The reviewed record of science sign in
Pith

arxiv: 2606.07914 · v1 · pith:FLZ3LF6X · submitted 2026-06-06 · stat.ML · cs.LG

Identifiability and Estimation for Unlabeled Finite Mixtures under Marginal Independence

Takafumi Kanamori , Yushi Hirose , Shohei Yamamoto This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-27 19:32 UTCgrok-4.3pith:FLZ3LF6Xrecord.jsonopen to challenge →

classification stat.ML cs.LG
keywords finite mixturesidentifiabilitymarginal independencemixing matrixcomponent recoverymaximum mean discrepancyunlabeled datamixture models
0
0 comments X

The pith

Marginal independence on one coordinate pair per component identifies all latent components and recovers the mixing matrix from unlabeled mixtures.

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

The paper studies recovery of shared latent components from finite mixtures that have unknown mixing weights and no labels or clean samples. It proves that if each component is independent on at least one coordinate pair, then all components become identifiable and the mixing matrix is recoverable under full-rank and no-cancellation conditions on the matrix. A structural result first shows that any independent affine combination of product components must equal a single component when the univariate marginals are linearly independent. This principle extends to the observable mixtures, so that marginally independent affine combinations isolate the latent components. The work introduces a Product-Marginal Maximum Mean Discrepancy estimator that converges uniformly and remains stable under approximate independence.

Core claim

When every component is independent on some coordinate pair, all components are identifiable, and the mixing matrix is recoverable under the stated completion conditions. Marginally independent affine combinations of the observable mixtures recover the corresponding latent components under full-rank and no-cancellation conditions on the mixing matrix.

What carries the argument

Marginal independence assumption on at least one coordinate pair per component, which lets marginally independent affine combinations isolate individual latent components from the mixtures.

If this is right

  • All components become identifiable once each satisfies marginal independence on some coordinate pair.
  • The mixing matrix is recoverable under the full-rank and no-cancellation completion conditions.
  • The PM-MMD estimator achieves uniform convergence and stability when marginal independence holds approximately.
  • Marginal independence supplies a candidate-level diagnostic through held-out PM-MMD, while irreducibility does not.

Where Pith is reading between the lines

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

  • The method may apply directly to flow-cytometry data where cell-type variables exhibit pairwise independence within each type.
  • The diagnostic based on held-out PM-MMD could serve as a practical check before applying the recovery procedure to new unlabeled datasets.
  • The separation of testable marginal independence from harder-to-check irreducibility suggests prioritizing coordinate pairs that show low dependence in exploratory analysis.

Load-bearing premise

Each latent component must be independent on at least one observed coordinate pair, together with the mixing matrix being full rank and satisfying no-cancellation conditions.

What would settle it

In a controlled experiment with known ground-truth components where each satisfies marginal independence on a coordinate pair, the PM-MMD estimator fails to recover the correct components or mixing matrix.

Figures

Figures reproduced from arXiv: 2606.07914 by Shohei Yamamoto, Takafumi Kanamori, Yushi Hirose.

Figure 1
Figure 1. Figure 1: Simplex visualization of the singleton GMM experiment over ten independent runs [PITH_FULL_IMAGE:figures/full_fig_p044_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Sample-size dependence in the controlled GMM experiments. Panels (a) and (b) [PITH_FULL_IMAGE:figures/full_fig_p045_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Mixture-separation sweep on the raw gated-pool DLBCL experiment. Points show [PITH_FULL_IMAGE:figures/full_fig_p049_3.png] view at source ↗
read the original abstract

