Pith. sign in

REVIEW 2 major objections 5 minor 29 references

A Modular Algorithm for Non-Stationary Online Convex-Concave Optimization

T0 review · 2 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A modular algorithm achieves, up to logarithmic factors, the minimax optimal dynamic duality gap for online convex-concave optimization, and uses multiple predictors to shrink the gap to near-constant in predictable environments.

desk verdict Introduces dynamic duality gap for OCCO with a modular algorithm and prediction-error bounds, but the solver's strong-monotonicity claim in Theorem 9 is unproven and likely false; needs a fix before publication. read the letter →

arxiv 2509.07901 v1 pith:BXCD2H2T submitted 2025-09-09 cs.LG math.OC

Qing-xin Meng , Xia Lei , Jian-wei Liu This is my paper · ORCID
classification cs.LGmath.OC MSC 68Q3290C2591A2690C47
keywords non-stationaryonlinelearningconvex-concaveoptimizationdynamicdualitygapmodularalgorithminterdependentupdatemeta-expertframeworkmulti-predictoraggregationminimaxoptimality
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

The paper studies two players who repeatedly pick strategies in a time-varying convex-concave game, with the goal of keeping the dynamic duality gap small against any hindsight comparator sequence. It claims that a modular algorithm—an adaptive pair of expert learners, a pool of predictors aggregated by clipped Hedge, and an integration layer with a coupled expert/meta update—achieves the minimax optimal duality gap up to logarithmic factors. The same algorithm also guarantees a prediction-error-driven bound: if any of $d$ predictors tracks the true payoff well, the gap drops to near-constant rather than growing like $\sqrt{T}$. This matters because existing algorithms for online convex-concave optimization either pay the full worst-case regret even in benign environments or lack tight guarantees. Experiments on stationary, periodic, and adversarial payoff sequences illustrate the predicted adaptability.

What carries the argument

The load-bearing objects are the dynamic duality gap itself, the modular expert/meta decomposition, and the coupled best-response system that joins the expert and meta layers. The D-DGap is split into two dynamic-regret terms plus a static duality-gap term; a pair of ADER or ADER-like learners controls the path-length part, an optimistic proximal update built on Fenchel coupling controls the prediction-error part, and a clipped-Hedge meta-layer selects among $d$ predictors. The distinctive mechanism is the interdependent update: the expert advice $(\hat x_t,\hat y_t)$ and the meta-weights $(w_t,\omega_t)$ are computed together as the solution of the coupled system in Equations (3b) and (4b), reformulated as a four-player best-response game. Existence and uniqueness come from the Browder–Minty theorem applied to a stacked gradient operator $G$, and Algorithm 1, a strongly monotone variational-inequality solver, computes the solution at a linear rate.

What would settle it

Take a simple instance with $X=Y=[-1,1]$, quadratic regularizers, and a bilinear predictor $h_t(x,y)=xy$, then compute the smallest value of $\langle G(x)-G(x'),x-x'\rangle/\|x-x'\|^2$ for the operator $G$ in Equation (7). If this infimum is below 1 (equivalently, if the symmetric part of the Jacobian of $G$ has an eigenvalue below 1), the strong-monotonicity premise of Theorem 9 fails, Algorithm 1's linear-rate guarantee does not apply, and the exact $(\hat x_t,\hat y_t,w_t,\omega_t)$ needed by Theorems 6–8 may not be computable by the proposed subroutine.

Watch

Extended reading notes

Core claim

The central claim is Theorem 14: the modular algorithm guarantees, for every comparator sequence, $$\mathrm{D-DGap}(u_{1:T},v_{1:T}) \le \tilde O\left(\min\left\{\min_{k\in\{1,\dots,d\}}\sum_{t=1}^T \rho(f_t,h^k_t),\ \sqrt{(1+\min\{P_T,C_T\})T}\right\}\right),$$ where $\rho(f,h)=\max_{x,y}|f(x,y)-h(x,y)|$ is the pointwise prediction error, $P_T$ is the path length of the comparator sequence, and $C_T$ is a data-dependent upper threshold for it. The first term means that when any of $d$ predictors models the payoff sequence well, the duality gap is driven by that predictor's cumulative error—near-constant in stationary or periodic environments. The second term is the minimax-optimal dynamic-regret floor, up to a logarithmic factor. The proof decomposes the D-DGap into two dynamic-regret terms handled by the Adaptive Module and a static duality-gap term handled by the meta-layer, then couples the expert and meta updates through a joint variational inequality solved by a linearly convergent algorithm.

