pith. sign in

arxiv: 2303.11789 · v12 · submitted 2023-03-20 · 💻 cs.LG · cs.DC· cs.SY· eess.SY· math.PR

Decentralized Online Learning for Random Inverse Problems Over Graphs

Xiwei Zhang , Tao Li , Yan Chen , Qianyuan Long This is my paper

Pith reviewed 2026-05-24 09:11 UTC · model grok-4.3

classification 💻 cs.LG cs.DCcs.SYeess.SYmath.PR
keywords decentralized learningonline learninginverse problemsHilbert spacespersistence of excitationconsistencyRKHSrandom difference equations
0
0 comments X

The pith

Connected graphs and persistent excitation on forward operators guarantee mean-square and almost-sure consistency of decentralized online estimates in Hilbert spaces.

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

The paper proposes a decentralized online learning algorithm that solves random inverse problems over network graphs from streaming measurements. It unifies distributed parameter estimation in Hilbert spaces with least-mean-square learning in reproducing kernel Hilbert spaces by recasting convergence as L2-asymptotic stability of inhomogeneous random difference equations driven by martingale terms. Under the assumptions of graph connectivity and an infinite-dimensional spatio-temporal persistence of excitation condition on the forward operators, every node’s estimate converges both in mean square and almost surely to the true solution. A reader cares because the result supplies rigorous convergence guarantees for distributed adaptive learning from non-stationary data without requiring a fusion center.

Core claim

If the network graph is connected and the sequence of forward operators satisfies the infinite-dimensional spatio-temporal persistence of excitation condition, then the estimates of all nodes are mean square and almost surely strongly consistent. The same conclusion holds for the RKHS version when the operators induced by random input data meet the analogous condition. The proof proceeds by establishing L2-asymptotic stability theory for the associated class of inhomogeneous random difference equations in Hilbert spaces.

What carries the argument

L2-asymptotic stability theory for inhomogeneous random difference equations in Hilbert spaces with L2-bounded martingale difference terms, which converts algorithm convergence into stability under the persistence-of-excitation condition.

If this is right

  • All nodes obtain strongly consistent estimates without any central data aggregation.
  • The same algorithm and proof cover both general Hilbert-space inverse problems and kernel-based RKHS learning from non-stationary streams.
  • Mean-square and almost-sure consistency hold simultaneously once the persistence condition is met.
  • The result extends classical adaptive filtering theory to infinite-dimensional distributed settings.

Where Pith is reading between the lines

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

  • Empirical checks of the persistence condition on observed operator sequences could serve as a practical diagnostic for expected convergence speed in deployed sensor networks.
  • The stability framework may transfer directly to other distributed stochastic approximation tasks that can be written as inhomogeneous random recursions in Hilbert spaces.
  • On graphs that are only weakly connected, the same persistence condition might still guarantee local consistency within each connected component.

Load-bearing premise

The sequence of forward operators must satisfy the infinite-dimensional spatio-temporal persistence of excitation condition.

What would settle it

Exhibit a connected graph together with a sequence of forward operators that violates the spatio-temporal persistence of excitation condition and show that the node estimates fail to converge in mean square or almost surely to the true parameter.

Figures

Figures reproduced from arXiv: 2303.11789 by Qianyuan Long, Tao Li, Xiwei Zhang, Yan Chen.

Figure 1
Figure 1. Figure 1: (a) estimates of nodes fi , i = 1, · · · , 10 for k = 1000; (b) estimates of nodes fi , i = 1, · · · , 10 for k = 100000. 6. CONCLUSIONS We have established a framework of random inverse problems with online measurements over graphs, and present a decentralized online learning algorithm with online data streams, which unifies the distributed parameter estimation in Hilbert spaces and the least mean square … view at source ↗
read the original abstract