We study component recovery and mixing-matrix estimation from unlabeled finite mixtures whose observable distributions share the same latent components but have unknown mixing weights. The main identifying signal is marginal independence: each component is assumed to be independent on at least one coordinate pair, but no labels, clean component samples, or mixing weights are observed. We first prove a structural result for product components: under linear independence of the univariate marginals, any independent affine combination of the components must coincide with a single component. We then extend this principle to observable mixtures and show that, under full-rank and no-cancellation conditions, marginally independent affine combinations recover the corresponding latent components. When every component is independent on some coordinate pair, all components are identifiable, and the mixing matrix is recoverable under the stated completion conditions. Finally, we propose a Product-Marginal Maximum Mean Discrepancy (PM-MMD) estimator over affine combinations of the observable mixtures and prove uniform convergence and stability under approximate marginal independence. This framework also separates the empirical roles of the assumptions: irreducibility is, in general, not directly testable from the unlabeled mixtures alone, whereas marginal independence yields a candidate-level diagnostic through held-out PM-MMD. Controlled and flow-cytometry experiments show when marginal independence provides a useful recovery signal. In the reported multi-component comparisons, condition-aware representative selection stabilizes PM-MMD and improves recovery relative to clustering, factorization, and pairwise mixture-proportion baselines using the same unlabeled mixtures.

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.

Referee Report

2 major / 2 minor

Summary. The paper studies identifiability and estimation of latent components and mixing weights in unlabeled finite mixtures, where the only identifying signal is that each latent component is marginally independent on at least one (possibly component-specific) coordinate pair. It first proves a structural result that, under linear independence of univariate marginals, any independent affine combination of fully product components must equal a single component. This is extended to observable mixtures under full-rank and no-cancellation conditions on the mixing matrix, yielding identifiability of all components and recoverability of the mixing matrix. A Product-Marginal MMD (PM-MMD) estimator over affine combinations is proposed, with proofs of uniform convergence and stability under approximate marginal independence; experiments on controlled data and flow-cytometry data compare it to clustering, factorization, and baseline methods.

Significance. If the central identifiability claims hold, the work provides a new route to component recovery that does not require labels, clean samples, or known proportions and instead exploits verifiable marginal independence. Strengths include the explicit separation of the roles of irreducibility versus marginal independence (with a held-out diagnostic for the latter) and the machine-checked-style proofs of uniform convergence and stability for the PM-MMD estimator. The framework could be useful in domains such as cytometry or genomics where marginal independence on selected coordinate pairs is plausible.

major comments (2)
  1. [extension argument following the product-component structural result] The structural result is stated for fully product components (linear independence of univariate marginals implies any independent affine combination equals one component). The extension to observable mixtures with component-specific (possibly distinct) independence pairs invokes full-rank and no-cancellation on the mixing matrix to conclude that marginally independent affine combinations recover the latent components. It is not clear from the lifting argument whether an affine combination chosen to enforce independence on one specific pair for one component can inadvertently produce a non-product structure that violates the structural result for the remaining components; this step is load-bearing for the claim that every component is identifiable when each has its own independence pair.
  2. [PM-MMD estimator and convergence theorem] The PM-MMD estimator is defined over affine combinations of the observable mixtures and is shown to converge uniformly under approximate marginal independence. However, the stability claim under the case of heterogeneous independence pairs across components relies on the same lifting that is questioned above; if the extension does not fully secure the product structure for all components simultaneously, the uniform convergence guarantee for the estimator may not transfer directly to the multi-component recovery task.
minor comments (2)
  1. [main identifiability theorem] Notation for the coordinate pairs and the selection of which pair is used for each component should be made explicit in the statement of the main identifiability theorem.
  2. [discussion of assumptions] The abstract states that 'irreducibility is, in general, not directly testable from the unlabeled mixtures alone'; a brief remark on whether any diagnostic for the full-rank/no-cancellation conditions is feasible would help readers assess practical applicability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and constructive comments. We address each major comment below. We agree that the lifting argument would benefit from expanded exposition for clarity, and we will revise the manuscript to make the correspondence between observable affine coefficients and effective latent weights fully explicit.

read point-by-point responses

  1. Referee: [extension argument following the product-component structural result] The structural result is stated for fully product components (linear independence of univariate marginals implies any independent affine combination equals one component). The extension to observable mixtures with component-specific (possibly distinct) independence pairs invokes full-rank and no-cancellation on the mixing matrix to conclude that marginally independent affine combinations recover the latent components. It is not clear from the lifting argument whether an affine combination chosen to enforce independence on one specific pair for one component can inadvertently produce a non-product structure that violates the structural result for the remaining components; this step is load-bearing for the claim that every component is identifiable when each has its own independence pair.

