phase transitions.

pith-pipeline@v0.9.0 · 5550 in / 1361 out tokens · 46754 ms · 2026-05-12T04:15:11.467961+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

297 extracted references · 297 canonical work pages · 4 internal anchors

[1]

C. K. Chow and C. N. Liu , title =. IEEE Transactions on Information Theory , year =

work page
[2]

1988 , address =

Judea Pearl , title =. 1988 , address =

work page 1988
[3]

2024 , journal =

Maity, Subha and Dutta, Diptavo and Terhorst, Jonathan and Sun, Yuekai and Banerjee, Moulinath , number =. 2024 , journal =. doi:10.1093/BIOMET/ASAD029 , issn =

work page doi:10.1093/biomet/asad029 2024
[4]

2022 , journal =

Hanneke, Steve and Kpotufe, Samory , number =. 2022 , journal =. doi:10.1214/22-AOS2189 , issn =

work page doi:10.1214/22-aos2189 2022
[5]

Tang, Minh and Athreya, Avanti and Sussman, Daniel L and Lyzinski, Vince and Priebe, Carey E , arxivId =

work page
[6]

and Hancock, Edwin R

Luo, Bin and Robles-Kelly, Antonio and Torsello, Andrea and Wilson, Richard C. and Hancock, Edwin R. , volume =. 2001 , journal =. doi:10.1109/CVPR.2001.990621 , issn =

work page doi:10.1109/cvpr.2001.990621 2001
[7]

doi:10.1056/NEJMOA1511939;PAGE:STRING:ARTICLE/CHAPTER , issn =

2015 , journal =. doi:10.1056/NEJMOA1511939;PAGE:STRING:ARTICLE/CHAPTER , issn =

work page doi:10.1056/nejmoa1511939;page:string:article/chapter 2015
[8]

Xu, F., Zhang, R.,

Weiss, Karl and Khoshgoftaar, Taghi M. and Wang, Ding Ding , number =. 2016 , journal =. doi:10.1186/S40537-016-0043-6 , issn =

work page doi:10.1186/s40537-016-0043-6 2016
[9]

IEEE Trans. Knowl. Data Eng , author =

work page
[10]

2017 , journal =

Greenewald, Kristjan and Tewari, Ambuj and Klasnja, Predrag and Murphy, Susan , month =. 2017 , journal =

work page 2017
[11]

and Cannings, Timothy I

Reeve, Henry W.J. and Cannings, Timothy I. and Samworth, Richard J. , number =. 2021 , journal =. doi:10.1214/21-AOS2102 , issn =

work page doi:10.1214/21-aos2102 2021
[12]

Ann. Statist. , author =

work page
[13]

2025 , journal =

Qin, Caihong and Xie, Jinhan and Li, Ting and Bai, Yang , number =. 2025 , journal =. doi:10.1080/01621459.2024.2403788 , issn =

work page doi:10.1080/01621459.2024.2403788 2025
[14]

2019 , journal =

Dzemski, Andreas , number =. 2019 , journal =. doi:10.1162/REST

work page doi:10.1162/rest 2019
[15]

Lattimore, Tor and Szepesv

work page
[16]

Waldron, Maja , arxivId =

work page
[17]

Chen, Li and Lin, D. Y. and Zeng, Donglin , number =. 2012 , journal =. doi:10.1093/BIOSTATISTICS/KXR017 , issn =

work page doi:10.1093/biostatistics/kxr017 2012
[18]

2025 , journal =

Yang, Jichen and Wang, Lei and Lian, Heng , number =. 2025 , journal =. doi:10.1007/S11222-025-10685-9 , issn =

work page doi:10.1007/s11222-025-10685-9 2025
[19]

2021 , journal =

Yan, Bowei and Sarkar, Purnamrita , number =. 2021 , journal =. doi:10.1080/01621459.2019.1706541 , issn =

work page doi:10.1080/01621459.2019.1706541 2021
[20]

Journal of the American Statistical Association , volume =

Niu, Yabo and Ni, Yang and Pati, Debdeep and Mallick, Bani K. , number =. 2024 , journal =. doi:10.1080/01621459.2023.2233744;REQUESTEDJOURNAL:JOURNAL:UASA20;WGROUP:STRING:PUBLICATION , issn =

work page doi:10.1080/01621459.2023.2233744;requestedjournal:journal:uasa20;wgroup:string:publication 2024
[21]

