Bilevel optimization is a fundamental framework for machine learning problems, where an upper-level (UL) objective depends on the solution of a nested lower-level (LL) problem. Existing bilevel algorithms typically assume that the LL objective is available for evaluation, differentiation, or unrolling. However, this assumption can fail when the LL response is produced by a black-box solver, simulator, adversary, or learned predictor. In this paper, we study LL agnostic bilevel optimization, where the UL learner has access only to an approximate LL solution yˉ(x) and cannot query the LL objective, its gradients, Hessian/Jacobian information, or the procedure that generates the LL response. We propose LeGo-BiO, a coordinate-wise pseudo-hypergradient method that estimates the missing Jacobian ∇xyˉ(x) using finite differences of yˉ(⋅). For each UL coordinate, LeGo-BiO constructs a decision vector subject to 1-sparse updates, thereby isolating the corresponding Jacobian column needed in the UL gradient computation. This avoids the directional-projection limitation of full-parameter response differences and, unlike derivative-free/zeroth-order approximations of the entire UL gradient, preserves the available first-order gradient information with respect to the UL variables for more accurate UL gradient evaluation. We establish convergence guarantees for LeGo-BiO in both deterministic and stochastic settings under an LL error-bound/Polyak-Lojasiewicz-type condition, without requiring LL strong convexity. Our bounds yield an O(T−1) rate in the deterministic setting and an O(T−1/2) rate in the stochastic setting when the LL approximation error decreases geometrically. Experiments demonstrate competitive performance of LeGo-BiO despite being agnostic to the LL objective.
Introduction
Bilevel optimization provides a natural framework for learning problems in which an upper-level (UL) decision depends on the solution of a nested lower-level (LL) problem. A standard bilevel optimization problem is formulated as
where f:Rdu×Rdl→R and g:Rdu×Rdl→R denote the UL and LL objectives, respectively. This framework has been widely adopted across a broad range of machine learning applications, including hyperparameter optimization [1, 2], meta-learning [3, 4], reinforcement learning [5, 6], neural architecture search [7, 8], adversarial training [9, 10, 11], data hyper-cleaning [2, 12], and dictionary learning [13, 14].
Most existing methods for solving Problem (1) are developed under the assumption that the LL objective is explicitly available and can be differentiated, queried, or optimized by the UL learner. However, this assumption may fail in many settings where the LL solution is produced by an external procedure whose objective or internal mechanism is inaccessible. For example, in adversarial training, the UL problem updates the model parameters, while the LL problem generates adversarial examples or perturbations against the current model. When this LL procedure is implemented through a black-box attack, simulator, or learned generator, the UL training algorithm may only observe the resulting adversarial examples, without direct access to the underlying LL objective, a differentiable description of the LL objective (i.e., gradient, Hessian, or Jacobian), or the internal optimization procedure that produce them.
Motivated by applications where the LL response is observable but the objective or mechanism generating it is unavailable to the UL learner, we study a new problem setting, which we call LL agnostic bilevel optimization. In this setting, the LL objective g still defines the target response y∗(x) in (1), but the UL algorithm does not have access to g, including its function values (zeroth-order information), gradients (first-order information), and Hessian/Jacobian information (second-order information). Instead, for a UL variable x, the algorithm only observes an approximate LL solution yˉ(x), which may be produced by a black-box solver, simulator, or learned predictor.
This LL-agnostic setting goes beyond the scope of existing bilevel optimization algorithms. When the LL problem is strongly convex, classical bilevel optimization methods update x using the gradient of the UL objective (UL-gradient), which can be computed using the chain rule as:
∇ℓ(x)=∇xf(x,y∗(x))+[∇xy∗(x)]T∇yf(x,y∗(x)).(2)
Approximate implicit differentiation (AID) methods [5, 15, 16, 17, 18, 19, 20, 21] estimate the Jacobian ∇xy∗(x) in Eq. (2) by applying the implicit function theorem [22]. This requires derivative information from the LL objective, i.e., ∇xy2g and ∇yy2g, to construct the resulting hypergradient. Iterative differentiation (ITD) methods [2, 12, 23, 24, 25, 26, 27] instead differentiate through the LL optimization trajectory, and therefore require access to the LL update rule or computational graph. Both classes of methods are incompatible with the LL-agnostic setting, where the UL algorithm cannot evaluate or differentiate the LL objective and cannot necessarily backpropagate through the procedure that returns yˉ(x).