    Authors: The lifting argument maps an affine combination of observable mixtures to an affine combination of the latent components whose effective weights are obtained by post-multiplying the mixing matrix by the vector of affine coefficients. Full rank of the mixing matrix guarantees that, for each target component, there is a unique choice of coefficients that isolates it exactly. The no-cancellation condition then ensures that any other choice of coefficients produces a non-zero mixture on the remaining components; because the latent components are fully product and their univariate marginals are linearly independent, such a non-isolating mixture cannot be marginally independent on the designated pair. Consequently, the only affine combinations that achieve marginal independence on a given pair are the isolating ones, each of which the structural result identifies with a single product component. No inadvertent non-product structure arises for the remaining components, because each isolation is performed separately and the structural result applies directly to any distribution that attains independence. We will insert an explicit lemma stating this coefficient-to-weight correspondence and its consequences for the product property. revision: yes

  2. Referee: [PM-MMD estimator and convergence theorem] The PM-MMD estimator is defined over affine combinations of the observable mixtures and is shown to converge uniformly under approximate marginal independence. However, the stability claim under the case of heterogeneous independence pairs across components relies on the same lifting that is questioned above; if the extension does not fully secure the product structure for all components simultaneously, the uniform convergence guarantee for the estimator may not transfer directly to the multi-component recovery task.

    Authors: The uniform convergence theorem is proved directly on the space of affine combinations of the observable mixtures, using only the approximate marginal independence of each latent component on its own pair and standard MMD continuity arguments; it does not invoke simultaneous recovery of all components. The estimator is applied once per independence pair, and the stability result follows from the fact that each pair isolates its corresponding component under the identifiability conditions. With the expanded lifting argument supplied in the revision, the transfer from per-pair convergence to multi-component recovery is immediate. We will add a remark after the convergence theorem that explicitly decouples the uniform convergence statement from the identifiability extension. revision: yes

Circularity Check

0 steps flagged

No circularity: identifiability rests on external linear independence and full-rank assumptions

full rationale

The derivation begins with an explicit structural theorem for product components (linear independence of univariate marginals implies independent affine combinations coincide with single components) and extends it to mixtures via stated full-rank and no-cancellation conditions on the mixing matrix. These are external mathematical assumptions, not quantities fitted inside the paper or defined in terms of the target identifiability result. The PM-MMD estimator is introduced as a new construction with separate uniform convergence and stability proofs under approximate marginal independence; no step renames a fitted parameter as a prediction or reduces the central claim to a self-citation chain. The argument is therefore self-contained against the stated assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

Ledger entries extracted from abstract only; full paper may contain additional fitted constants or domain assumptions not visible here.

axioms (2)
  • domain assumption linear independence of the univariate marginals
    Invoked in the structural result for product components.
  • domain assumption full-rank and no-cancellation conditions on the mixing matrix
    Required to lift the product result to observable mixtures and recover the mixing matrix.
invented entities (1)
  • PM-MMD estimator no independent evidence
    purpose: Estimator over affine combinations of observable mixtures that exploits approximate marginal independence
    New construction proposed in the paper; no independent evidence supplied beyond the convergence claim.

pith-pipeline@v0.9.1-grok · 5796 in / 1423 out tokens · 19297 ms · 2026-06-27T19:32:44.547471+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 6 canonical work pages · 3 internal anchors

  1. [1]

    Stefan Aeberhard and M. Forina. Wine. UCI Machine Learning Repository, 1992

    1992

  2. [2]