2023 , journal =

Xu, Shirong and Zhen, Yaoming and Wang, Junhui , number =. 2023 , journal =. doi:10.1080/07350015.2022.2085726 , issn =

work page doi:10.1080/07350015.2022.2085726 2023
[22]

and Vogelstein, J

Binkiewicz, N. and Vogelstein, J. T. and Rohe, K. , number =. 2017 , journal =. doi:10.1093/BIOMET/ASX008 , issn =

work page doi:10.1093/biomet/asx008 2017
[23]

2024 , journal =

Giessing, Alexander and Wang, Jingshen , number =. 2024 , journal =. doi:10.1093/JRSSSB/QKAD075 , issn =

work page doi:10.1093/jrsssb/qkad075 2024
[24]

2025 , journal =

Yang, Jichen and Wang, Lei and Lian, Heng , number =. 2025 , journal =. doi:10.1007/S11222-025-10607-9 , issn =

work page doi:10.1007/s11222-025-10607-9 2025
[25]

2022 , journal =

Zhao, Junlong and Liu, Xiumin and Wang, Hansheng and Leng, Chenlei , number =. 2022 , journal =. doi:10.1093/BIOMET/ASAB006 , issn =

work page doi:10.1093/biomet/asab006 2022
[26]

2024 , journal =

Wang, Ziyuan and Wang, Lei and Lian, Heng , number =. 2024 , journal =. doi:10.1111/SJOS.12723 , issn =

work page doi:10.1111/sjos.12723 2024
[27]

2025 , journal =

Zhou, Doudou and Liu, Molei and Li, Mengyan and Cai, Tianxi , number =. 2025 , journal =. doi:10.1080/01621459.2024.2356291 , issn =

work page doi:10.1080/01621459.2024.2356291 2025
[28]

2025 , journal =

Zhao, Pan and Josse, Julie and Yang, Shu , pages =. 2025 , journal =

work page 2025
[29]

2024 , journal =

Zhou, Xingcai and Zheng, Haotian and Zhang, Haoran and Huang, Chao , number =. 2024 , journal =. doi:10.1002/STA4.70004 , issn =

work page doi:10.1002/sta4.70004 2024
[30]

1996 , journal =

Huang, Jian , number =. 1996 , journal =. doi:10.1214/AOS/1032894452 , issn =

work page doi:10.1214/aos/1032894452 1996
[31]

J. Am. Statist. Assoc , author =

work page
[32]

and Hall, Georgina , number =

Abbe, Emmanuel and Bandeira, Afonso S. and Hall, Georgina , number =. 2016 , journal =. doi:10.1109/TIT.2015.2490670 , issn =

work page doi:10.1109/tit.2015.2490670 2016
[34]

2023 , journal =

Wang, Jiangzhou and Zhang, Jingfei and Liu, Binghui and Zhu, Ji and Guo, Jianhua , number =. 2023 , journal =. doi:10.1080/01621459.2021.1996378;REQUESTEDJOURNAL:JOURNAL:UASA20;WGROUP:STRING:PUBLICATION , issn =

work page doi:10.1080/01621459.2021.1996378;requestedjournal:journal:uasa20;wgroup:string:publication 2023
[35]

2025 , journal =

Chen, Chen and Xu, Dawei and Ding, Juan and Zhang, Junjian and Xiong, Wenjun , number =. 2025 , journal =. doi:10.1002/STA4.70041 , issn =

work page doi:10.1002/sta4.70041 2025
[36]

doi:10.1214/23-EJS2147 , arxivId =

2023 , author =. doi:10.1214/23-EJS2147 , arxivId =

work page doi:10.1214/23-ejs2147 2023
[37]

2022 , journal =

Tan, Kean Ming and Wang, Lan and Zhou, Wen Xin , number =. 2022 , journal =. doi:10.1111/RSSB.12485 , issn =

work page doi:10.1111/rssb.12485 2022
[38]

2021 , journal =

Arroyo, Jesús and Chen, Guodong and Priebe, Carey E and Vogelstein, Joshua T , pages =. 2021 , journal =

work page 2021
[39]

MacDonald, P. W. and Levina, E. and Zhu, J. , number =. 2022 , journal =. doi:10.1093/BIOMET/ASAB058 , issn =

work page doi:10.1093/biomet/asab058 2022
[40]

and Li, Andrew A

Farias, Vivek F. and Li, Andrew A. , number =. 2019 , journal =. doi:10.1287/MNSC.2018.3092 , issn =

work page doi:10.1287/mnsc.2018.3092 2019
[41]