Load-bearing premise

The whole guarantee rests on the assumption that the joint best-response map used to solve the two coupled updates is strongly monotone—a property that does not automatically follow from the individual blocks being strongly convex, so the cross-coupling terms must actually preserve it for the linear-convergence solver to work.

Editorial extensions

If this is right

  • In stationary or fully predictable environments, where some predictor has cumulative error $\sum_t \rho(f_t,h^k_t)=O(1)$, the D-DGap collapses to $\tilde O(1)$ instead of growing with $T$.
  • When no predictor helps, the bound falls back to $\tilde O(\sqrt{(1+\min\{P_T,C_T\})T})$, which Proposition 3 shows is minimax optimal up to logarithmic factors.
  • The meta-layer alone achieves a static duality gap of $\tilde O(\min\{\sum_t \rho(f_t,h_t), \sqrt{T\ln T}\})$, so the algorithm is never worse than parameter-free Hedge even with no side information.
  • Because the Adaptive Module is exchangeable, any ADER-like base learner with its own dynamic-regret guarantee can be plugged in and inherit the same overall D-DGap bound.
  • With $d$ predictors, the algorithm automatically identifies the best one after the fact, so environments whose period matches one of the predictors (periods 2, 3, 4, 7, 8 in the experiments) enjoy near-constant duality gap, while adversarial environments remain at the minimax floor.

Reading between the lines

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

  • Editorial inference: replacing Algorithm 1 by any other linearly convergent monotone-operator solver with the same Lipschitz constant would preserve the D-DGap theorem while removing dependence on the specific strong-monotonicity assertion.
  • Editorial inference: the aggregator's loss vector $L_t$ is built from evaluations at the algorithm's own query points, so the same multi-predictor construction would work for arbitrary finite sets of side-information signals, not only for payoff functions as predictors.
  • Editorial inference: the bound $\min_k \sum_t \rho(f_t,h^k_t)$ suggests that deliberately injecting a family of candidate-world models (for example, periodic models with different periods) turns model selection into an online subproblem; the paper's own experimental predictor set could be used to test how the bound degrades when the true period is not among the candidates.
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

2 major / 5 minor

Summary. The paper studies non-stationary Online Convex-Concave Optimization (OCCO) with the dynamic duality gap (D-DGap) as the performance metric. It proposes a modular algorithm with three components: an Adaptive Module running two ADER-like algorithms, a Prediction-Error Expert using optimistic implicit updates, and a Meta-Layer combining the two experts via clipped Hedge; a Multi-Predictor Aggregator further selects among several predictors. The main claim, Theorem 14, is an Õ(min{min_k Σ_t ρ(f_t,h^k_t), sqrt((1+min{P_T,C_T})T)}) bound on D-DGap, with the minimax-optimal adaptive term and a prediction-error-driven term. The analysis is developed in Sections 3.2–3.4, with proofs in Appendix A, and Section 4 reports experiments on synthetic quadratic games. The paper's central technical innovation is an interdependent update between the expert and meta layers, whose coupled saddle-point equations are solved by a strongly monotone VI solver (Algorithm 1).

Significance. The manuscript introduces a clean modular architecture for non-stationary OCCO and a new decomposition of the dynamic duality gap into expert and meta terms. Theorems 6 and 7, which bound the prediction-error expert and the meta layer, are plausible adaptations of established optimistic implicit OCO templates, and the multi-predictor aggregation via clipped Hedge is a useful add-on. The paper also ships an experimental evaluation on several non-stationary regimes. If the two proof gaps identified below are repaired, the claimed bound would be a meaningful contribution to the online learning literature. As it stands, however, two load-bearing steps in Theorems 8 and 9 are not supported, so the main theorem is not established in the current form.

