Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Exploring the Generalization Capabilities of AID-based Bi-level Optimization

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper shows AID-based bi-level optimizers have the same uniform-stability order as single-level nonconvex SGD, with diminishing step sizes beating constant ones at matched optimization error.

desk verdict Real stability first for AID-based bilevel methods, but the convergence proof has an unaccounted O(K) variance term that breaks the advertised generalization trade-off. read the letter →

arxiv 2411.16081 v1 pith:YO234LLM submitted 2024-11-25 cs.LG stat.ML

Congliang Chen , Li Shen , Zhiqiang Xu , Wei Liu , Zhi-Quan Luo , Peilin Zhao This is my paper · ORCID
classification cs.LGstat.ML MSC 68T0590C26
keywords bi-leveloptimizationapproximateimplicitdifferentiationuniformstabilitygeneralizationboundsstep-sizeschedulenonconvexSGDwithmomentum
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Bi-level optimization trains a model and its hyperparameters together, but the standard approximate implicit differentiation (AID) family cannot be collapsed into a single-level problem, so its generalization behavior has been open. This paper establishes that AID-based methods are uniformly stable: replacing one validation sample changes their output by at most $O(T^q/n)$ with an exponent $q<1$ under diminishing step sizes, the same sample order as nonconvex single-level SGD and ITD-based bi-level methods. Combined with a new convergence analysis for time-varying step sizes, the paper derives a generalization gap at $\epsilon$-accurate solutions and concludes that a diminishing learning rate beats a constant one at equal optimization error. The practical stakes are direct: memory-efficient implicit differentiation carries no theoretical generalization penalty, and decaying schedules are the safe choice for small validation sets, at the price of a slower $O(1/\log T)$ convergence in the pure decaying regime.

What carries the argument

The load-bearing object is a coupled-run stability recursion over the four iterate sequences of Algorithm 1 — the outer variable $x_t$, the inner variable $y_t$, the momentum $m_t$, and the inner-loop estimate $z^k_t$ — executed on two validation sets that differ in one sample. The key structural fact (Lemmas 8–11) is that the $z$-difference at step $t$ depends only on the previous $(x,y)$-difference and not on its own history, so the $z$-loop can be eliminated; the remaining differences then contract under a product of factors $1+\eta_{xt}\eta_{mt}C_m+\eta_{mt}C_m+\eta_{yt}L_1$, whose telescoping yields the $O(T^q/n)$ stability order. On the convergence side, two descent lemmas — one bounding $\|y_t-y^*(x_t)\|^2$ and one bounding the momentum/outer potential — telescope provided $\eta_{mt}$, $\eta_{mt}/\eta_{xt}$, and $\eta_{mt}/\eta_{yt}$ are non-increasing, which is precisely the diminishing-step-size regime the stability analysis favors.

What would settle it

On any bi-level problem that satisfies the assumptions (the ridge-regression toy example in the paper qualifies), fix $T$, run Algorithm 1 once per validation set for two validation sets of size $n$ that differ in a single sample, and measure the expected output difference: the theorem predicts a gap of order $T^q/n$, so a gap that fails to shrink with $n$, or that grows faster than $T^q$ with iterations, refutes Theorem IV.7. A sharper probe is to shrink the strong-convexity parameter $\mu$ toward zero, since $C_m$ and the exponent $q$ then blow up and the bound diverges: a problem in that regime where replacing one validation sample still moves the output by an $n$-independent amount would falsify the claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that the uniform stability of AID-based bi-level optimization matches the classical nonconvex single-level benchmark despite the two-level structure. The obstruction is that in Algorithm 1 the inner variable $y_t$, the momentum $m_t$, and the implicit-gradient estimate $z^k_t$ all depend on the full past of the outer variable $x_t$, so a one-sample change in the validation set propagates through three coupled sequences. By tracking all of these differences between two runs, the paper proves $\epsilon_{\mathrm{stab}} = O(T^q/n)$ with $q=(2C_m\alpha+L_1\beta)/(2C_m\alpha+L_1\beta+1)<1$ for $\eta_{xt}=\eta_{mt}=\alpha/t$ and $\eta_{yt}=\beta/t$ (Corollary IV.8), and it gives a convergence theorem for time-varying step sizes that yields $O(1/\sqrt{T})$ accuracy for $\Theta(1/\sqrt{T})$ rates. Joining the two results, the paper shows that at a fixed optimization accuracy $\epsilon$ the log-stability of the diminishing-rate schedule grows as $O(1/\epsilon)$, while the constant-rate schedule can only reach an exponential level, and it concludes that the diminishing schedule attains a smaller generalization gap at the same optimization error.