Existing Hessian/Jacobian-free and derivative-free bilevel methods are also not directly applicable. Although these methods avoid second-order derivative computations, they assume an information model in which the LL objective is accessible in some form. Hessian/Jacobian-free first-order methods [28, 29, 30, 31, 32] avoid explicit second-order derivatives, but still rely on gradients of the LL objective or reformulations based on LL optimality conditions. Zeroth-order bilevel methods [33, 34, 35, 36] typically construct finite-difference or smoothing-based estimators using function-value queries of the UL and LL objectives. A different line of derivative-free work considers direct-search methods that assume access to a black-box LL oracle returning approximate LL solutions and optimize the resulting inexact bilevel objective directly through function-value comparisons [37]. Such methods avoid differentiating the LL problem, but treat the induced UL objective as a derivative-free black box rather than exploiting its available first-order structure. In contrast, the LL-agnostic setting provides no function-value or gradient access to g, while the UL objective f remains differentiable and accessible. Our goal is therefore not to treat the entire bilevel problem as a black box. Instead, we exploit the available gradients of f and use only the observable LL response yˉ(x) to approximate the missing sensitivity ∇xyˉ(x). Thus, the key difficulty is not merely the absence of Hessian/Jacobian information or derivative information from the induced bilevel objective, but the absence of a usable LL objective altogether.
To address this difficulty, we propose LeGo-BiO (Lower-level Agnostic Bilevel Optimization), a coordinate-wise pseudo-hypergradient method that uses the observable LL response map x↦yˉ(x) to estimate the missing Jacobian ∇xy∗(x) in Eq. (2). Once yˉ(x) is used as a tractable proxy for y∗(x), the UL-gradient in Eq. (2) can be approximated by replacing y∗(x) with yˉ(x). The remaining challenge is to estimate the Jacobian ∇xyˉ(x) without differentiating through g or through the mechanism that produces yˉ. LeGo-BiO addresses this challenge by updating one UL coordinate at a time. Specifically, for each UL coordinate, LeGo-BiO constructs a decision vector subject to 1-sparse updates, i.e., restricting modifications to a single coordinate per step, thereby isolating the corresponding Jacobian column needed in the UL gradient computation. This coordinate-wise Jacobian estimate is then combined with the UL gradients (i.e., ∇xf and ∇yf) to form a pseudo-hypergradient. As a result, LeGo-BiO requires only gradients of the UL objective and queries to the approximate LL solution map without needing g, ∇g, ∇xy2g, or ∇yy2g, making it applicable to LL-agnostic setting.
The coordinate-wise design is important for both statistical and algorithmic reasons. A finite difference based on a full-parameter change in x only provides one directional derivative of yˉ along the chosen displacement, and therefore cannot identify the entire Jacobian matrix information needed to form a full UL-gradient without additional perturbations or structural assumptions. In contrast, LeGo-BiO isolates one Jacobian column at a time, which aligns naturally with coordinate-wise UL updates and avoids mixing the effects of multiple coordinates in a single response difference. Moreover, LeGo-BiO uses finite differences only to estimate the missing Jacobian within the UL-gradient, while retaining the available first-order gradients of the UL objective. This distinguishes it from the zeroth-order approaches that approximate the entire UL-gradient through function-value differences of the composite UL objective f(x,yˉ(x)), which are typically more sensitive to noise and dimensionality [38].
Our main contributions are summarized as follows.
We formulate and analyze a LL agnostic bilevel optimization setting in which the LL objective is unavailable to the UL algorithm, and the only available LL-side information is an approximate solution yˉ(x). This information model distinguishes the problem from both classical gradient-based bilevel optimization and standard zeroth-order bilevel optimization.
We develop LeGo-BiO, a coordinate-wise pseudo-hypergradient algorithm that estimates the Jacobian of the approximate LL solution (i.e., ∇xyˉ in Eq. (2)) using only differences between LL responses. This method avoids differentiating through the LL objective or the procedure that generates yˉ(x), and is therefore applicable to the LL-agnostic setting.
We establish convergence guarantees for LeGo-BiO with nonconvex UL objectives in both deterministic and stochastic settings. The analysis does not require strong convexity of the LL objective (LLSC); instead, it relies on an LL error-bound/Polyak–Łojasiewicz-type condition. LeGo-BiO achieves an O(T−1) convergence rate in the deterministic setting and an O(T−1/2) rate in the stochastic setting when the LL approximation error decreases geometrically. Despite operating in the more challenging LL-agnostic setting, these rates match the state-of-the-art deterministic and stochastic convergence rates achieved by standard bilevel methods under the LLSC assumption.