We propose a decentralized online learning algorithm for distributed random inverse problems over network graphs with online measurements, and unifies the distributed parameter estimation in Hilbert spaces and the least mean square problem in reproducing kernel Hilbert spaces (RKHS-LMS). We transform the convergence of the algorithm into the asymptotic stability of a class of inhomogeneous random difference equations in Hilbert spaces with $L_{2}$-bounded martingale difference terms and develop the $L_2$-asymptotic stability theory in Hilbert spaces. We show that if the network graph is connected and the sequence of forward operators satisfies the infinite-dimensional spatio-temporal persistence of excitation condition, then the estimates of all nodes are mean square and almost surely strongly consistent. Moreover, we propose a decentralized online learning algorithm in RKHS based on non-stationary online data streams, and prove that the algorithm is mean square and almost surely strongly consistent if the operators induced by the random input data satisfy the infinite-dimensional spatio-temporal persistence of excitation condition.

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

1 major / 0 minor

Summary. The paper proposes a decentralized online learning algorithm for random inverse problems over network graphs, unifying distributed parameter estimation in Hilbert spaces with RKHS-LMS. It reduces algorithm convergence to L2-asymptotic stability of inhomogeneous random difference equations driven by martingale-difference terms, develops the corresponding stability theory in Hilbert spaces, and concludes that node estimates are mean-square and almost-surely strongly consistent whenever the graph is connected and the sequence of forward operators (or induced operators in the RKHS case) satisfies an infinite-dimensional spatio-temporal persistence-of-excitation condition.

Significance. If the L2-stability theory is rigorously established, the unification of the Hilbert-space and RKHS settings and the explicit reduction to a verifiable excitation condition would constitute a useful contribution to distributed online learning for inverse problems. The explicit statement of the persistence condition as a hypothesis is a strength, as is the focus on both mean-square and almost-sure convergence.

major comments (1)
  1. [Abstract / §3] Abstract and §3 (stability-to-consistency step): the upgrade from L2-asymptotic stability of the inhomogeneous random difference equation to mean-square and a.s. strong consistency is obtained solely by invoking the infinite-dimensional spatio-temporal persistence-of-excitation condition. No argument is supplied showing that this condition produces a uniform lower bound on the accumulated operator in the strong operator topology (or on the covariance operator in the RKHS setting) that is sufficient for the infinite-dimensional case; the manuscript treats the condition as a black-box hypothesis whose sufficiency is left unverified.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and valuable feedback on our manuscript. We address the major comment below and indicate the planned revisions.

read point-by-point responses

  1. Referee: [Abstract / §3] Abstract and §3 (stability-to-consistency step): the upgrade from L2-asymptotic stability of the inhomogeneous random difference equation to mean-square and a.s. strong consistency is obtained solely by invoking the infinite-dimensional spatio-temporal persistence-of-excitation condition. No argument is supplied showing that this condition produces a uniform lower bound on the accumulated operator in the strong operator topology (or on the covariance operator in the RKHS setting) that is sufficient for the infinite-dimensional case; the manuscript treats the condition as a black-box hypothesis whose sufficiency is left unverified.

    Authors: We agree with the referee that the manuscript would benefit from an explicit argument linking the infinite-dimensional spatio-temporal persistence-of-excitation condition to the required uniform lower bound. While the condition is introduced as a hypothesis and the consistency claims are stated in terms of it, a self-contained derivation showing that the condition implies the necessary bound in the strong operator topology (Hilbert-space case) and on the covariance operator (RKHS case) is not detailed in §3. In the revised version we will add a supporting lemma (placed immediately after the stability theorem) that rigorously establishes this implication, thereby removing any appearance of a black-box step and making the stability-to-consistency transition fully explicit. revision: yes

Circularity Check

0 steps flagged

No circularity; consistency follows from external L2-stability theory plus stated excitation assumption.

full rationale

The derivation chain first develops an L2-asymptotic stability result for inhomogeneous random difference equations driven by martingale-difference terms in Hilbert space, then applies the infinite-dimensional spatio-temporal persistence of excitation condition (an external hypothesis on the forward operators) to obtain mean-square and a.s. consistency. The abstract and extracted claim treat this condition as a black-box input whose verification is left to the user; it is not obtained by fitting, self-definition, or self-citation within the paper. No load-bearing step reduces by construction to the algorithm outputs or to prior self-citations. The result is therefore independent of its own fitted values or renamed inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard assumptions from distributed optimization and functional analysis plus the key excitation condition introduced as a hypothesis.