Load-bearing premise

The inner-level function $g(x,\cdot,\zeta)$ must be $\mu$-strongly convex in $y$ with Lipschitz gradients and Lipschitz second derivatives, so that the optimal-response map $y^*(x)$ is Lipschitz and the implicit-gradient estimate stays bounded; the image-classification experiments in the paper do not satisfy this curvature condition.

Editorial extensions

If this is right

  • AID-based bi-level methods obtain a generalization gap of the same sample order ($O(T^q/n)$) as ITD-based methods and single-level nonconvex SGD, so the memory-efficient implicit-differentiation family carries no theoretical generalization penalty.
  • With $\Theta(1/t)$ step sizes the stability bound is $O(T^q/n)$ with $q<1$, yielding $\log \epsilon_{\mathrm{stab}} = O(1/\epsilon)$ at $\epsilon$-accuracy; with $\Theta(1/\sqrt{T})$ step sizes the convergence is $O(1/\sqrt{T})$ while stability stays bounded, so both regimes are usable.
  • At the same optimization error, a diminishing learning rate produces a smaller generalization gap than a constant one across the schedules analyzed, making decaying step sizes the theoretically preferred choice for AID-based bi-level training.
  • For constant step sizes the achievable log-stability at $\epsilon$-accuracy can only reach an exponential-in-$1/\epsilon$ level as $T$ grows, so the diminishing schedule is strictly better in the regime where both attain the same error.
  • The convergence result holds for any monotone non-increasing schedule with $\eta_{mt}/\eta_{xt}$ and $\eta_{mt}/\eta_{yt}$ non-increasing and $K=\Theta(\log T)$ inner-loop steps, so the analysis covers whole families of schedules, not a single tuned configuration.

Reading between the lines

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

  • An editorial extension: since the stability proof uses no convexity of the outer level and only the inner problem's strong convexity, the same $O(T^q/n)$ order should carry over to AID variants with conjugate-gradient or variance-reduced hypergradient estimation; the paper does not state this, but nothing in Lemmas 8–11 blocks it.
  • An editorial extension: the image-classification experiments violate the theorem's strong-convexity requirement on the inner problem, so the empirical agreement supports a conjecture the paper leaves implicit — that the stability order survives under a much weaker curvature condition, such as a gradient-dominance inequality.
  • An editorial extension: the theory suggests a concrete rule for hyperparameter tuning with small validation sets — use the fastest decaying schedule that still meets the optimization target, since the stability gap shrinks like $1/n$ while the iteration cost grows only as $T^q$ with $q<1$.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies the generalization behavior of AID-based bi-level optimization. It proves a uniform stability bound for Algorithm 1 under Lipschitz and strong-convexity assumptions, obtaining ε_stab = O(T^q/n) with q < 1 for diminishing step sizes. It then presents a convergence analysis for time-varying step sizes and combines stability with convergence to claim generalization-gap trade-offs, arguing that diminishing rates give a smaller gap than constant rates at the same optimization error. The paper includes experiments on a toy transfer-learning problem, MNIST data selection, and Fashion-MNIST data mixture.