We evaluate the numerical performance of LeGo-BiO on three tasks: deep hyper-representation, adversarial training, and data weighting for LLM alignment. The results show that, despite having no access to the LL objective, LeGo-BiO achieves competitive or superior performance compared with baselines that use LL objective information during training.
Related Work
In this section, we provide an overview on three closely related lines of research to place our work in comparative perspectives.
1) Gradient-based Bilevel Optimization: Gradient-based bilevel optimization methods typically compute or approximate the UL-gradient of the value function ℓ(x)=f(x,y∗(x)). Two major classes of methods are approximate implicit differentiation (AID) and iterative differentiation (ITD). AID-based methods use the implicit function theorem to approximate the implicit gradient of ℓ(x)[5, 15, 16, 17, 18, 19, 20, 21]. These methods generally require LL derivative information, such as ∇xy2g and ∇yy2g, or Hessian-vector and Jacobian-vector products derived from the LL objective. ITD-based methods instead approximate the UL-gradient by differentiating through the unrolled LL optimization trajectory [2, 12, 23, 24, 25, 26, 27]. As a result, ITD methods require access to the LL update rule or computational graph, since the UL-gradient is obtained by backpropagating through the LL solver.
These approaches are effective when the LL objective is available and differentiable, but they are inapplicable for the LL-agnostic setting studied in this paper. In our setting, the UL learner cannot query the LL objective g, compute ∇xy2g and ∇yy2g, form Hessian/Jacobian-vector products, or backpropagate through the procedure that produces the approximate LL solution yˉ(x). Therefore, neither AID nor ITD can be directly applied. In contrast, LeGo-BiO requires only gradients of the UL objective and queries to the observable LL response map x↦yˉ(x). This distinction is also reflected in the convergence analysis: many gradient-based bilevel methods rely on LL strong convexity to control the solution map and hypergradient approximation errors, whereas our analysis uses an LL error-bound condition together with an explicit approximation-error sequence for yˉ(x).
2) Hessian/Jacobian-free First-order Bilevel Optimization: A recent line of work develops Hessian/Jacobian-free first-order bilevel methods to reduce the computational burden of second-order hypergradient estimation. These methods avoid explicitly computing LL Hessian or Jacobian matrices and instead replace second-order quantities with first-order approximations, optimality-based constraints, or penalty/value-function reformulations. For example, [28] approximates second-order information using first-order gradient information. Other methods reformulate the bilevel problem through LL optimality conditions and penalty/value-function-based objectives so that the resulting algorithms only require first-order gradient information [29, 30, 31, 32].
Although these methods avoid explicit second-order derivatives, they still assume access to the LL objective in a usable form. In particular, they typically require LL gradients, LL optimality residuals, or penalty/value functions involving the LL objective. Hence, they remain incompatible with the LL-agnostic setting. The challenge in our setting is not merely that Hessian/Jacobian information is unavailable; rather, the LL objective itself is unavailable to the UL learner. LeGo-BiO therefore differs from Hessian/Jacobian-free first-order bilevel algorithms by estimating the missing Jacobian ∇xyˉ directly from differences of approximate LL solutions, without using g or its gradients.
3) Derivative-free/Zeroth-order Bilevel Optimization: Derivative-free/zeroth-order bilevel optimization studies settings where derivative information is unavailable and algorithms rely on function-value queries or direct evaluations of the induced bilevel objective. Several works develop finite-difference, evolution-strategy, or Gaussian-smoothing estimators for bilevel optimization. For example, evolution-strategy-based methods have been used to avoid second-order derivatives in meta-learning [33]. Gaussian-smoothing-based methods estimate derivatives of both UL and LL objectives from noisy function values [34, 35]. More recent single-loop zeroth-order bilevel methods further improve the efficiency of finite-difference UL-gradient approximation [36]. Beyond gradient-estimation-based approaches, derivative-free direct-search methods have also been studied for bilevel optimization. In particular, [37] develop inexact direct-search schemes that use a fixed-accuracy black-box LL solver and optimize the resulting approximate bilevel objective through function-value comparisons, without explicitly constructing gradient estimators.