axioms (2)
  • domain assumption The network graph is connected.
    Explicitly required for the consistency result to hold across all nodes.
  • domain assumption The sequence of forward operators satisfies the infinite-dimensional spatio-temporal persistence of excitation condition.
    Load-bearing hypothesis that enables the stability theory to imply strong consistency.

pith-pipeline@v0.9.0 · 5707 in / 1142 out tokens · 32545 ms · 2026-05-24T09:11:01.955892+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

76 extracted references · 76 canonical work pages

  1. [1]

    Bertero and P

    M. Bertero and P . Boccacci, Introduction to Inverse Problems in Imaging . Boca Raton: CRC Press, 1998

    work page 1998

  2. [2]

    Colton and R

    D. Colton and R. Kress, Inverse Acoustic and Electromagnetic Scattering Theory . Berlin: Springer Science and Business Media, 2012

    work page 2012

  3. [3]

    H. W. Engl, M. Hanke and A. Neubauer, Regularization of Inverse Problems . Berlin: Springer Science and Business Media, 1996

    work page 1996

  4. [4]

    V . A. Morozov, Methods for Solving Incorrectly Posed Problems . Berlin: Springer Science and Business Media, 2012

    work page 2012

  5. [5]

    A. G. Werschulz, The Computational Complexity of Differential and Integral Equations: An Information-Based Approach . Oxford: Oxford University Press, 1991

    work page 1991

  6. [6]

    Regularization of incorrectly posed pr oblems,

    A. N. Tikhonov, “Regularization of incorrectly posed pr oblems,” Soviet Mathematics Doklady, vol. 4, pp. 1624–1627, 1963

    work page 1963

  7. [7]

    Consistency and rate s of convergence of nonlinear Tikhonov regularization with random noise,

    N. Bissantz, T. Hohage and A. Munk, “Consistency and rate s of convergence of nonlinear Tikhonov regularization with random noise,” Inverse Probl., vol. 20, no. 6, pp. 1773, 2004

    work page 2004

  8. [8]

    Nonparametric statistical inverse probl ems,

    L. Cavalier, “Nonparametric statistical inverse probl ems,” Inverse Probl., vol. 24, no. 3, pp. 034004, 2008

    work page 2008

  9. [9]

    Analysis of regu larized inversion of data corrupted by white Gaussian noise,

    H. Kekkonen, M. Lassas and S. Siltanen, “Analysis of regu larized inversion of data corrupted by white Gaussian noise,” Inverse Probl., vol. 30, no. 4, pp. 045009, 2014

    work page 2014

  10. [10]

    Gine and R

    E. Gine and R. Nickl, Mathematical F oundations of Infinite-Dimensional Statistical Models. Cambridge: Cambridge University Press, 2015. August 29, 2024 DRAFT JOURNAL OF LATEX CLASS FILES, JUNE 2023 74

    work page 2015

  11. [11]

    Inverse problems with Poisson data: Statistical regularization theory, applications and algorithms,

    T. Hohage and F. Werner, “Inverse problems with Poisson data: Statistical regularization theory, applications and algorithms,” Inverse Probl., vol. 32, no. 9, pp. 093001, 2016

    work page 2016

  12. [12]

    Discrepancy based model selection in statistical inverse problems,

    S. Lu and P . Math´ e , “Discrepancy based model selection in statistical inverse problems,” J. Complex., vol. 30, no. 3, pp. 290–308, 2014

    work page 2014

  13. [13]

    Regularization of some linear ill-posed problems with discretized random noisy data,

    P . Math´ e and S. V . Pereverzev, “Regularization of some linear ill-posed problems with discretized random noisy data,” Math. Comput. , vol. 75, no. 256, pp. 1913–1929, 2006

    work page 1913

  14. [14]