Significance. If the main claims are established, the stability analysis is a valuable contribution: AID-based methods have previously been difficult to analyze because the coupled sequences (x, y, m, z) prevent a direct reduction to single-level optimization, and the paper's induction over four coupled sequences is nontrivial. Matching the O(T^q/n) sample order of ITD-based and single-level nonconvex SGD would be a noteworthy result. The paper also addresses a timely question, since generalization guarantees for AID-based bi-level methods are largely missing in the literature. The strengths are the formal stability framework and the explicit treatment of the momentum and inner-loop sequences; the main weakness, detailed below, is that the convergence proof contains an unhandled variance term that is load-bearing for the advertised rates.

major comments (4)
  1. [Appendix B, proof of Theorem IV.11, Lemma 16] Lemma 16 upper-bounds the variance E||Δ_t − EΔ_t||^2 by an O(K) expression. In the final display of the proof of Theorem IV.11, the term Σ_t 2η_mt^2 times this bound is present, and it is then dropped with the sentence "when K is large (1 − η_z)^{2K} will small". With η_mt = α/t and K = Θ(log T), we have Σ_t η_mt^2 = Θ(1), so the dropped term is Θ(K) = Θ(log T). After division by Σ_t η_mt = Θ(log T), it remains Θ(1), not o(1). The stated rate min_t E||∇Φ(x_t)||^2 = O(1/log T) therefore does not follow from the displayed inequalities, and Corollaries IV.14 and IV.16 inherit this gap. The proof needs either a sharper, summable variance bound, a different choice of K or step sizes that makes the term vanish, or an explicit revision of the claimed rate.
  2. [Section III-B and Section IV-D, Corollaries IV.14 and IV.16] The generalization error is decomposed into four terms, with term (II) identified as an optimization error. However, no bound is provided for term (II): the convergence theorem controls min_t E||∇Φ(x_t)||^2, which is a gradient-norm measure, and the corollaries translate this only into a statement about the stability bound log ε_stab. Since Φ is nonconvex, a small gradient norm does not imply a small function-value gap, so the gap between the empirical training error and the best-in-class error (term (II)) is not controlled. The claimed trade-off between "optimization error" and "generalization gap" is therefore not established for the total generalization error defined in Section III-B. The authors should either restrict their claims to the stability term (term (I)) or add an assumption such as a Polyak-Łojasiewicz condition that connects gradient norm to function suboptimality.
  3. [Theorem IV.7 and Lemma 9 (Appendix A)] The proof of Lemma 9 uses the condition μη_yt ≥ L1^2 η_yt^2 (equivalently η_yt ≤ μ/L1^2) to obtain the contraction (1 − μη_yt/2) for the y-sequence. This condition is not stated in Theorem IV.7 or Corollary IV.8, which only require η_z ≤ 1/L1. Without an explicit upper bound on η_yt, the coefficient (1 − μη_yt/2) may not be a contraction, and the O(T^q/n) stability bound is not proven as stated. The theorem and corollary should include the implied step-size restriction, or the proof should be revised to avoid this condition.
  4. [Sections V-B and V-C, Assumption IV.2] The MNIST and Fashion-MNIST experiments train a LeNet5 network with cross-entropy loss as the lower-level problem. The lower-level objective is then neither strongly convex in the network parameters nor does it have uniformly Lipschitz second derivatives, so Assumption IV.2, which is used in the stability proof (Lemmas 7, 8, 9) and in the definition of the hypergradient, is not satisfied. The experiments therefore do not directly corroborate the theoretical results. They should be explicitly described as outside the scope of the theory, or the theory should be extended to non-strongly-convex lower-level problems.
minor comments (5)
  1. [Throughout] There are several typos: "Initilize" should be "Initialize" in Algorithm 1; "Baot et al." should be "Bao et al." in Section I; "Hard et al." should be "Hardt et al." in Section II; "wih" should be "with" in Assumption IV.1; "vairable" should be "variable" in Section IV; "tunned" should be "tuned" in Remark IV.6.
  2. [Lemma 3 and proof of Theorem IV.7] The expression "∥mt−1 + ˜mt−1∥" appears twice and should read "∥mt−1 − ˜mt−1∥"; otherwise the induction is bounding a sum of two independent norms rather than the difference.
  3. [Theorem IV.11] The theorem uses the quantity D2 without defining it. From the proof, the intended quantity appears to be Dz or possibly L2; the notation should be made consistent.
  4. [Lemma 16] In the final bound, the term "2KL^2_0" is missing its denominator "/µ^2" in the typeset formula; the displayed inequality is otherwise dimensionally inconsistent.
  5. [Corollary IV.16] The proof states that "By taking the derivative, the function value is decreasing while T increases," but the derivative is not shown and T is discrete; a short monotonicity argument would make the step more transparent.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the stability and convergence claims are derived from stated assumptions and external prior results, not defined in terms of their conclusions.

