pith. sign in

arxiv: 2604.22510 · v1 · submitted 2026-04-24 · 🧮 math.PR

Asymptotics of Multi-Scale McKean--Vlasov Diffusions with Super-Linear Kernels: a Lifted Semigroup Approach

Wei Hong , Shanshan Hu , Wei Liu , Shiyuan Yang This is my paper

Pith reviewed 2026-05-08 10:17 UTC · model grok-4.3

classification 🧮 math.PR
keywords multi-scale McKean-Vlasov diffusionssuper-linear kernelssmall-noise asymptoticsfunctional law of large numberslarge deviation principlelifted semigroupnonlinear Markov processesviable pairs
0
0 comments X

The pith

Multi-scale McKean-Vlasov diffusions with super-linear kernels obey a functional law of large numbers and large deviation principle in the small-noise limit.

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

This paper establishes the small-noise asymptotic behavior for multi-scale McKean-Vlasov diffusions in which the interaction kernels grow faster than linearly and depend on the probability laws of both the slow and fast components. The frozen system therefore becomes a nonlinear Markov process, which complicates standard averaging arguments. The authors introduce a lifted semigroup construction together with a generalized Khasminskii discretization to obtain the deterministic limit of the slow variable along with explicit convergence rates. They further prove the large deviation principle by means of lifted viable pairs and a generalized functional occupation measure method. The results apply to consensus-based optimization algorithms arising in machine learning and multilevel optimization.

Core claim

In the small-noise limit the slow component converges to the solution of a deterministic nonlinear equation that arises from the McKean-Vlasov dynamics of the frozen system, and the paths of the slow component satisfy a large deviation principle whose rate function is characterized through the lifted viable pair. The proof combines a lifted semigroup argument, a generalized Khasminskii time-discretization scheme that yields explicit rates, and a functional occupation-measure approach that establishes the equivalent Laplace principle. The framework requires only local Lipschitz continuity on the super-linear kernels.

What carries the argument

The lifted semigroup, an extension of the Markov semigroup that incorporates dependence on the joint laws of the slow and fast variables to construct the nonlinear frozen dynamics.

If this is right

  • The slow variable converges to its deterministic limit with explicit quantitative rates.
  • The large deviation principle holds and is equivalent to the Laplace principle.
  • The results remain valid under local Lipschitz continuity rather than global Lipschitz conditions on the kernels.
  • The same framework applies to multi-scale consensus-based optimization methods used in machine learning and multilevel optimization.

Where Pith is reading between the lines

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

  • The lifting technique may extend averaging principles to other classes of super-linear mean-field models that are not diffusive.
  • The explicit convergence rates could inform step-size selection in numerical schemes for simulating these multi-scale systems.
  • The occupation-measure method may connect to large-deviation analysis of mean-field control problems with fast-slow separation.

Load-bearing premise

The system obtained by freezing the slow variable must still define a well-posed nonlinear Markov process even though the interaction kernels grow super-linearly, and the kernels must satisfy local Lipschitz conditions.

What would settle it

Numerical simulation of a concrete super-linear kernel example in which the slow component fails to converge to the predicted deterministic limit as the noise intensity tends to zero, or in which the observed large-deviation rate function differs from the one derived via the lifted viable pair.

read the original abstract

In this work, we establish the small-noise asymptotic behaviour (namely, the functional law of large numbers and the large deviation principle) for multi-scale McKean--Vlasov diffusions with super-linear kernels. In this setting, the interaction depends on the laws of both the slow component and the fast oscillating process. Consequently, the frozen (parameterized) system exhibits McKean--Vlasov dynamics, forming a nonlinear Markov process and thereby rendering the analysis more complex compared to existing works. We develop a lifted semigroup argument and employ a generalized Khasminskii time discretization scheme to derive the small-noise limit of the slow variable, providing explicit convergence rates. Furthermore, we introduce the notion of a lifted viable pair and utilize a generalized functional occupation measure approach to establish the Laplace principle, which is equivalent to the large deviation principle. The main results of this work find broad applications in multi-scale models arising in fields such as machine learning and optimization theory. In particular, our results can be employed to analyze the dynamics of multi-scale consensus-based methods for multilevel optimization, where the coefficients typically satisfy local Lipschitz continuity on the interaction kernels.