2023 , journal =

Hanneke, Steve and Kpotufe, Samory and Mahdaviyeh, Yasaman and Neu, Gergely and Rosasco, Lorenzo , pages =. 2023 , journal =

work page 2023
[42]

2021 , journal =

Kpotufe, Samory and Martinet, Guillaume , number =. 2021 , journal =. doi:10.1214/21-AOS2084 , issn =

work page doi:10.1214/21-aos2084 2021
[43]

Zeng, Donglin and Mao, Lu and Lin, D. Y. , number =. 2016 , journal =. doi:10.1093/BIOMET/ASW013 , issn =

work page doi:10.1093/biomet/asw013 2016
[44]

, number =

Gentleman, Robert and Geyer, Charles J. , number =. 1994 , journal =. doi:10.1093/BIOMET/81.3.618 , issn =

work page doi:10.1093/biomet/81.3.618 1994
[45]

2022 , journal =

Maity, Subha and Sun, Yuekai and Banerjee, Moulinath , number =. 2022 , journal =

work page 2022
[46]

2022 , journal =

Maity, Subha and Sun, Yuekai and Banerjee, Moulinath , pages =. 2022 , journal =

work page 2022
[47]

2025 , journal =

Chen, Shuxiao and Li, Sai and Zhang, Bo and Ye, Ting , number =. 2025 , journal =. doi:10.1515/JCI-2024-0024/XML , issn =

work page doi:10.1515/jci-2024-0024/xml 2025
[48]

and Zhou, Harrison H

Zhang, Anderson Y. and Zhou, Harrison H. , number =. 2016 , journal =

work page 2016
[49]

2021 , journal =

Rozemberczki, Benedek and Allen, Carl and Sarkar, Rik , pages =. 2021 , journal =. doi:10.1093/comnet/xxx000 , arxivId =

work page doi:10.1093/comnet/xxx000 2021
[50]

2024 , journal =

Chernozhukov, Victor and Fern. 2024 , journal =. doi:10.1016/j.jeconom.2020.08.009 , issn =

work page doi:10.1016/j.jeconom.2020.08.009 2024
[51]

and Wang, W

Hu, Y. and Wang, W. , number =. 2024 , journal =. doi:10.1093/BIOMET/ASAE011 , issn =

work page doi:10.1093/biomet/asae011 2024
[52]

2021 , journal =

Chen, Mingli and Fern. 2021 , journal =. doi:10.1016/J.JECONOM.2020.04.004 , issn =

work page doi:10.1016/j.jeconom.2020.04.004 2021
[53]

2020 , journal =

Gao, Wayne Yuan , number =. 2020 , journal =. doi:10.1016/j.jeconom.2019.09.005 , issn =

work page doi:10.1016/j.jeconom.2019.09.005 2020
[54]

2025 , journal =

Zheng, Cheng and Dasgupta, Sayan and Xie, Yuxiang and Haris, Asad and Chen, Ying Qing , number =. 2025 , journal =. doi:10.3390/math13030441 , issn =

work page doi:10.3390/math13030441 2025
[55]

2019 , journal =

Hanneke, Steve and Kpotufe, Samory , volume =. 2019 , journal =

work page 2019
[56]

Li, Sai and Zhang, Linjun , arxivId =

work page
[57]

2020 , journal =

Bastani, Hamsa , number =. 2020 , journal =. doi:10.1287/MNSC.2020.3729 , issn =

work page doi:10.1287/mnsc.2020.3729 2020
[58]

Sci , author =

Manag. Sci , author =

work page
[59]

and Chatterjee, Snigdhansu , publisher =

Chandna, Swati and Bagozzi, Benjamin E. and Chatterjee, Snigdhansu , publisher =. 2025 , journal =. doi:10.1109/TNSE.2025.3598705 , issn =

work page doi:10.1109/tnse.2025.3598705 2025
[61]

2020 , journal =

Zhou, Zhixin and Li, Ping , pages =. 2020 , journal =. doi:10.1214/20-EJS1686 , issn =

work page doi:10.1214/20-ejs1686 2020
[62]

Cox, D. R. , number =. 1972 , journal =. doi:10.1111/J.2517-6161.1972.TB00899.X , issn =

work page doi:10.1111/j.2517-6161.1972.tb00899.x 1972
[63]

, number =

Fan, Jianqing and Gao, Cheng and Klusowski, Jason M. , number =. 2025 , journal =. doi:10.1214/25-AOS2534 , issn =

work page doi:10.1214/25-aos2534 2025
[64]