full rationale

The paper's derivation chain is self-contained rather than circular. Theorem IV.7 and Corollary IV.8 establish uniform stability from the Lipschitz, strong-convexity, and smoothness assumptions (Assumptions IV.1 and IV.2) via recursive perturbation bounds over the coupled x, y, m, and z sequences; no fitted parameter is renamed as a stability prediction. Theorem IV.11 is derived from descent lemmas, a variance-growth assumption (Assumption IV.10), and explicit step-size conditions, with the proof sketched in Section IV-E and detailed in Appendix B; the reference to Chen et al. [33] is contextual and not load-bearing because the paper supplies its own lemmas and proof. The diminishing-versus-constant generalization comparison is a mathematical consequence of the paper's proven upper bounds, not a self-defined target: Corollary IV.14 and Corollary IV.16 combine the convergence rate with the stability bound and compare the resulting trade-offs. The reviewer-flagged O(K) variance term in Lemma 16 that appears to be dropped in the final display of Theorem IV.11 is a potential correctness gap in the convergence proof, but it is not circularity: the claimed rate does not reduce to an input by construction, and the issue is whether the displayed inequalities imply O(1/log T), not whether the theorem restates an assumption. Similarly, the MNIST/Fashion-MNIST experiments using LeNet5 with cross-entropy loss do not satisfy the strong-convexity assumption of Assumption IV.2, but that is a scope mismatch rather than a circular derivation. Because no prediction is fitted to the data it claims to predict, and no load-bearing argument depends on an unverified self-citation, the circularity score is 0.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical or algorithmic entities. Its free parameters are schedule constants and an inner-loop count. The main assumptions are standard Lipschitz and strong-convexity conditions, plus a generalized variance assumption. The strong-convexity assumption is especially fragile because the empirical neural-network experiments violate it.

free parameters (3)
  • alpha and beta in diminishing step-size schedules = eta_xt = eta_mt = alpha/t, eta_yt = beta/t
    Analysis parameters governing the step-size schedules; the stability exponent q depends on them. They are chosen by hand and not fitted to data.
  • Inner-loop step size eta_z = eta_z <= 1/L_1
    Required for the boundedness of z iterates in Lemma 7; a constraint, not a fitted value.
  • Inner-loop iteration count K = K = Theta(log T)
    Set to balance the hypergradient estimation error; the convergence proof requires K to grow with T, but the final rate does not track K-dependent variance terms.
assumptions (6)
  • domain assumption Assumption IV.1: f is lower bounded, L_0-Lipschitz, and has L_1-Lipschitz gradients in both arguments.
    Used throughout the stability argument to bound the effect of replacing one validation sample.
  • domain assumption Assumption IV.2: g is mu-strongly convex in y, has L_1-Lipschitz gradients, and has L_2-Lipschitz second derivatives.
    Provides the contraction factors in Lemmas 7 and 9 and the hypergradient formula in Proposition III.1.
  • domain assumption Assumption IV.10: the stochastic gradient of g has variance bounded by D_1 times the squared gradient norm plus D_0.
    Generalized variance condition used in Lemma 19 to control the y-subproblem error; the constants D_0 and D_1 are not estimated.
  • standard math Proposition III.2 from Hardt et al.: uniform stability in the output implies an upper bound on the generalization gap.
    Bridges the beta-stability definition of Definition IV.3 to the generalization decomposition in Section III-B.
  • standard math Implicit Function Theorem for the hypergradient representation in Proposition III.1.
    Gives the exact gradient of Phi(x) under strong convexity and smoothness of g.
  • domain assumption Normalization f in [0,1] in Corollary IV.8.
    Used to convert output stability into a function-value stability bound; the main stability theorem does not need it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploring the Generalization Capabilities of AID-based Bi-level Optimization." pith.science (2026). https://pith.science/paper/YO234LLM