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 / 1 minor

Summary. The paper establishes the small-noise functional law of large numbers (with explicit rates) and large deviation principle for multi-scale McKean-Vlasov diffusions whose interaction kernels are super-linear and depend on the joint laws of slow and fast components. The analysis proceeds via a lifted semigroup construction combined with a generalized Khasminskii discretization for the LLN, and via lifted viable pairs together with a generalized functional occupation-measure argument to obtain the Laplace principle (hence the LDP). The setting is motivated by applications to multi-scale consensus-based optimization methods.

Significance. If the well-posedness and regularity hypotheses can be verified, the results would constitute a non-trivial extension of existing small-noise asymptotics for McKean-Vlasov systems to the super-linear regime and to genuinely nonlinear frozen processes. The explicit rates and the occupation-measure approach for the LDP are potentially useful for analyzing multi-scale algorithms in machine learning and optimization.

major comments (2)
  1. [Abstract (and presumed §2–3 on the frozen process)] The central claims rest on the well-posedness of the frozen parameterized McKean-Vlasov process for each fixed slow variable. The abstract invokes only local Lipschitz continuity of the kernels, yet super-linear growth precludes the standard global-Lipschitz existence/uniqueness theorems. No a priori moment bounds, uniqueness argument, or Feller property for the lifted semigroup are indicated; without these, both the functional LLN with rates and the viable-pair compactness for the LDP are unsupported.
  2. [Abstract (and presumed §4–5 on the LDP)] The generalized Khasminskii discretization and the lifted viable-pair construction are asserted to yield explicit convergence rates and the Laplace principle, respectively. However, the passage from the occupation-measure tightness to the identification of the limiting rate functional appears to require uniform integrability or exponential-moment controls that are not obviously available under super-linear growth; a concrete estimate or truncation argument is needed.
minor comments (1)
  1. [Abstract] The abstract states that the interaction depends on the laws of both components, but the precise form of the multi-scale scaling (e.g., the relative speed of the fast process) is not written explicitly; this should be displayed as an equation early in the introduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough and constructive report. The comments identify key points that require clarification regarding well-posedness and the passage to the rate functional. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract (and presumed §2–3 on the frozen process)] The central claims rest on the well-posedness of the frozen parameterized McKean-Vlasov process for each fixed slow variable. The abstract invokes only local Lipschitz continuity of the kernels, yet super-linear growth precludes the standard global-Lipschitz existence/uniqueness theorems. No a priori moment bounds, uniqueness argument, or Feller property for the lifted semigroup are indicated; without these, both the functional LLN with rates and the viable-pair compactness for the LDP are unsupported.

    Authors: We agree that explicit well-posedness statements are essential. The full manuscript (Section 2) establishes existence and uniqueness for the frozen McKean-Vlasov process under local Lipschitz continuity combined with super-linear growth via a truncation procedure that yields uniform moment bounds (Proposition 2.4). Uniqueness of the lifted process follows from a pathwise argument, and the Feller property of the lifted semigroup is proved in Theorem 2.7 using these moment controls. These results underpin both the LLN and the viable-pair compactness. We will add a brief statement of these facts to the abstract and the opening of Section 1 to make the dependence explicit. revision: partial

  2. Referee: [Abstract (and presumed §4–5 on the LDP)] The generalized Khasminskii discretization and the lifted viable-pair construction are asserted to yield explicit convergence rates and the Laplace principle, respectively. However, the passage from the occupation-measure tightness to the identification of the limiting rate functional appears to require uniform integrability or exponential-moment controls that are not obviously available under super-linear growth; a concrete estimate or truncation argument is needed.