major comments (2)
  1. [Section 3.3, Theorem 9 and Eq. (7)] The proof of Theorem 9 asserts that since each ℓ_i in Eq. (7) is 1-strongly convex in its own coordinate, the stacked operator G is 1-strongly monotone with respect to the norm ‖x‖²=‖x‖²+‖y‖²+w²+ω². This implication is false: component-wise strong convexity of the diagonal blocks does not control the off-diagonal cross terms. A concrete instance satisfying the assumptions is f_t(x,y)=xy on X=Y=[-1,1], φ(z)=ψ(z)=z²/2, ex_t=ey_t=0, η_t=γ_t=100, θ_t=ϑ_t=1, w=0.999, ω=0.01. Using the explicit formula for G in Eq. (25), at x=(1,1,w,ω) and x'=(0,0,w,ω) the first two blocks of G(x)-G(x') are (2,-98.9), so ⟨G(x)-G(x'),x-x'⟩ = 2-98.9 = -96.9, whereas 1-strong monotonicity would require at least ‖x-x'‖²=2. Thus the Browder-Minty argument does not apply, and Proposition 13's linear rate, which explicitly sets the strong-monotonicity constant μ=1, is unsupported. Because Algorithm 1 is the proposed method for computing the coupled solutions (3b)-(4b) used in Theorems 6-8 and hence in Theorem 14, the implementability of the central algorithm is not established.
  2. [Section 3.3, Theorem 8 and Eq. (6)] The inequality in Eq. (6) is not an identity and is not generally true. The middle term M_t := w_t^T A_t [0;1] - [0,1] A_t ω_t equals w_t[f_t(bx_t,by_t)-f_t(x_t,by_t)] + (1-ω_t)[f_t(x_t,by_t)-f_t(x_t,y_t)], which can be negative. For the instance f_t(x,y)=xy, X=Y=[-1,1], φ=ψ=z²/2, x_t=y_t=1, ex_t=ey_t=0, ew_t=eω_t=1/2, and η_t=γ_t=θ_t=ϑ_t=1, a numerical solution of the coupled first-order conditions for (3b)-(4b) gives approximately w_t≈0.676, ω_t≈0.502, bx_t≈-0.495, by_t≈-0.012, and M_t≈-0.49. With comparators u_t=v_t=0.5, the left-hand side of Eq. (6) is approximately -0.251, while the right-hand side is approximately -0.74, contradicting the claimed inequality. Consequently the adaptive-module bound in Theorem 8 and the second term inside the minimum of Theorem 14 are not proven.
minor comments (5)
  1. [Algorithm 1, line 9] The stopping rule 'until (L/(L+1))^{k/2}↓0' is asymptotic; please specify a finite tolerance and analyze how the approximation error propagates into the bounds of Theorems 6-8.
  2. [Proposition 13] The rate in Proposition 13 is quoted from Nesterov and Scrimali (2006) after setting μ=1, but the dependence on the initial residual ||G(y_0)|| and on the dimension of K is not stated; the statement should be self-contained.
  3. [Equation (7)] The notation w^TΛ_t(x,y)ω/w in ℓ1 and the analogous /ω in ℓ2 is confusing because w and ω denote both scalars and vectors; please add an explicit explanation of the intended scaling.
  4. [Section 4] The text refers to 'Figure 2c' and 'Figure 2f' for panels that appear in Figure 3; the cross-references need to be corrected.
  5. [Algorithm 2, lines 5-6] Please clarify how the output of Algorithm 1 is split into (bx_t,by_t) and (w_t,ω_t), and note that the Lipschitz constant L in Algorithm 1 must be recomputed at each round from the current predictor h_t.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the D-DGap bound is competitive against external predictors, learning rates are adaptive/self-bounding, and the sole Meng-Liu self-citation is contextual. Theorem 9's strong-monotonicity gap is a correctness issue, not a circular one.