    Allman, Catherine Matias, and John A

    Elizabeth S. Allman, Catherine Matias, and John A. Rhodes. Identifiability of parameters in latent structure models with many observed variables.Annals of Statistics, 37(6A):3099– 3132, 2009

    2009

  3. [3]

    Optical recognition of handwritten digits

    Ethem Alpaydin and Cenk Kaynak. Optical recognition of handwritten digits. UCI Machine Learning Repository, 1998

    1998

  4. [4]

    Tensor decompositions for learning latent variable models.J

    Animashree Anandkumar, Rong Ge, Daniel J Hsu, Sham M Kakade, Matus Telgarsky, et al. Tensor decompositions for learning latent variable models.J. Mach. Learn. Res., 15(1):2773–2832, 2014

    2014

  5. [5]

    Semi-supervised novelty detection.The Journal of Machine Learning Research, 11:2973–3009, 2010

    Gilles Blanchard, Gyemin Lee, and Clayton Scott. Semi-supervised novelty detection.The Journal of Machine Learning Research, 11:2973–3009, 2010. 22

    2010

  6. [6]

    Decontamination of Mutually Contaminated Models

    Gilles Blanchard and Clayton Scott. Decontamination of Mutually Contaminated Models. In Samuel Kaski and Jukka Corander, editors,Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics, volume 33 ofProceedings of Machine Learning Research, pages 1–9, Reykjavik, Iceland, 22–25 Apr 2014. PMLR

    2014

  7. [7]

    Oxford University Press, 2013

    St´ ephane Boucheron, G´ abor Lugosi, and Pascal Massart.Concentration Inequalities: A Nonasymptotic Theory of Independence. Oxford University Press, 2013

    2013

  8. [8]

    Nearly optimal sample complexity for learning with label proportions

    Robert Istvan Busa-Fekete, Travis Dick, Claudio Gentile, Haim Kaplan, Tomer Koren, and Uri Stemmer. Nearly optimal sample complexity for learning with label proportions. In Forty-second International Conference on Machine Learning, 2025

    2025

  9. [9]

    Class-prior estimation for learning from positive and unlabeled data

    Marthinus Christoffel, Gang Niu, and Masashi Sugiyama. Class-prior estimation for learning from positive and unlabeled data. InAsian Conference on Machine Learning, pages 221–

  10. [10]

    Hejblum, and Rodolphe Thi´ ebaut

    Daniel Commenges, Chariff Alkhassim, Raphael Gottardo, Boris P. Hejblum, and Rodolphe Thi´ ebaut. cytometree: a binary tree algorithm for automatic gating in cytometry analysis. Cytometry Part A, 93(11):1132–1140, 2018

    2018

  11. [11]

    Archetypal analysis.Technometrics, 36(4):338–347, 1994

    Adele Cutler and Leo Breiman. Archetypal analysis.Technometrics, 36(4):338–347, 1994

    1994

  12. [12]

    Dempster, Nan M

    Arthur P. Dempster, Nan M. Laird, and Donald B. Rubin. Maximum likelihood from incomplete data via the EM algorithm.Journal of the Royal Statistical Society: Series B, 39(1):1–38, 1977

    1977

  13. [13]

    Mixture weight estimation and model prediction in multi-source multi-target domain adaptation.Advances in Neural In- formation Processing Systems, 36:4845–4898, 2023

    Yuyang Deng, Ilja Kuzborskij, and Mehrdad Mahdavi. Mixture weight estimation and model prediction in multi-source multi-target domain adaptation.Advances in Neural In- formation Processing Systems, 36:4845–4898, 2023

    2023

  14. [14]

    Learning classifiers from only positive and unlabeled data

    Charles Elkan and Keith Noto. Learning classifiers from only positive and unlabeled data. InProceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 213–220, 2008

    2008

  15. [15]