    Authors: We concur that the identification step must be justified carefully. In Section 5 we apply a truncation argument (Lemma 5.2) that exploits the moment bounds already obtained for the frozen process to secure uniform integrability of the occupation measures. The limiting rate functional is then identified by lower semicontinuity and the characterization of lifted viable pairs. We will expand the proof of the Laplace principle to include an explicit statement of this truncation estimate and its consequences for exponential moments. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on explicit semigroup constructions and occupation measures from stated assumptions.

full rationale

The paper derives functional LLN and LDP for the slow variable via lifted semigroup arguments and generalized viable-pair occupation measures applied to the frozen nonlinear McKean-Vlasov process. These steps are constructed from the well-posedness assumption (local Lipschitz kernels plus moment bounds) rather than reducing any limit result to a fitted parameter or self-referential definition. No self-citation chain is load-bearing for the core asymptotics, and the lifted viable pair is introduced as a technical device to handle the multi-scale interaction without renaming a known empirical pattern. The derivation chain remains independent of its inputs once the frozen process is granted.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard well-posedness assumptions for McKean-Vlasov SDEs plus the new technical notions of lifted semigroup and lifted viable pair introduced in the paper; no free parameters or invented physical entities are stated.

axioms (2)
  • domain assumption Existence and uniqueness of solutions to the multi-scale McKean-Vlasov SDEs under the stated kernel conditions.
    Necessary before any asymptotic analysis can begin.
  • domain assumption The frozen parameterized system forms a nonlinear Markov process.
    Directly stated in the abstract as the source of added complexity.

pith-pipeline@v0.9.0 · 5518 in / 1315 out tokens · 47072 ms · 2026-05-08T10:17:18.776348+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

64 extracted references · 64 canonical work pages

  1. [1]

    Bayraktar, Z

    E. Bayraktar, Z. Ding, I. Ekren and H. Zhou.Uniform-in-time propagation of chaos for consensus-based minimax algorithm. arXiv:2602.14403

    work page arXiv

  2. [2]

    Z. W. Bezemek, and K. Spiliopoulos.Rate of homogenization for fully-coupled McKean–Vlasov SDEs. Stoch. Dyn. 23, Paper No. 2350013, 65 pp, 2023

    work page 2023

  3. [3]

    Z. W. Bezemek and K. Spiliopoulos.Large deviations for interacting multiscale particle systems. Stochastic Process. Appl. 155, 27–108, 2023

    work page 2023

  4. [4]

    Billingsley.Convergence of Probability Measures

    P. Billingsley.Convergence of Probability Measures. Second edition. Wiley Series in Probability and Statistics. John Wiley & Sons, Inc., New York, 1999

    work page 1999

  5. [5]

    G. D. Birkhoff,Proof of the ergodic theorem.Proceedings of the National Academy of Sciences, 17, 656–660, 1931

    work page 1931

  6. [6]

    N. N. Bogoliubov and Y. A. Mitropolsky.Asymtotic Methods in the Theory of Non-linear Oscillations. Hindustan, 1961

    work page 1961

  7. [7]

    Bou´ e and P

    M. Bou´ e and P. Dupuis.A variational representation for certain functionals of Brownian motion. Ann. Probab. 26(4), 1641–1659, 1998

    work page 1998

  8. [8]

    V. S. Borkar and V. R. Konda.The actor-critic algorithm as multi-time-scale stochastic approximation. Sadhana 22(4), 525–543, 1997

    work page 1997

  9. [9]

    Brehier.Strong and weak orders in averaging for SPDEs

    C.-E. Brehier.Strong and weak orders in averaging for SPDEs. Stochastic Process. Appl. 122, 2553–2593, 2012

    work page 2012

  10. [10]

    Brehier.Orders of convergence in the averaging principle for SPDEs: the case of a stochastically forced slow component

    C.-E. Brehier.Orders of convergence in the averaging principle for SPDEs: the case of a stochastically forced slow component. Stochastic Process. Appl. 130, 3325–3368, 2020

    work page 2020

  11. [11]

    Budhiraja and P

    A. Budhiraja and P. Dupuis.A variational representation for positive functionals of infinite dimensional Brownian motion. Probab. Math. Statist. 20(1), 39–61, 2000

    work page 2000

  12. [12]

    Budhiraja and P

    A. Budhiraja and P. Dupuis.Analysis and Approximation of Rare Events. Probability Theory and Stochastic Modelling, Vol. 94. Springer, New York, 2019

    work page 2019

  13. [13]

    Budhiraja, P

    A. Budhiraja, P. Dupuis and V. Maroulas.Large deviations for infinite dimensional stochastic dynamical systems. Ann. Probab. 36(4), 1390–1420, 2008

    work page 2008

  14. [14]