    Filter bas ed methods for statistical linear inverse problems,

    M. A. Iglesias, K. Lin, S. Lu and A. M. Stuart, “Filter bas ed methods for statistical linear inverse problems,” Commun. Math. Sci. , vol. 15, no. 7, pp. 1867–1896, 2017

    work page 2017

  15. [15]

    On the asymptotical regular ization for linear inverse problems in presence of white noise,

    S. Lu, P . Niu and F. Werner, “On the asymptotical regular ization for linear inverse problems in presence of white noise,” SIAM-ASA J. Uncertain. Quantif. , vol. 9, no. 1, pp. 1–28, 2021

    work page 2021

  16. [16]

    Stochastic gradient descent for li near inverse problems in Hilbert spaces,

    S. Lu and P . Math´ e, “Stochastic gradient descent for li near inverse problems in Hilbert spaces,” Math. Comput. , vol. 91, no. 336, pp. 1763–1788, 2017

    work page 2017

  17. [17]

    Randomized matri x approximation to enhance regularized projection schemes in inverse problems,

    S. Lu, P . Math´ e and S. V . Pereverzev, “Randomized matri x approximation to enhance regularized projection schemes in inverse problems,” Inverse Probl. , vol. 36, no. 8, pp. 085013, 2020

    work page 2020

  18. [18]

    Iterate averaging, the Kalm an filter, and 3DV AR for linear inverse problems,

    F. G. Jones and G. Simpson, “Iterate averaging, the Kalm an filter, and 3DV AR for linear inverse problems,” Numer . Algorithms, vol. 92, no. 2, pp. 1–21, 2022

    work page 2022

  19. [19]

    Optimal discretization of inverse problems in Hilbert scales. Regularization and self-regularization of projection met hods,

    P . Math´ e and S. V . Pereverzev, “Optimal discretization of inverse problems in Hilbert scales. Regularization and self-regularization of projection met hods,” SIAM J. Numer . Anal. , vol. 38, no. 6 , pp. 1999–2021, 2001

    work page 1999

  20. [20]

    On the discrepancy principle for sto chastic gradient descent,

    T. Jahn and B. Jin, “On the discrepancy principle for sto chastic gradient descent,” Inverse Probl., vol. 36, no. 9, pp. 095009, 2020

    work page 2020

  21. [21]

    On the regularizing property of stocha stic gradient descent,

    B. Jin and X. Lu, “On the regularizing property of stocha stic gradient descent,” Inverse Probl., vol. 35, no. 1, pp. 015004, 2018

    work page 2018

  22. [22]

    On the optimality of ave raging in distributed statistical learning,

    J. D. Rosenblatt and B. Nadler, “On the optimality of ave raging in distributed statistical learning,” Inf. Inference, vol. 5, no. 4, pp. 379–404, 2016

    work page 2016

  23. [23]

    Diffusion least-mean squar es over adaptive networks: Formulation and performance analysis,

    C. G. Lopes and A. H. Sayed, “Diffusion least-mean squar es over adaptive networks: Formulation and performance analysis,” IEEE Trans. Signal Process. , vol. 56, no. 7, pp. 3122–3136, 2008

    work page 2008

  24. [24]

    Convergence rate analysis of d istributed gossip (linear parameter) estimation: Fundamental limits and tradeoff,

    S. Kar and J. M. F. Moura, “Convergence rate analysis of d istributed gossip (linear parameter) estimation: Fundamental limits and tradeoff,” IEEE J. Sel. Top. Signal Process. , August 29, 2024 DRAFT JOURNAL OF LATEX CLASS FILES, JUNE 2023 75 vol. 5, no. 4, pp. 674–690, 2011

    work page 2024

  25. [25]

    Diffusion strategies for distributed Kalman filtering and smoothing,

    F. S. Cattivelli and A. H. Sayed, “Diffusion strategies for distributed Kalman filtering and smoothing,” IEEE Trans. Autom. Control , vol. 55, no. 9, pp. 2069–2084, 2010

    work page 2069

  26. [26]