2025 , journal =

Cai, Tianxi and Li, Mengyan and Liu, Molei , number =. 2025 , journal =. doi:10.1080/01621459.2024.2393463 , issn =

work page doi:10.1080/01621459.2024.2393463 2025
[65]

2018 , journal =

Jochmans, Koen , number =. 2018 , journal =. doi:10.1080/07350015.2017.1286242;CTYPE:STRING:JOURNAL , issn =

work page doi:10.1080/07350015.2017.1286242;ctype:string:journal 2018
[66]

2002 , journal =

Chen, Kani and Jin, Zhezhen and Ying, Zhiliang , number =. 2002 , journal =. doi:10.1093/BIOMET/89.3.659 , issn =

work page doi:10.1093/biomet/89.3.659 2002
[67]

2025 , journal =

Mao, Guangcai and Yang, Shu and Wang, Xiaofei , number =. 2025 , journal =. doi:10.1093/BIOMTC/UJAF131 , issn =

work page doi:10.1093/biomtc/ujaf131 2025
[68]

and Leng, Chenlei , number =

Yan, Ting and Jiang, Binyan and Fienberg, Stephen E. and Leng, Chenlei , number =. 2019 , journal =. doi:10.1080/01621459.2018.1448829 , issn =

work page doi:10.1080/01621459.2018.1448829 2019
[69]

2024 , journal =

Xie, Wenyi and Zeng, Donglin and Wang, Yuanjia , number =. 2024 , journal =. doi:10.1214/24-AOAS1875 , keywords =

work page doi:10.1214/24-aoas1875 2024
[70]

2024 , journal =

Duan, Junting and Pelger, Markus and Xiong, Ruoxuan , number =. 2024 , journal =. doi:10.1016/J.JECONOM.2023.105521 , issn =

work page doi:10.1016/j.jeconom.2023.105521 2024
[71]

2025 , journal =

Yao, Chengyuan and Cortez, Carmen and Yu, Renzhe , month =. 2025 , journal =. doi:10.1145/3706468.3706567 , arxivId =

work page doi:10.1145/3706468.3706567 2025
[72]

2023 , journal =

Wang, Yu Mei and Sun, Yuzhi and Wang, Beiying and Wu, Zhiping and He, Xiao Ying and Zhao, Yuansong , number =. 2023 , journal =. doi:10.1093/BIB/BBAD426 , issn =

work page doi:10.1093/bib/bbad426 2023
[73]

2025 , journal =

Wu, Jou Chin and Chen, Li Pang , number =. 2025 , journal =. doi:10.1002/SIM.70163 , issn =

work page doi:10.1002/sim.70163 2025
[74]

2026 , journal =

Zheng, Zejing and Zheng, Shengbing and Zhao, Junlong , month =. 2026 , journal =. doi:10.1016/J.CSDA.2025.108292 , issn =

work page doi:10.1016/j.csda.2025.108292 2026
[75]

2025 , journal =

Liu, Jiaxin and Song, Yunquan , publisher =. 2025 , journal =. doi:10.1080/03610918.2025.2578277 , issn =

work page doi:10.1080/03610918.2025.2578277 2025
[76]

Fu, Bo and Jiang, Dandan , arxivId =

work page
[77]

Tony and Li, Hongzhe , number =

Li, Sai and Cai, T. Tony and Li, Hongzhe , number =. 2022 , journal =. doi:10.1111/RSSB.12479 , issn =

work page doi:10.1111/rssb.12479 2022
[78]

J. R. Statist. Soc. B , author =

work page
[79]

2022 , journal =

Zhang, Yijiao and Zhu, Zhongyi , month =. 2022 , journal =. doi:10.5705/ss.202022.0396 , issn =

work page doi:10.5705/ss.202022.0396 2022
[80]

2025 , journal =

Zhang, Yuhao and Yu, Yang and Sheng, Danshu and Liang, Wanfeng , publisher =. 2025 , journal =. doi:10.1080/00949655.2025.2539481 , issn =

work page doi:10.1080/00949655.2025.2539481 2025
[81]

Tony and Wei, Hongji , number =

Cai, T. Tony and Wei, Hongji , number =. 2021 , journal =. doi:10.1214/20-AOS1949 , issn =

work page doi:10.1214/20-aos1949 2021
[82]

and Yu, Y

Wang, F. and Yu, Y. , number =. 2025 , journal =. doi:10.1093/BIOMET/ASAF018 , issn =

work page doi:10.1093/biomet/asaf018 2025

Showing first 80 references.