@misc{pith2026241116081,
  author       = {Pith},
  title        = {Pith review of: Exploring the Generalization Capabilities of AID-based Bi-level Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YO234LLM}},
  note         = {Machine review of arXiv:2411.16081}
}
read the original abstract

Bi-level optimization has achieved considerable success in contemporary machine learning applications, especially for given proper hyperparameters. However, due to the two-level optimization structure, commonly, researchers focus on two types of bi-level optimization methods: approximate implicit differentiation (AID)-based and iterative differentiation (ITD)-based approaches. ITD-based methods can be readily transformed into single-level optimization problems, facilitating the study of their generalization capabilities. In contrast, AID-based methods cannot be easily transformed similarly but must stay in the two-level structure, leaving their generalization properties enigmatic. In this paper, although the outer-level function is nonconvex, we ascertain the uniform stability of AID-based methods, which achieves similar results to a single-level nonconvex problem. We conduct a convergence analysis for a carefully chosen step size to maintain stability. Combining the convergence and stability results, we give the generalization ability of AID-based bi-level optimization methods. Furthermore, we carry out an ablation study of the parameters and assess the performance of these methods on real-world tasks. Our experimental results corroborate the theoretical findings, demonstrating the effectiveness and potential applications of these methods.

Figures

Figures reproduced from arXiv: 2411.16081 by the authors.

Figure 1
Figure 1. Results for Toy Example. The left figure shows the results when learning rates are constant, the middle figure shows the results when we use [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Results for Data selection on MNIST. The first figure shows the result with constant learning rates. The second figure shows the results with diminishing [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Results for Dataset Mixture on Fashion MNIST. The figure shows the [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 35 canonical work pages

  1. [1]

    Stability and generalization of bilevel programming in hyperpa- rameter optimization,

    F. Bao, G. Wu, C. Li, J. Zhu, and B. Zhang, “Stability and generalization of bilevel programming in hyperpa- rameter optimization,” Advances in Neural Information Processing Systems, vol. 34, pp. 4529–4541, 2021

    work page 2021

  2. [2]

    Gradient-based optimization of hyperparam- eters,

    Y . Bengio, “Gradient-based optimization of hyperparam- eters,” Neural computation , vol. 12, no. 8, pp. 1889– 1900, 2000

    work page 1900

  3. [3]

    Forward and reverse gradient-based hyperparameter optimization,

    L. Franceschi, M. Donini, P. Frasconi, and M. Pon- til, “Forward and reverse gradient-based hyperparameter optimization,” in International Conference on Machine Learning. PMLR, 2017, pp. 1165–1173

    work page 2017

  4. [4]

    Bilevel programming for hyperparameter optimization and meta-learning,

    L. Franceschi, P. Frasconi, S. Salzo, R. Grazzi, and M. Pontil, “Bilevel programming for hyperparameter optimization and meta-learning,” in International Con- ference on Machine Learning . PMLR, 2018, pp. 1568– 1577

    work page 2018

  5. [5]

    Optimizing millions of hyperparameters by implicit differentiation,

    J. Lorraine, P. Vicol, and D. Duvenaud, “Optimizing millions of hyperparameters by implicit differentiation,” in International conference on artificial intelligence and statistics. PMLR, 2020, pp. 1540–1552

    work page 2020

  6. [6]

    On the iteration complexity of hypergradient computation,

    R. Grazzi, L. Franceschi, M. Pontil, and S. Salzo, “On the iteration complexity of hypergradient computation,” in International Conference on Machine Learning. PMLR, 2020, pp. 3748–3758

    work page 2020

  7. [7]

    Implicit differentiation of lasso-type models for hyperparameter optimization,

    Q. Bertrand, Q. Klopfenstein, M. Blondel, S. Vaiter, A. Gramfort, and J. Salmon, “Implicit differentiation of lasso-type models for hyperparameter optimization,” in International Conference on Machine Learning. PMLR, 2020, pp. 810–821

    work page 2020

  8. [8]

    Meta-learning with differentiable closed-form solvers,

    L. Bertinetto, J. F. Henriques, P. H. Torr, and A. Vedaldi, “Meta-learning with differentiable closed-form solvers,” arXiv preprint arXiv:1805.08136 , 2018

    arXiv 2018

Show all 49 references
  1. [9]