    J. A. Carrillo, Y.-P. Choi, C. Totzeck and O. Tse.An analytical framework for consensus-based global optimization method. Math. Model Methods Appl. Sci. 28(6), 1037–1066, 2018

    work page 2018

  15. [15]

    Cerrai.Normal deviations from the averaged motion for some reaction-diffusion equations with fast oscillating perturbation

    S. Cerrai.Normal deviations from the averaged motion for some reaction-diffusion equations with fast oscillating perturbation. J. Math. Pures Appl. 91, 614–647, 2009

    work page 2009

  16. [16]

    Cerrai and M

    S. Cerrai and M. Freidlin.Averaging principle for a class of stochastic reaction-diffusion equations. Probab. Theory Related Fields 144(1-2), 137–177, 2009

    work page 2009

  17. [17]

    Chaintron and A

    L.-P. Chaintron and A. Diez.Propagation of chaos: a review of models, methods and applications. I. Models and methods. Kinet. Relat. Models 15(6), 895–1015, 2022

    work page 2022

  18. [18]

    Cheng, Z

    M. Cheng, Z. Hao and M. R ¨ockner.Strong and weak convergence for the averaging principle of DDSDE with singular drift. Bernoulli 30(2), 1586–1610, 2024

    work page 2024

  19. [19]

    Delgadino, R.S

    M.G. Delgadino, R.S. Gvalani and G.A. Pavliotis.On the diffusive-mean field limit for weakly interacting diffusions exhibiting phase transitions. Arch. Ration. Mech. Anal. 241, 91–148, 2021

    work page 2021

  20. [20]

    Dembo and O

    A. Dembo and O. Zeitouni.Large Deviations Techniques and Applications. Second edition. Springer- Verlag, New York, 1998

    work page 1998

  21. [21]

    K. Du, Y. Jiang and J. Li.Empirical approximation to invariant measures for McKean–Vlasov processes: mean-field interaction vs self-interaction. Bernoulli 29, 2492–2518, 2023. 64

    work page 2023

  22. [22]

    Duncan, N

    A. Duncan, N. N ¨usken and L. Szpruch.On the geometry of Stein variational gradient descent. J. Mach. Learn. Res. 24, 1–39, 2023

    work page 2023

  23. [23]

    Dupuis and R

    P. Dupuis and R. S. Ellis.A Weak Convergence Approach to the Theory of Large Deviations. Wiley Series in Probability and Statistics. John Wiley & Sons, Inc., New York, 1997

    work page 1997

  24. [24]

    Dupuis and K

    P. Dupuis and K. Spiliopoulos.Large deviations for multiscale diffusion via weak convergence methods. Stochastic Process. Appl. 122, 1947–1987, 2012

    work page 1947

  25. [25]

    Garbuno-Inigo, F

    A. Garbuno-Inigo, F. Hoffmann, W. Li and A. M. Stuart.Interacting Langevin diffusions: gradient structure and ensemble Kalman sampler. SIAM J. Appl. Dyn. Syst. 19, 412–441, 2020

    work page 2020

  26. [26]

    Gasteratos, M

    I. Gasteratos, M. Salins and K. Spiliopoulos.Moderate deviations for systems of slow-fast stochastic reaction-diffusion equations. Stoch. Partial Differ. Equ. Anal. Comput. 11, 503–598, 2023

    work page 2023

  27. [27]

    N. J. Gerber, F. Hoffmann and U. Vaes.Mean-field limits for consensus-based optimization and sampling. arXiv:2312.07373, 2024

    work page arXiv 2024

  28. [28]

    N. J. Gerber, F. Hoffmann, D. Kim and U. Vaes.Uniform-in-time propagation of chaos for consensus- based optimization. arXiv:2505.08669, 2025

    work page arXiv 2025

  29. [29]

    B. Gess, R. S. Gvalani and S. Hu.Random dynamical systems for McKean–Vlasov SDEs via rough path theory. arXiv:2507.02449, 2025

    work page arXiv 2025

  30. [30]

    S.-Y. Ha, S. Jin and D. Kim.Convergence of a first-order consensus-based global optimization algorithm. Math. Models Methods Appl. Sci. 30(12), 2417–2444, 2020

    work page 2020

  31. [31]