    Mixture proportion estimation and PU learning: A modern approach.Advances in Neural Information Processing Systems, 34:8532–8544, 2021

    Saurabh Garg, Yifan Wu, Alexander J Smola, Sivaraman Balakrishnan, and Zachary Lip- ton. Mixture proportion estimation and PU learning: A modern approach.Advances in Neural Information Processing Systems, 34:8532–8544, 2021

    2021

  16. [16]

    Large sample analysis of the median heuristic

    Damien Garreau, Wittawat Jitkrittum, and Motonobu Kanagawa. Large sample analysis of the median heuristic.arXiv preprint arXiv:1707.07269, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  17. [17]

    Gonzalez

    Teofilo F. Gonzalez. Clustering to minimize the maximum intercluster distance.Theoretical Computer Science, 38:293–306, 1985

    1985

  18. [18]

    Borgwardt, Malte J

    Arthur Gretton, Karsten M. Borgwardt, Malte J. Rasch, Bernhard Sch¨ olkopf, and Alexan- der Smola. A kernel two-sample test.Journal of Machine Learning Research, 13(25):723– 773, 2012

    2012

  19. [19]

    A kernel statistical test of independence.Advances in neural information processing systems, 20, 2007

    Arthur Gretton, Kenji Fukumizu, Choon Teo, Le Song, Bernhard Sch¨ olkopf, and Alex Smola. A kernel statistical test of independence.Advances in neural information processing systems, 20, 2007

    2007

  20. [20]

    Nonparametric estimation of component distributions in a multivariate mixture.The annals of statistics, 31(1):201–224, 2003

    Peter Hall and Xiao-Hua Zhou. Nonparametric estimation of component distributions in a multivariate mixture.The annals of statistics, 31(1):201–224, 2003

    2003

  21. [21]

    DEDPUL: Difference-of-estimated-densities-based positive-unlabeled learn- ing.arXiv preprint arXiv:1902.06965, 2019

    Dmitry Ivanov. DEDPUL: Difference-of-estimated-densities-based positive-unlabeled learn- ing.arXiv preprint arXiv:1902.06965, 2019. 23

    work page arXiv 1902

  22. [22]

    Estimating the class prior and pos- terior from noisy positives and unlabeled data.Advances in neural information processing systems, 29, 2016

    Shantanu Jain, Martha White, and Predrag Radivojac. Estimating the class prior and pos- terior from noisy positives and unlabeled data.Advances in neural information processing systems, 29, 2016

    2016

  23. [23]

    Decontamination of mutual contamination models.Journal of Machine Learning Research, 20(41):1–57, 2019

    Julian Katz-Samuels, Gilles Blanchard, and Clayton Scott. Decontamination of mutual contamination models.Journal of Machine Learning Research, 20(41):1–57, 2019

    2019

  24. [24]

    Positive- unlabeled learning with non-negative risk estimator.Advances in neural information pro- cessing systems, 30, 2017

    Ryuichi Kiryo, Gang Niu, Marthinus C Du Plessis, and Masashi Sugiyama. Positive- unlabeled learning with non-negative risk estimator.Advances in neural information pro- cessing systems, 30, 2017

    2017

  25. [25]

    Lee and H

    Daniel D. Lee and H. Sebastian Seung. Learning the parts of objects by non-negative matrix factorization.Nature, 401(6755):788–791, 1999

    1999

  26. [26]

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

    Xuefeng Li, Tongliang Liu, Bo Han, Gang Niu, and Masashi Sugiyama. Provably end- to-end label-noise learning without anchor points. InProceedings of the 38th International Conference on Machine Learning, volume 139 ofProceedings of Machine Learning Research, 2021

    2021

  27. [27]

    Classification with noisy labels by importance reweighting

    Tongliang Liu and Dacheng Tao. Classification with noisy labels by importance reweighting. IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(3):447–461, 2016

    2016

  28. [28]