    Convergence of meta-learning with task-specific adaptation over partial parameters,

    K. Ji, J. D. Lee, Y . Liang, and H. V . Poor, “Convergence of meta-learning with task-specific adaptation over partial parameters,” Advances in Neural Information Processing Systems, vol. 33, pp. 11 490–11 500, 2020

    2020

  2. [10]

    Model-agnostic meta-learning for fast adaptation of deep networks,

    C. Finn, P. Abbeel, and S. Levine, “Model-agnostic meta-learning for fast adaptation of deep networks,” in International conference on machine learning . PMLR, 2017, pp. 1126–1135

    2017

  3. [11]

    Meta-learning with implicit gradients,

    A. Rajeswaran, C. Finn, S. M. Kakade, and S. Levine, “Meta-learning with implicit gradients,” Advances in neural information processing systems , vol. 32, 2019

    2019

  4. [12]

    Autoaugment: Learning augmentation strate- gies from data,

    E. D. Cubuk, B. Zoph, D. Mane, V . Vasudevan, and Q. V . Le, “Autoaugment: Learning augmentation strate- gies from data,” in Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition , 2019, pp. 113–123

    2019

  5. [13]

    Cadda: Class-wise automatic differentiable data augmentation for eeg signals,

    C. Rommel, T. Moreau, J. Paillard, and A. Gramfort, “Cadda: Class-wise automatic differentiable data augmentation for eeg signals,” arXiv preprint arXiv:2106.13695, 2021

    2021 arXiv

  6. [14]

    Darts: Differentiable architecture search,

    H. Liu, K. Simonyan, and Y . Yang, “Darts: Differentiable architecture search,” arXiv preprint arXiv:1806.09055 , 2018

    2018 arXiv

  7. [15]

    Deep bilevel learning,

    S. Jenni and P. Favaro, “Deep bilevel learning,” in Proceedings of the European conference on computer vision (ECCV), 2018, pp. 618–633

    2018

  8. [16]

    Automatic design of cnns via differentiable neural architecture search for polsar image classification,

    H. Dong, B. Zou, L. Zhang, and S. Zhang, “Automatic design of cnns via differentiable neural architecture search for polsar image classification,” IEEE Transac- tions on Geoscience and Remote Sensing , vol. 58, no. 9, pp. 6362–6375, 2020

    2020

  9. [17]

    Advancing model pruning via bi-level optimization,

    Y . Zhang, Y . Yao, P. Ram, P. Zhao, T. Chen, M. Hong, Y . Wang, and S. Liu, “Advancing model pruning via bi-level optimization,” Advances in Neural Information Processing Systems, vol. 35, pp. 18 309–18 326, 2022

    2022

  10. [18]

    Anti-makeup: Learning a bi-level adversarial network for makeup- invariant face verification,

    Y . Li, L. Song, X. Wu, R. He, and T. Tan, “Anti-makeup: Learning a bi-level adversarial network for makeup- invariant face verification,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1, 2018

    2018

  11. [19]

    Connecting generative adver- sarial networks and actor-critic methods,

    D. Pfau and O. Vinyals, “Connecting generative adver- sarial networks and actor-critic methods,” arXiv preprint arXiv:1610.01945, 2016

    2016 arXiv

  12. [20]

    On the global optimality of model-agnostic meta-learning,

    L. Wang, Q. Cai, Z. Yang, and Z. Wang, “On the global optimality of model-agnostic meta-learning,” in International conference on machine learning . PMLR, 2020, pp. 9837–9846

    2020

  13. [21]

    A two- timescale stochastic algorithm framework for bilevel op- timization: Complexity analysis and application to actor- critic,

    M. Hong, H.-T. Wai, Z. Wang, and Z. Yang, “A two- timescale stochastic algorithm framework for bilevel op- timization: Complexity analysis and application to actor- critic,” SIAM Journal on Optimization, vol. 33, no. 1, pp. 147–180, 2023

    2023

  14. [22]

    Randomized stochastic variance- reduced methods for stochastic bilevel optimization,