    S.-Y. Ha, M. Kang, D. Kim, J. Kim and I. Yang.Stochastic consensus dynamics for nonconvex optimization on the Stiefel manifold: mean-field limit and convergence. Math. Models Methods Appl. Sci. 32(3), 533–617, 2022

    work page 2022

  32. [32]

    Hairer and X.-M

    M. Hairer and X.-M. Li.Averaging dynamics driven by fractional Brownian motion. Ann. Probab. 48, 1826–1860, 2020

    work page 2020

  33. [33]

    Herty, Y

    M. Herty, Y. Huang, D. Kalise and H. Kouhkouh.A multiscale consensus-based algorithm for multilevel optimization. Math. Models Methods Appl. Sci. 35, 2207–2243, 2025

    work page 2025

  34. [34]

    W. Hong, S. Hu and W. Liu.McKean–Vlasov SDE and SPDE with locally monotone coefficients. Ann. Appl. Probab. 34, 2136–2189, 2024

    work page 2024

  35. [35]

    W. Hong, G. Li and S. Li.Multi-scale McKean–Vlasov SDEs: moderate deviation principle in different regimes. to appear in Ann. Inst. Henri Poincar´ e Probab. Stat., arXiv:2306.11569, 2023

    work page arXiv 2023

  36. [36]

    W. Hong, S. Li and W. Liu.Strong convergence rates in averaging principle for slow-fast McKean–Vlasov SPDEs. J. Differential Equations 316, 94–135, 2022

    work page 2022

  37. [37]

    W. Hong, S. Li and W. Liu.McKean–Vlasov stochastic partial differential equations: existence, uniqueness and propagation of chaos. Probab. Theory Related Fields 193, 717–793, 2025

    work page 2025

  38. [38]

    W. Hong, S. Li and W. Liu.Mean field stochastic partial differential equations with nonlinear kernels. Ann. Appl. Probab. 36, 206–274, 2026

    work page 2026

  39. [39]

    W. Hong, S. Li, W. Liu and X. Sun.Central limit type theorem and large deviation principle for multi-scale McKean–Vlasov SDEs. Probab. Theory Related Fields 187, 133–201, 2023

    work page 2023

  40. [40]

    Y. Hou, Y. Li and L. Xie.Asymptotic limit of fully coupled multi-scale non-linear stochastic system: the non-autonomous approximation method. arXiv:2412.13430

    work page arXiv

  41. [41]

    W. Hu, M. Salins and K. Spiliopoulos.Large deviations and averaging for systems of slow-fast stochastic reaction-diffusion equations. Stoch. Partial Differ. Equ. Anal. Comput. 7, 808–874, 2019

    work page 2019

  42. [42]

    Huang, J

    H. Huang, J. Qiu and K. Riedl.Consensus-based optimization for saddle point problems. SIAM J. Control Optim. 62, 1093–1121, 2024

    work page 2024

  43. [43]

    Huang and F.-Y

    X. Huang and F.-Y. Wang.Distribution dependent SDEs with singular coefficients. Stochastic Process. Appl. 129, 4747–4770, 2019

    work page 2019

  44. [44]

    Jabin and Z

    P.-E. Jabin and Z. Wang.Quantitative estimates of propagation of chaos for stochastic systems with W −1,∞ kernels. Invent. Math. 214, 523–591, 2018

    work page 2018

  45. [45]

    Khas’minskii.The averaging principle for stochastic differential equations

    R. Khas’minskii.The averaging principle for stochastic differential equations. Problemy Peredachi Informatsii 4, 86–87, 1968

    work page 1968

  46. [46]

    Khas’minskii.Stochastic Stability of Differential Equations

    R. Khas’minskii.Stochastic Stability of Differential Equations. Second edition. Springer, Heidelberg, 2012

    work page 2012

  47. [47]

    M. Koß, S. Weissmann and J. Zech.On the mean-field limit of consensus-based methods. Math. Methods Appl. Sci. 2025

    work page 2025

  48. [48]

    Y. Li, F. Wu and L. Xie.Poisson equation on Wasserstein space and diffusion approximations for McKean–Vlasov equation. SIAM J. Math. Anal. 56, 1495–1524, 2024

    work page 2024

  49. [49]