    Robust distri buted estimation by networked agents,

    S. Al-Sayed, A. M. Zoubir and A. H. Sayed, “Robust distri buted estimation by networked agents,” IEEE Trans. Signal Process. , vol. 65, no. 15, pp. 3909–3921, 2017

    work page 2017

  27. [27]

    D iffusion estimation over cooperative multi-agent networks with missing data,

    M. R. Gholami, M. Jansson, E. G. Str¨ om and A. H. Sayed, “D iffusion estimation over cooperative multi-agent networks with missing data,” IEEE Trans. Signal Inf. Proc. Netw. , vol. 2, no. 3, pp. 276–289, 2016

    work page 2016

  28. [28]

    Diffusion ada ptation over multi-agent networks with wireless link impairments,

    R. Abdolee, B. Champagne and A. H. Sayed, “Diffusion ada ptation over multi-agent networks with wireless link impairments,” IEEE. Trans. Mob. Comput. , vol. 15, no. 6, pp. 1362–1376, 2016

    work page 2016

  29. [29]

    Distributed param eter estimation in sensor networks: Nonlinear observation models and imperfect comm unication,

    S. Kar, J. M. F. Moura and K. Ramanan, “Distributed param eter estimation in sensor networks: Nonlinear observation models and imperfect comm unication,” IEEE Trans. Inf. Theory, vol. 58, no. 6, pp. 3575–3605, 2012

    work page 2012

  30. [30]

    Consensus+innovations distr ibuted inference over networks: Cooperation and sensing in networked systems,

    S. Kar and J. M. F. Moura, “Consensus+innovations distr ibuted inference over networks: Cooperation and sensing in networked systems,” IEEE Signal Process. Mag. , vol. 30, no. 3, pp. 99–109, 2013

    work page 2013

  31. [31]

    Diffusion LMS with correlate d regressors I: Realization-wise stability,

    M. J. Piggott and V . Solo, “Diffusion LMS with correlate d regressors I: Realization-wise stability,” IEEE Trans. Signal Process. , vol. 64, no. 21, pp. 5473–5484, 2016

    work page 2016

  32. [32]

    Diffusion LMS with correlate d regressors II: Performance,

    M. J. Piggott and V . Solo, “Diffusion LMS with correlate d regressors II: Performance,” IEEE Trans. Signal Process. , vol. 65, no. 15, pp. 3934-3947, 2017

    work page 2017

  33. [33]

    Decentralized cooperative online estimation with random observation matrices, communication graphs and time delay s,

    J. Wang, T. Li and X. Zhang, “Decentralized cooperative online estimation with random observation matrices, communication graphs and time delay s,” IEEE Trans. Inf. Theory , vol. 67, no. 6, pp. 4035–4059, 2021

    work page 2021

  34. [34]

    Diffusion LMS filter for distributed estimation of systems with stochastic state transition and observation m atrices,

    J. Y . Ishihara and S. C. Alghunaim, “Diffusion LMS filter for distributed estimation of systems with stochastic state transition and observation m atrices,” in Proc. 2017 Am. Control Conf., Seattle, USA, May. 24-26, 2017, pp. 5199–5204

    work page 2017

  35. [35]

    Sup ervised learning as an inverse problem based on non-smooth loss function,

    S. Lyaqini, M. Quafafou, M. Nachaoui and A. Chakib, “Sup ervised learning as an inverse problem based on non-smooth loss function,” Knowl. Inf. Syst. , vol. 62, no. 8, pp. 3039– 3058, 2020

    work page 2020

  36. [36]

    On the mathematical foundation s of learning,

    S. Smale and F. Cucker, “On the mathematical foundation s of learning,” Bull. Amer . Math. Soc., vol. 39, no.1, pp. 1–49, 2002. August 29, 2024 DRAFT JOURNAL OF LATEX CLASS FILES, JUNE 2023 76

    work page 2002

  37. [37]