    Z. Guo and T. Yang, “Randomized stochastic variance- reduced methods for stochastic bilevel optimization,” JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 10 arXiv e-prints, pp. arXiv–2105, 2021

    2021

  15. [23]

    Approximation methods for bilevel programming,

    S. Ghadimi and M. Wang, “Approximation methods for bilevel programming,” arXiv preprint arXiv:1802.02246, 2018

    2018 arXiv

  16. [24]

    A two- timescale framework for bilevel optimization: Complex- ity analysis and application to actor-critic,

    M. Hong, H.-T. Wai, Z. Wang, and Z. Yang, “A two- timescale framework for bilevel optimization: Complex- ity analysis and application to actor-critic,”arXiv preprint arXiv:2007.05170, 2020

    2007 arXiv

  17. [25]

    Closing the gap: Tighter analysis of alternating stochastic gradient methods for bilevel problems,

    T. Chen, Y . Sun, and W. Yin, “Closing the gap: Tighter analysis of alternating stochastic gradient methods for bilevel problems,” Advances in Neural Information Pro- cessing Systems, vol. 34, pp. 25 294–25 307, 2021

    2021

  18. [26]

    Provably faster algorithms for bilevel optimization,

    J. Yang, K. Ji, and Y . Liang, “Provably faster algorithms for bilevel optimization,” Advances in Neural Informa- tion Processing Systems , vol. 34, 2021

    2021

  19. [27]

    A single-timescale stochastic bilevel optimization method,

    T. Chen, Y . Sun, and W. Yin, “A single-timescale stochastic bilevel optimization method,” arXiv preprint arXiv:2102.04671, 2021

    2021 arXiv

  20. [28]

    Bilevel optimization: Con- vergence analysis and enhanced design,

    K. Ji, J. Yang, and Y . Liang, “Bilevel optimization: Con- vergence analysis and enhanced design,” in International Conference on Machine Learning . PMLR, 2021, pp. 4882–4892

    2021

  21. [29]

    A framework for bilevel optimization that enables stochas- tic and global variance reduction algorithms,

    M. Dagr ´eou, P. Ablin, S. Vaiter, and T. Moreau, “A framework for bilevel optimization that enables stochas- tic and global variance reduction algorithms,” Advances in Neural Information Processing Systems , vol. 35, pp. 26 698–26 710, 2022

    2022

  22. [30]

    Amortized implicit differenti- ation for stochastic bilevel optimization,

    M. Arbel and J. Mairal, “Amortized implicit differenti- ation for stochastic bilevel optimization,” arXiv preprint arXiv:2111.14580, 2021

    2021 arXiv

  23. [31]

    Projection-free stochastic bi-level optimization,

    Z. Akhtar, A. S. Bedi, S. T. Thomdapu, and K. Rajawat, “Projection-free stochastic bi-level optimization,” IEEE Transactions on Signal Processing , vol. 70, pp. 6332– 6347, 2022

    2022

  24. [32]

    Fednest: Federated bilevel, minimax, and com- positional optimization,

    D. A. Tarzanagh, M. Li, C. Thrampoulidis, and S. Oy- mak, “Fednest: Federated bilevel, minimax, and com- positional optimization,” in International Conference on Machine Learning. PMLR, 2022, pp. 21 146–21 179

    2022

  25. [33]

    Decentralized stochastic bilevel optimization with improved per-iteration complexity,

    X. Chen, M. Huang, S. Ma, and K. Balasubra- manian, “Decentralized stochastic bilevel optimization with improved per-iteration complexity,” arXiv preprint arXiv:2210.12839, 2022

    2022 arXiv

  26. [34]

    Stability and generaliza- tion,

    O. Bousquet and A. Elisseeff, “Stability and generaliza- tion,” The Journal of Machine Learning Research, vol. 2, pp. 499–526, 2002

    2002

  27. [35]

    Stability of randomized learning algorithms

    A. Elisseeff, T. Evgeniou, M. Pontil, and L. P. Kaelbing, “Stability of randomized learning algorithms.” Journal of Machine Learning Research , vol. 6, no. 1, 2005

    2005

  28. [36]

    Train faster, gener- alize better: Stability of stochastic gradient descent,

    M. Hardt, B. Recht, and Y . Singer, “Train faster, gener- alize better: Stability of stochastic gradient descent,” in International conference on machine learning . PMLR, 2016, pp. 1225–1234

    2016

  29. [37]

    Stability and convergence trade-off of iterative optimization algorithms,

    Y . Chen, C. Jin, and B. Yu, “Stability and convergence trade-off of iterative optimization algorithms,” arXiv preprint arXiv:1804.01619, 2018

    2018 arXiv

  30. [38]

    What is a good metric to study generalization of minimax learners?