    Li and L

    Y. Li and L. Xie.Functional law of large numbers and central limit theorem for slow-fast McKean–Vlasov equations. Discrete Contin. Dyn. Syst. Ser. S 16, 846–877, 2023

    work page 2023

  50. [50]

    Liu and M

    W. Liu and M. R ¨ockner.Stochastic Partial Differential Equations: An Introduction. Universitext. Springer, Cham, 2015

    work page 2015

  51. [51]

    Pinnau, C

    R. Pinnau, C. Totzeck, O. Tse and S. Martin.A consensus-based model for global optimization and its mean-field limit. Math. Models Methods Appl. Sci. 27, 183–204, 2017. 65

    work page 2017

  52. [52]

    P. Ren, M. R¨ockner and F.-Y. Wang.Linearization of nonlinear Fokker-Planck equations and applica- tions. J. Differential Equations 322, 1–37, 2022

    work page 2022

  53. [53]

    R¨ockner, X

    M. R¨ockner, X. Sun and Y. Xie.Strong convergence order for slow-fast McKean–Vlasov stochastic differential equations. Ann. Inst. Henri Poincar´ e Probab. Stat. 57, 547–576, 2021

    work page 2021

  54. [54]

    R¨ockner and L

    M. R¨ockner and L. Xie.Averaging principle and normal deviations for multiscale stochastic systems. Comm. Math. Phys. 383, 1889–1937, 2021

    work page 1937

  55. [55]

    Scheutzow.Uniqueness and nonuniqueness of solutions of Vlasov-McKean equations

    M. Scheutzow.Uniqueness and nonuniqueness of solutions of Vlasov-McKean equations. J. Aust. Math. Soc. Ser. A 43, 246–256, 1987

    work page 1987

  56. [56]

    Serfaty.Mean field limit for Coulomb-type flows

    S. Serfaty.Mean field limit for Coulomb-type flows. Duke Math. J. 169, 2887–2935, 2020

    work page 2020

  57. [57]

    G. Shen, J. Xiang and J.-L. Wu.Averaging principle for distribution dependent stochastic differential equations driven by fractional Brownian motion and standard Brownian motion. J. Differential Equations 321, 381–414, 2022

    work page 2022

  58. [58]

    G. Shen, H. Zhou and J.-L. Wu.Large deviation principle for multi-scale distribution dependent stochastic differential equations driven by fractional Brownian motions. Journal of Evolution Equations. 24, Paper No. 35, 30pp., 2024

    work page 2024

  59. [59]

    Sznitman.Topics in propagation of chaos

    A.-S. Sznitman.Topics in propagation of chaos. In ´Ecole d’´Et´ e de Probabilit´ es de Saint-Flour XIX—1989, Lecture Notes in Math. 1464, pp. 165–251. Springer, Berlin, 1991

    work page 1989

  60. [60]

    Vaes.Sharp propagation of chaos for the ensemble Langevin sampler

    U. Vaes.Sharp propagation of chaos for the ensemble Langevin sampler. J. London Math. Soc. 110, 34 pp., 2024

    work page 2024

  61. [61]

    Varadhan.Asymptotic probabilities and differential equations

    S.R.S. Varadhan.Asymptotic probabilities and differential equations. Comm. Pure Appl. Math. 19, 261–286, 1966

    work page 1966

  62. [62]

    Wang.Distribution dependent SDEs for Landau type equations

    F.-Y. Wang.Distribution dependent SDEs for Landau type equations. Stochastic Process. Appl. 128, 595–621, 2018

    work page 2018

  63. [63]

    H. Wu, J. Hu and C. Yuan.Large deviation for slow-fast McKean–Vlasov stochastic differential equations driven by fractional Brownian motions and Brownian motions. Stoch. Dyn., Paper No. 2450044., 2023

    work page 2023

  64. [64]

    Zhang, G

    Y. Zhang, G. Zhang, P. Khanduri, M. Hong, S. Chang and S. Liu.Revisiting and advancing fast adversarial training through the lens of bi-level optimization. Int. Conf. Machine Learning (PMLR), 26693–26712, 2022

    work page 2022