    Online gradient descent learnin g algorithms,

    Y . Ying and M. Pontil, “Online gradient descent learnin g algorithms,” F ound. Comput. Math., vol. 8, pp. 561–596, 2008

    work page 2008

  38. [38]

    Online learning as stochastic ap proximation of regularization paths: Optimality and almost-sure convergence,

    P . Tarr` es and Y . Y ao, “Online learning as stochastic ap proximation of regularization paths: Optimality and almost-sure convergence,” IEEE Trans. Inf. Theory , vol. 60, no. 9, pp. 5716–5735, 2014

    work page 2014

  39. [39]

    Nonparametric stochastic a pproximation with large step-sizes,

    A. Dieuleveut and F. Bach, “Nonparametric stochastic a pproximation with large step-sizes,” Ann. Stat. , vo. 44, no. 4, pp. 1363–1399, 2016

    work page 2016

  40. [40]

    D istributed adaptive learning with multiple kernels in diffusion networks,

    B. Shin, M. Y ukawa, R. L. G. Cavalcante and A. Dekorsy, “D istributed adaptive learning with multiple kernels in diffusion networks,” IEEE Trans. Signal Process. , vol. 66, no. 21, pp. 5505–5519, 2018

    work page 2018

  41. [41]

    Learning theory wit h consensus in reproducing kernel Hilbert spaces,

    Z. Deng, J. Gregory and A. Kurdila, “Learning theory wit h consensus in reproducing kernel Hilbert spaces,” in Proc. 2012 Am. Control Conf. , Montreal, Canada, Jun. 27-29, 2012, pp. 1400–1405

    work page 2012

  42. [42]

    The diffusion-KLMS algorithm,

    R. Mitra and V . Bhatia, “The diffusion-KLMS algorithm, ” in Proc. 2014 ICCIT , Bhubaneswar, India, Dec. 22-24, 2014, pp. 256–259

    work page 2014

  43. [43]

    A diffusion kernel LMS alg orithm for nonlinear adaptive networks,

    R. Chouvardas and M. Draief, “A diffusion kernel LMS alg orithm for nonlinear adaptive networks,” in Proc. 2016 IEEE Int. Conf. Acoust. Speech Signal Process. , Shanghai, China, Mar. 20-25, 2016, pp. 4164–4168

    work page 2016

  44. [44]

    Onlin e distributed learning over networks in RKHS spaces using random Fourier features,

    P . Bouboulis, S. Chouvardas and S. Theodoridis, “Onlin e distributed learning over networks in RKHS spaces using random Fourier features,” IEEE Trans. Signal Process. , vol. 66, no. 7, pp. 1920–1932, 2018

    work page 1920

  45. [45]

    Online gradient desc ent algorithms for functional data learning,

    X. Chen, B. Tang, J. Fan and X. Guo, “Online gradient desc ent algorithms for functional data learning,” J. Complex. , vol. 70, pp. 101635, 2022

    work page 2022

  46. [46]

    Convergence of unregulariz ed online learning algorithms,

    Y . Lei, L. Shi and Z. C. Guo, “Convergence of unregulariz ed online learning algorithms,” J. Mach. Learn. Res. , vo. 18, no. 171, pp. 1-33, 2018

    work page 2018

  47. [47]

    Distributed learning wit h multi-penalty regularization,

    Z. C. Guo, S. B. Lin and L. Shi, “Distributed learning wit h multi-penalty regularization,” Appl. Comput. Harmon. Anal. , vol. 46, no. 3, pp. 478–499, 2019

    work page 2019

  48. [48]

    Learning theory of dis tributed spectral algorithms,

    Z. C. Guo, S. B. Lin and D. X. Zhou, “Learning theory of dis tributed spectral algorithms,” Inverse Probl., vol. 33, no. 7, pp. 074009, 2017

    work page 2017

  49. [49]

    Optimal convergence for distribu ted learning with stochastic gradient methods and spectral algorithms,

    J. Lin and V . Cevher, “Optimal convergence for distribu ted learning with stochastic gradient methods and spectral algorithms,” J. Mach. Learn. Res. , vol. 21, no. 147, pp. 1–63, 2020

    work page 2020

  50. [50]