    Some methods for classification and analysis of multivariate observations

    James MacQueen. Some methods for classification and analysis of multivariate observations. InProceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, volume 1, pages 281–297. University of California Press, 1967

    1967

  29. [29]

    Prior shift estimation for positive unlabeled data through the lens of kernel embedding

    Jan Mielniczuk, Wojciech Rejchel, and Pawel Teisseyre. Class prior estimation for positive- unlabeled learning when label shift occurs.arXiv preprint arXiv:2502.21194, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  30. [30]

    Kernel mean embedding of distributions: A review and beyond.Foundations and Trends in Machine Learning, 10(1–2):1–141, 2017

    Krikamol Muandet, Kenji Fukumizu, Bharath Sriperumbudur, and Bernhard Sch¨ olkopf. Kernel mean embedding of distributions: A review and beyond.Foundations and Trends in Machine Learning, 10(1–2):1–141, 2017

    2017

  31. [31]

    Ng, Michael I

    Andrew Y. Ng, Michael I. Jordan, and Yair Weiss. On spectral clustering: Analysis and an algorithm. InAdvances in Neural Information Processing Systems, volume 14, 2002

    2002

  32. [32]

    Making deep neural networks robust to label noise: A loss correction approach

    Giorgio Patrini, Alessandro Rozza, Aditya Krishna Menon, Richard Nock, and Lizhen Qu. Making deep neural networks robust to label noise: A loss correction approach. InProceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1944– 1952, 2017

    1944

  33. [33]

    Optimum bounds for the distributions of martingales in banach spaces.The Annals of Probability, 22(4):1679–1706, 1994

    Iosif Pinelis. Optimum bounds for the distributions of martingales in banach spaces.The Annals of Probability, 22(4):1679–1706, 1994

    1994

  34. [34]

    Smola, Tib´ erio S

    Novi Quadrianto, Alex J. Smola, Tib´ erio S. Caetano, and Quoc V. Le. Estimating labels from label proportions.Journal of Machine Learning Research, 10:2349–2374, 2009

    2009

  35. [35]

    Mixture proportion estimation via kernel embeddings of distributions

    Harish Ramaswamy, Clayton Scott, and Ambuj Tewari. Mixture proportion estimation via kernel embeddings of distributions. InInternational conference on machine learning, pages 2052–2060, 2016

    2052

  36. [36]

    Unsupervised learning under latent label shift.Advances in Neural Information Processing Systems, 35:18763– 18778, 2022

    Manley Roberts, Pranav Mani, Saurabh Garg, and Zachary Lipton. Unsupervised learning under latent label shift.Advances in Neural Information Processing Systems, 35:18763– 18778, 2022

    2022

  37. [37]

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

    Clayton Scott. A rate of convergence for mixture proportion estimation, with application to learning from noisy labels. InArtificial Intelligence and Statistics, pages 838–846, 2015. 24

    2015

  38. [38]

    Classification with asymmetric label noise: Consistency and maximal denoising

    Clayton Scott, Gilles Blanchard, and Gregory Handy. Classification with asymmetric label noise: Consistency and maximal denoising. In Shai Shalev-Shwartz and Ingo Steinwart, editors,Proceedings of the 26th Annual Conference on Learning Theory, volume 30 ofPro- ceedings of Machine Learning Research, pages 489–511, Princeton, NJ, USA, 12–14 Jun

  39. [39]

    Identifying Finite Mixtures of Nonparametric Product Distributions and Causal Inference of Confounders

    Eleni Sgouritsa, Dominik Janzing, Jonas Peters, and Bernhard Sch¨ olkopf. Identifying finite mixtures of nonparametric product distributions and causal inference of confounders.arXiv preprint arXiv:1309.6860, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  40. [40]

    Kernel distribution embeddings: Uni- versal kernels, characteristic kernels and kernel metrics on distributions.Journal of Machine Learning Research, 19(44):1–29, 2018