Our method is related to this literature because it also uses finite differences, but both the information model and the role of function evaluations are different. Standard zeroth-order bilevel methods assume access to function values of the objectives, including the LL objective g, and use finite differences or smoothing to estimate gradients of the UL and LL objectives. Likewise, direct-search bilevel methods treat the induced UL objective as a derivative-free black box and make optimization decisions based on its function values. In contrast, the LL-agnostic setting provides no function-value access to g. At the same time, the UL objective f remains differentiable and accessible. Therefore, LeGo-BiO does not treat either level, or the induced bilevel objective, as a fully black-box optimization problem. Instead, it preserves the chain-rule structure of the UL-gradient by using ∇xf and ∇yf directly, while applying finite differences only to estimate the missing Jacobian ∇xyˉ(x) of the approximate LL solution. This targeted use of finite differences distinguishes LeGo-BiO from both gradient-estimation-based zeroth-order methods and direct-search-based derivative-free methods, and is central to its LL agnostic design. Since function-value-based zeroth-order estimators often incur additional sampling and dimension-dependent errors [38], retaining the available UL gradients in LeGo-BiO can lead to a more accurate and less noisy UL-gradient approximation.
The LeGo-BiO Algorithm
In this section, we propose LeGo-BiO, a coordinate-wise pseudo-hypergradient method for LL agnostic bilevel optimization. The main idea is to estimate the Jacobian of the observable LL response map x↦yˉ(x) by updating one UL coordinate at a time and measuring the resulting change in the approximate LL solution. This construction enables LeGo-BiO to approximate the missing Jacobian information of yˉ without accessing the LL objective g, its gradient ∇g, or its second-order derivatives, i.e., ∇xy2g and ∇yy2g.
The LeGo-BiO algorithm is summarized in Algorithm 1. For each UL coordinate, LeGo-BiO constructs a decision vector subject to 1-sparse updates, which restricts modifications to a single coordinate per step, thereby isolating the corresponding Jacobian column needed in the UL gradient computation. Specifically, at iteration t and for each coordinate j∈{1,…,du}, define the coordinate displacement as
stj≜xt−1j−xtj,xtj≜xt−1−stjej,(3)
where ej denotes the j-th canonical basis vector in Rdu. The point xtj differs from xt−1 only in the j-th coordinate. Therefore, the difference yˉ(xt−1)−yˉ(xtj) captures how the approximate LL response changes when only the j-th UL coordinate is updated. This provides a one-dimensional finite-difference estimate of the j-th column of the Jacobian ∇xyˉ(x):
stjyˉ(xt−1)−yˉ(xtj)≈∇xyˉ(xt)ej.
Throughout the paper, we assume that stj=0 to ensure that the finite-difference estimate is well-defined. This column-wise estimate is exactly the quantity needed for the j-th UL-gradient component, because
[1]F. Bao, G. Wu, C. Li, J. Zhu, and B. Zhang, “Stability and generalization of bilevel programming in hyperparameter optimization,” Advances in neural information processing systems, vol. 34, pp. 4529–4541, 2021.arxiv.org/abs/2106.04188
[2]L. Franceschi, P. Frasconi, S. Salzo, R. Grazzi, and M. Pontil, “Bilevel programming for hyperparameter optimization and meta-learning,” in International Conference on Machine Learning. PMLR, 2018, pp. 1568–1577.
[3]A. Rajeshwaran, C. Finn, S. M. Kakade, and S. Levine, “Meta-learning with implicit gradients,” Advances in neural information processing systems, vol. 32, 2019.arxiv.org/abs/1909.04630
[4]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.arxiv.org/abs/2006.09486
[5]M. Hong, H.-T. Wai, Z. Wang, and Z. Yang, “A two-timescale stochastic algorithm framework for bilevel optimization: Complexity analysis and application to actor-critic,” SIAM Journal on Optimization, vol. 33, no. 1, pp. 147–180, 2023.DOI
[6]H. Shen, Z. Yang, and T. Chen, “Principled penalty-based methods for bilevel reinforcement learning and rlhf,” Journal of Machine Learning Research, vol. 26, no. 114, pp. 1–49, 2025.DOI
[7]D. Lian, Y. Zheng, Y. Xu, Y. Lu, L. Lin, P. Zhao, J. Huang, and S. Gao, “Towards fast adaptation of neural architectures with meta learning,” in International Conference on Learning Representations, 2019.
[8]Y. Hu, X. Wu, and R. He, “Tf-nas: Rethinking three search freedoms of latency-constrained differentiable neural architecture search,” in European Conference on Computer Vision. Springer, 2020, pp. 123–139.arxiv.org/abs/2008.05314
[9]Y. Tian, L. Shen, G. Su, Z. Li, and W. Liu, “Alphagan: Fully differentiable architecture search for generative adversarial networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 10, pp. 6752–6766, 2021.DOI
[10]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,” in International Conference on Machine Learning. PMLR, 2022, pp. 26 693–26 712.