    Online learning algorithms,

    S. Smale and Y . Y ao, “Online learning algorithms,” F ound. Comput. Math. , vol. 6, pp. August 29, 2024 DRAFT JOURNAL OF LATEX CLASS FILES, JUNE 2023 77 145–170, 2006

    work page 2024

  51. [51]

    Mean stability of a stoc hastic difference equation,

    V . M. Ungureanu and S. S. Cheng, “Mean stability of a stoc hastic difference equation,” Ann. Pol. Math. , vol. 1, no. 93, pp. 33–52, 2008

    work page 2008

  52. [52]

    Applied stochastic approximation alg orithms in Hilbert space,

    S. C. Kubrusly, “Applied stochastic approximation alg orithms in Hilbert space,” Int. J. Control, vol. 28, no. 1, pp. 23–31, 1978

    work page 1978

  53. [53]

    Formalization of a stochastic approximation theorem,

    K. V ajjha, B. Trager, A. Shinnar and V . Pestun, “Formalization of a stochastic approximation theorem,” arXiv:2202.05959, 2022

    work page arXiv 2022

  54. [54]

    Stability, stabilizability and dete ctability for Markov jump discrete-time linear systems with multiplicative noise in Hilbert spaces ,

    V . M. Ungureanu, “Stability, stabilizability and dete ctability for Markov jump discrete-time linear systems with multiplicative noise in Hilbert spaces ,” Optimization, vol. 63, no. 11, pp. 1689–1712, 2014

    work page 2014

  55. [55]

    Optimal control for linear discrete- time systems with Markov pertur- bations in Hilbert spaces,

    V . M. Ungureanu, “Optimal control for linear discrete- time systems with Markov pertur- bations in Hilbert spaces,” IMA J. Math. Control Inf. , vol. 26, no. 1, pp. 105–127, 2009

    work page 2009

  56. [56]

    Detectability, obs ervability and Lyapunov-type theorems of linear discrete time-varying stochastic syste ms with multiplicative noise,

    W. Zhang, W. X. Zheng and B. S. Chen, “Detectability, obs ervability and Lyapunov-type theorems of linear discrete time-varying stochastic syste ms with multiplicative noise,” Int. J. Control, vol. 90, no. 11, pp. 2490–2507, 2017

    work page 2017

  57. [57]

    Efficient algorithms for the regularization of dynamic inverse problems: I. Theory,

    U. Schmitt and A. K. Louis, “Efficient algorithms for the regularization of dynamic inverse problems: I. Theory,” Inverse Probl., vol. 18, no. 3, pp. 6, 2002

    work page 2002

  58. [58]

    Modern regularization metho ds for inverse problems,

    M. Benning and M. Burger, “Modern regularization metho ds for inverse problems,” Acta Numer ., vol. 27, pp. 1–111, 2018

    work page 2018

  59. [59]

    Convergence of generic in finite products of affine operators,

    S. Reich and A. J. Zaslavski, “Convergence of generic in finite products of affine operators,” Abstract Appl. Anal. , vol. 4, no. 1, pp. 1–19, 1999

    work page 1999

  60. [60]

    Infinite products of arbitra ry operators and intersections of subspaces in Hilbert space,

    E. Pustylnik and S. Reich, “Infinite products of arbitra ry operators and intersections of subspaces in Hilbert space,” J. Approx. Theory , vol. 178, pp. 91–102, 2014

    work page 2014

  61. [61]

    Convergenc e of non-cyclic infinite products of operators,

    E. Pustylnik, S. Reich and A. J. Zaslavski, “Convergenc e of non-cyclic infinite products of operators,” J. Math. Anal. Appl. , vol. 380, no. 2, pp. 759–767, 2011

    work page 2011

  62. [62]

    Convergenc e of non-periodic infinite products of orthogonal projections and nonexpansive operators in Hi lbert space,