    Carl-Johann Simon-Gabriel and Bernhard Sch¨ olkopf. Kernel distribution embeddings: Uni- versal kernels, characteristic kernels and kernel metrics on distributions.Journal of Machine Learning Research, 19(44):1–29, 2018

    2018

  41. [41]

    Nonparametric estimation of multi-view latent variable models

    Le Song, Animashree Anandkumar, Bo Dai, and Bo Xie. Nonparametric estimation of multi-view latent variable models. InInternational Conference on Machine Learning, pages 640–648. PMLR, 2014

    2014

  42. [42]

    Sriperumbudur, Arthur Gretton, Kenji Fukumizu, Bernhard Sch¨ olkopf, and Gert R

    Bharath K. Sriperumbudur, Arthur Gretton, Kenji Fukumizu, Bernhard Sch¨ olkopf, and Gert R. G. Lanckriet. Hilbert space embeddings and metrics on probability measures. Journal of Machine Learning Research, 11:1517–1561, 2010

    2010

  43. [43]

    Invariance assumptions for class distribution estimation.arXiv preprint arXiv:2311.17225, 2023

    Dirk Tasche. Invariance assumptions for class distribution estimation.arXiv preprint arXiv:2311.17225, 2023

    work page arXiv 2023

  44. [44]

    Minimax estima- tion of maximum mean discrepancy with radial kernels.Advances in Neural Information Processing Systems, 29, 2016

    Ilya O Tolstikhin, Bharath K Sriperumbudur, and Bernhard Sch¨ olkopf. Minimax estima- tion of maximum mean discrepancy with radial kernels.Advances in Neural Information Processing Systems, 29, 2016

    2016

  45. [45]

    Dry Bean

    UCI Machine Learning Repository. Dry Bean. UCI Machine Learning Repository, 2020

    2020

  46. [46]

    An efficient and provable approach for mixture proportion estimation using linear independence assumption

    Xiyu Yu, Tongliang Liu, Mingming Gong, Kayhan Batmanghelich, and Dacheng Tao. An efficient and provable approach for mixture proportion estimation using linear independence assumption. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4480–4489, 2018

    2018

  47. [47]

    Mixture proportion estimation beyond irreducibility

    Yilun Zhu, Aaron Fjeldsted, Darren Holland, George Landon, Azaree Lintereur, and Clay- ton Scott. Mixture proportion estimation beyond irreducibility. InInternational Conference on Machine Learning, pages 42962–42982, 2023. 25 Appendix Contents A Notation 26 B Proof of Theorem in Section 3 27 B.1 Proof of Theorem 1 . . . . . . . . . . . . . . . . . . . . ...

    2023

  48. [48]

    Moreover, d(ri)< c, d(r)> cforr∈∆ ¯r\ ∪i∈R(η) 12 Bi

    Thereforedis strongly convex on eachB i ∩∆ ¯r. Moreover, d(ri)< c, d(r)> cforr∈∆ ¯r\ ∪i∈R(η) 12 Bi. Thus, with Ai :={r∈∆ ¯r:d(r)≤c} ∩B i, the level set{r∈∆ ¯r:d(r)≤c}is the disjoint union of the nonempty connected setsA i, and eachA i contains exactly oner i. Letr ∗ be a local minimizer ofdon this level set, and suppose r∗ ∈A i. Sincedis convex onA i,r ∗ ...

  49. [49]

    Max vtx. dist

    Hence the same argument as in part (a) gives {r∈∆ ¯r: bd(r)≤c ′}= [ i∈R(η) 12 bAi, bAi :={r∈∆ ¯r: bd(r)≤c ′} ∩B i, where the bAi are nonempty, pairwise disjoint, and connected, and each contains exactly one ri. Moreover, bdis strongly convex on eachB i ∩∆ ¯r. Thus any local minimizer br∈ bAi is the minimizer of bd, equivalently of MMD(r TbU), over bAi. Th...

    work page arXiv 2000