    A. Ozdaglar, S. Pattathil, J. Zhang, and K. Zhang, “What is a good metric to study generalization of minimax learners?” arXiv preprint arXiv:2206.04502 , 2022

    2022 arXiv

  31. [39]

    Sta- bility analysis and generalization bounds of adversarial training,

    J. Xiao, Y . Fan, R. Sun, J. Wang, and Z.-Q. Luo, “Sta- bility analysis and generalization bounds of adversarial training,” arXiv preprint arXiv:2210.00960 , 2022

    2022 arXiv

  32. [40]

    Stability and gener- alization of the decentralized stochastic gradient descent ascent algorithm,

    M. Zhu, L. Shen, B. Du, and D. Tao, “Stability and gener- alization of the decentralized stochastic gradient descent ascent algorithm,” arXiv preprint arXiv:2310.20369 , 2023

    2023 arXiv

  33. [41]

    Stability and generalization of decentralized stochastic gradient descent,

    T. Sun, D. Li, and B. Wang, “Stability and generalization of decentralized stochastic gradient descent,” in Proceed- ings of the AAAI Conference on Artificial Intelligence , vol. 35, no. 11, 2021, pp. 9756–9764

    2021

  34. [42]

    Topology-aware generalization of decentralized sgd,

    T. Zhu, F. He, L. Zhang, Z. Niu, M. Song, and D. Tao, “Topology-aware generalization of decentralized sgd,” in International Conference on Machine Learning. PMLR, 2022, pp. 27 479–27 503

    2022

  35. [43]

    A closer look at the training strategy for modern meta-learning,

    J. Chen, X.-M. Wu, Y . Li, Q. Li, L.-M. Zhan, and F.- l. Chung, “A closer look at the training strategy for modern meta-learning,” Advances in Neural Information Processing Systems, vol. 33, pp. 396–406, 2020

    2020

  36. [44]

    Generalization of model-agnostic meta-learning algorithms: Recurring and unseen tasks,

    A. Fallah, A. Mokhtari, and A. Ozdaglar, “Generalization of model-agnostic meta-learning algorithms: Recurring and unseen tasks,” Advances in Neural Information Pro- cessing Systems, vol. 34, pp. 5469–5480, 2021

    2021

  37. [45]

    Will bilevel optimizers benefit from loops,

    K. Ji, M. Liu, Y . Liang, and L. Ying, “Will bilevel optimizers benefit from loops,” arXiv preprint arXiv:2205.14224, 2022

    2022 arXiv

  38. [46]

    Opti- mal algorithms for stochastic bilevel optimization un- der relaxed smoothness conditions,

    X. Chen, T. Xiao, and K. Balasubramanian, “Opti- mal algorithms for stochastic bilevel optimization un- der relaxed smoothness conditions,” arXiv preprint arXiv:2306.12067, 2023

    2023 arXiv

  39. [47]

    The mnist database of handwritten digit im- ages for machine learning research [best of the web],

    L. Deng, “The mnist database of handwritten digit im- ages for machine learning research [best of the web],” IEEE signal processing magazine , vol. 29, no. 6, pp. 141–142, 2012

    2012

  40. [48]

    Gradient-based learning applied to document recogni- tion,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recogni- tion,” Proceedings of the IEEE, vol. 86, no. 11, pp. 2278– 2324, 1998

    1998

  41. [49]

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,

    H. Xiao, K. Rasul, and R. V ollgraf, “Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,” arXiv preprint arXiv:1708.07747 , 2017. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 11 APPENDIX A. Proof of Theorem IV .7 Notation A.1. We us...

    2017 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.