[11]Y. Li, L. Song, X. Wu, R. He, and T. Tan, “Learning a bi-level adversarial network with global and local perception for makeup-invariant face verification,” Pattern Recognition, vol. 90, pp. 99–108, 2019.
[12]A. Shaban, C.-A. Cheng, N. Hatch, and B. Boots, “Truncated back-propagation for bilevel optimization,” in The 22nd International Conference on Artificial Intelligence and Statistics. PMLR, 2019, pp. 1723–1732.arxiv.org/abs/1810.10667
[13]B. Lecouat, J. Ponce, and J. Mairal, “A flexible framework for designing trainable priors with adaptive smoothing and game encoding,” Advances in Neural Information Processing Systems, vol. 33, pp. 15 664–15 675, 2020.
[14]J. Mairal, F. Bach, and J. Ponce, “Task-driven dictionary learning,” IEEE transactions on pattern analysis and machine intelligence, vol. 34, no. 4, pp. 791–804, 2011.DOI
[15]S. Ghadimi and M. Wang, “Approximation methods for bilevel programming,” arXiv preprint arXiv:1802.02246, 2018.arxiv.org/abs/1802.02246
[16]K. Ji, J. Yang, and Y. Liang, “Bilevel optimization: Convergence analysis and enhanced design,” in International Conference on Machine Learning. PMLR, 2021, pp. 4882–4892.arxiv.org/abs/2010.07962
[17]T. Chen, Y. Sun, and W. Yin, “Closing the gap: Tighter analysis of alternating stochastic gradient methods for bilevel problems,” Advances in Neural Information Processing Systems, vol. 34, pp. 25 294–25 307, 2021.
[18]M. Arbel and J. Mairal, “Amortized implicit differentiation for stochastic bilevel optimization,” in International Conference on Learning Representations, 2022.arxiv.org/abs/2111.14580
[19]M. Dagréou, P. Ablin, S. Vaiter, and T. Moreau, “A framework for bilevel optimization that enables stochastic and global variance reduction algorithms,” Advances in Neural Information Processing Systems, vol. 35, pp. 26 698–26 710, 2022.arxiv.org/abs/2201.13409
[20]P. Khanduri, S. Zeng, M. Hong, H.-T. Wai, Z. Wang, and Z. Yang, “A near-optimal algorithm for stochastic bilevel optimization via double-momentum,” Advances in Neural Information Processing Systems, vol. 34, pp. 30 271–30 283, 2021.arxiv.org/abs/2102.07367
[21]F. Pedregosa, “Hyperparameter optimization with approximate gradient,” in International conference on machine learning. PMLR, 2016, pp. 737–746.arxiv.org/abs/1602.02355
[22]W. Rudin et al., Principles of mathematical analysis. McGraw-hill New York, 1976, vol. 3.
[23]J. Domke, “Generic methods for optimization-based modeling,” in Artificial Intelligence and Statistics. PMLR, 2012, pp. 318–326.
[24]D. Maclaurin, D. Duvenaud, and R. Adams, “Gradient-based hyperparameter optimization through reversible learning,” in International conference on machine learning. PMLR, 2015, pp. 2113–2122.DOI
[25]L. Franceschi, M. Donini, P. Frasconi, and M. Pontil, “Forward and reverse gradient-based hyperparameter optimization,” in International Conference on Machine Learning. PMLR, 2017, pp. 1165–1173.arxiv.org/abs/1703.01785
[26]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.