    E. Pustylnik, S. Reich and A. J. Zaslavski, “Convergenc e of non-periodic infinite products of orthogonal projections and nonexpansive operators in Hi lbert space,” J. Approx. Theory , vol. 164, no. 5, pp. 611–624, 2012

    work page 2012

  63. [63]

    Stability of recursive stochastic tracking al gorithms,

    L. Guo, “Stability of recursive stochastic tracking al gorithms,” SIAM J. Control Optim. , vol. 32, no. 5, pp. 1195–1225, 1994

    work page 1994

  64. [64]

    Estimating time-varying parameters by the Kal man filter based algorithm: Stability and convergence,

    L. Guo, “Estimating time-varying parameters by the Kal man filter based algorithm: Stability and convergence,” IEEE Trans. Autom. Control , vol. 35, no. 2, pp. 141–147, 1990. August 29, 2024 DRAFT 78

    work page 1990

  65. [65]

    Exponential stability of general t racking algorithms,

    L. Guo and L. Ljung, “Exponential stability of general t racking algorithms,” IEEE Trans. Autom. Control , vol. 40, no. 8, pp. 1376–1387, 1995

    work page 1995

  66. [66]

    Performance analysis of general tr acking algorithms,

    L. Guo and L. Ljung, “Performance analysis of general tr acking algorithms,” IEEE Trans. Autom. Control , vol. 40, no. 8, pp. 1388–1402, 1995

    work page 1995

  67. [67]

    Performance analysis of forgetting factor RLS algorithms,

    L. Guo, L. Ljung and P . Priouret, “Performance analysis of forgetting factor RLS algorithms,” Int. J. Adapt. Control Signal Process. , vol. 7, no. 6, pp. 525–537, 1993

    work page 1993

  68. [68]

    Necessary and sufficient co nditions for stability of LMS,

    L. Guo, L. Ljung and G. Wang, “Necessary and sufficient co nditions for stability of LMS,” IEEE Trans. Autom. Control , vol. 42, no. 6, pp. 761–770, 1997

    work page 1997

  69. [69]

    Hyt¨ onen, J

    T. Hyt¨ onen, J. V an Neerven, M. V eraar and L. Weis, Analysis in Banach spaces . Berlin: Springer, 2016

    work page 2016

  70. [70]

    Hyt¨ onen, J

    T. Hyt¨ onen, J. V an Neerven, M. V eraar and L. Weis, Analysis in Banach Spaces: V olume II: Probabilistic Methods and Operator Theory . Berlin: Springer, 2018

    work page 2018

  71. [71]

    Analysis of normalized least mean squ ares-based consensus adaptive filters under a general information condition,

    S. Xie and L. Guo, “Analysis of normalized least mean squ ares-based consensus adaptive filters under a general information condition,” SIAM J. Control Optim. , vol. 56, no. 5, pp. 3404–3431, 2018

    work page 2018

  72. [72]

    Zhang, T

    X. Zhang, T. Li and X. Fu, “Decentralized online regular ized learning over random time- varying graphs, arXiv: 2206.03861, 2024

    work page arXiv 2024

  73. [73]

    Hille and R

    E. Hille and R. C. Phillips, Functional Analysis and Semi-groups . Providence: American Mathematical Society, 1996

    work page 1996

  74. [74]

    Remarks on measurabil ity of operator-valued functions,

    O. Blasco and I. Garcia-Bayona, “Remarks on measurabil ity of operator-valued functions,” Mediterr . J. Math., no. 13, pp. 5147–5162, 2016

    work page 2016

  75. [75]

    Y . S. Chow and H. Teicher, Probability Theory: Independence, Tnterchangeability, M artin- gales. Berlin: Springer Science and Business Media, 2012

    work page 2012

  76. [76]

    A convergence theorem for n on-negative almost super- martingales and some applications,

    H. Robbins and D. Siegmund, “A convergence theorem for n on-negative almost super- martingales and some applications,” Optimizing Methods in Statistics , pp. 233–257, 1971. August 29, 2024 DRAFT

    work page 1971