[27]M. Mackay, P. Vicol, J. Lorraine, D. Duvenaud, and R. Grosse, “Self-tuning networks: Bilevel optimization of hyperparameters using structured best-response functions,” in International Conference on Learning Representations, 2019.arxiv.org/abs/1903.03088
[28]Y. Yang, P. Xiao, and K. Ji, “Achieving O (ε⁻¹·⁵) complexity in hessian/jacobian-free stochastic bilevel optimization,” Advances in Neural Information Processing Systems, 2023.DOI
[29]J. Kwon, D. Kwon, S. Wright, and R. D. Nowak, “A fully first-order method for stochastic bilevel optimization,” in International Conference on Machine Learning. PMLR, 2023, pp. 18 083–18 113.arxiv.org/abs/2301.10945
[30]——, “On penalty methods for nonconvex bilevel optimization and first-order stochastic approximation,” in The Twenty-First International Conference on Learning Representations, 2024.arxiv.org/abs/2309.01753
[31]B. Liu, M. Ye, S. Wright, P. Stone, and Q. Liu, “‘Bome!’ bilevel optimization made easy: A simple first-order approach,” Advances in neural information processing systems, vol. 35, pp. 17 248–17 262, 2022.arxiv.org/abs/2209.08709
[32]H. Shen and T. Chen, “On penalty-based bilevel gradient descent method,” in International conference on machine learning. PMLR, 2023, pp. 30 992–31 015.arxiv.org/abs/2302.05185
[33]X. Song, W. Gao, Y. Yang, K. Choromanski, A. Pacchiano, and Y. Tang, “Es-maml: Simple hessian-free meta learning,” in International Conference on Learning Representations, 2020.arxiv.org/abs/1910.01215
[34]A. Aghasi and S. Ghadimi, “Fully zeroth-order bilevel programming via gaussian smoothing,” Journal of Optimization Theory and Applications, vol. 205, no. 2, p. 31, 2025.arxiv.org/abs/2404.00158
[35]A. Aghasi, J. Kwon, and S. Ghadimi, “Optimal zeroth-order bilevel optimization,” arXiv preprint arXiv:2510.03646, 2025.arxiv.org/abs/2510.03646
[36]M. Rando and S. Vaiter, “Zoba: An efficient single-loop zeroth-order bilevel optimization algorithm,” arXiv preprint arXiv:2601.21836, 2026.arxiv.org/abs/2601.21836
[37]Y. Diouane, V. Kungurtsev, F. Rinaldi, and D. Zeffiro, “Inexact direct-search methods for bilevel optimization problems,” Computational Optimization and Applications, vol. 88, no. 2, pp. 469–490, 2024.arxiv.org/abs/2307.09924
[38]J. C. Duchi, M. I. Jordan, M. J. Wainwright, and A. Wibisono, “Optimal rates for zero-order convex optimization: The power of two function evaluations,” IEEE Transactions on Information Theory, vol. 61, no. 5, pp. 2788–2806, 2015.arxiv.org/abs/1312.2139
[39]Z. Liu, X. Zhang, P. Khanduri, S. Lu, and J. Liu, “Interact: achieving low sample and communication complexities in decentralized bilevel learning over networks,” in Proceedings of the Twenty-Third International Symposium on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing, 2022, pp. 61–70.arxiv.org/abs/2207.13283
[40]P. Qiu, Y. Li, Z. Liu, P. Khanduri, J. Liu, N. B. Shroff, E. S. Bentley, and K. Turck, “Diamond: Taming sample and communication complexities in decentralized bilevel optimization,” in IEEE INFOCOM 2023-IEEE conference on computer communications. IEEE, 2023, pp. 1–10.arxiv.org/abs/2212.02376
[41]Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, 1998.
[42]X. Chen, T. Xiao, and K. Balasubramanian, “Optimal algorithms for stochastic bilevel optimization under relaxed smoothness conditions,” Journal of Machine Learning Research, vol. 25, no. 151, pp. 1–51, 2024.arxiv.org/abs/2306.12067
[43]K. He, X. Zhang, S. Ren, and J. Sun, “Identity mappings in deep residual networks,” in European conference on computer vision. Springer, 2016, pp. 630–645.arxiv.org/abs/1603.05027
[44]A. Krizhevsky, “Learning multiple layers of features from tiny images,” Master’s thesis, University of Toronto, 2009.
[45]E. Wong, L. Rice, and J. Z. Kolter, “Fast is better than free: Revisiting adversarial training,” in International Conference on Learning Representations, 2020.arxiv.org/abs/2001.03994
[46]M. Andriushchenko and N. Flammarion, “Understanding and improving fast adversarial training,” Advances in Neural Information Processing Systems, vol. 33, pp. 16 048–16 059, 2020.arxiv.org/abs/2007.02617
[48]Y. Bai, A. Jones, K. Ndousse, A. Askell, A. Chen, N. DasSarma, D. Drain, S. Fort, D. Ganguli, T. Henighan et al., “Training a helpful and harmless assistant with reinforcement learning from human feedback,” arXiv preprint arXiv:2204.05862, 2022.arxiv.org/abs/2204.05862
Ronald L. P. D. de Jong, Tim J. M. Jaspers, Raf A. H. Vervoort, Aron F. H. A. Bakker, Anna L. Dijk, Yiping Li, Jip L. Tolenaar, Jelle P. Ruurda, Willem M. Brinkman, Daan de Geus et al. ·