full rationale

The derivation is self-contained with respect to circularity. Theorem 14's bound is a competitive guarantee: min_k sum_t rho(f_t,h^k_t) is an external predictor-error term, not a fitted value, and the sqrt((1+min{P_T,C_T})T) term comes from the ADER-type adaptive module (Proposition 4), not from the target D-DGap. The learning rates eta_t, gamma_t, theta_t, upsilon_t, and zeta_t are self-bounding: each is determined from previously observed residuals (delta^x_t, delta^y_t, Delta^x_t, Delta^y_t, and Delta_t), so the algorithm adapts to, rather than fits, the quantity being bounded. The only self-citation, Meng and Liu (2025), appears in Related Work as motivation for D-DGap and carries none of the proof weight; Theorems 6, 7, 8, and 14 are proved from optimality conditions, Fenchel couplings, and cited external tools (ADER, clipped Hedge, Nesterov-Scrimali). The flagged weakness is a correctness gap, not circularity: Theorem 9's proof asserts that 'By construction, each ell_i is 1-strongly convex in its own coordinate, so G is 1-strongly monotone,' which does not follow because off-diagonal cross terms in the stacked Jacobian can destroy strong monotonicity; this affects whether Algorithm 1 provably computes the exact (bx_t, by_t, w_t, omega_t), but it is not a case of a prediction being equivalent to an input by construction. Hence no circular step is present.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

The paper relies on standard OCO and variational inequality machinery, plus a domain assumption on predictors. The only ad hoc assumption is the unproved strong monotonicity of G in Theorem 9, which is critical for the solver's convergence. No new physical or mathematical entities are postulated.

free parameters (1)
  • epsilon = unspecified positive constant
    Added to every learning rate denominator to prevent division by zero at initialization (e.g., Lemma 5, Theorem 6, Theorem 7, Theorem 14). It is a smoothing parameter, not fitted to data.
assumptions (6)
  • standard math Minimax lower bound for dynamic regret in OCO (Theorem 2 of Zhang et al. 2018)
    Used in Proposition 3 to lower bound the D-DGap by decomposing into two independent OCO regrets.
  • standard math ADER upper bound for dynamic regret (Zhang et al. 2018)
    Used in Proposition 4 to establish the minimax-optimal D-DGap bound from running two independent ADER algorithms.
  • standard math Browder-Minty theorem for strongly monotone variational inequalities (Brezis 2011)
    Used in Theorem 9 to guarantee existence and uniqueness of the solution to the coupled expert/meta update system.
  • domain assumption Bounded domains and bounded subgradients (Assumptions 1 and 2)
    Standard in OCCO; used throughout the proofs for finite constants such as D_X, D_Y, G_X, G_Y.
  • domain assumption Predictors have Lipschitz-continuous gradients (Assumption 11)
    Needed to compute the Lipschitz constant L of the operator G and to run Algorithm 1.
  • ad hoc to paper The stacked gradient operator G is 1-strongly monotone
    Asserted in Theorem 9's proof from component-wise strong convexity of each ell_i, but this implication is not generally valid and is load-bearing for the convergence of Algorithm 1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Modular Algorithm for Non-Stationary Online Convex-Concave Optimization." pith.science (2026). https://pith.science/paper/BXCD2H2T

@misc{pith2026250907901,
  author       = {Pith},
  title        = {Pith review of: A Modular Algorithm for Non-Stationary Online Convex-Concave Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BXCD2H2T}},
  note         = {Machine review of arXiv:2509.07901}
}
read the original abstract

This paper investigates the problem of Online Convex-Concave Optimization, which extends Online Convex Optimization to two-player time-varying convex-concave games. The goal is to minimize the dynamic duality gap (D-DGap), a critical performance measure that evaluates players' strategies against arbitrary comparator sequences. Existing algorithms fail to deliver optimal performance, particularly in stationary or predictable environments. To address this, we propose a novel modular algorithm with three core components: an Adaptive Module that dynamically adjusts to varying levels of non-stationarity, a Multi-Predictor Aggregator that identifies the best predictor among multiple candidates, and an Integration Module that effectively combines their strengths. Our algorithm achieves a minimax optimal D-DGap upper bound, up to a logarithmic factor, while also ensuring prediction error-driven D-DGap bounds. The modular design allows for the seamless replacement of components that regulate adaptability to dynamic environments, as well as the incorporation of components that integrate ``side knowledge'' from multiple predictors. Empirical results further demonstrate the effectiveness and adaptability of the proposed method.

Figures

Figures reproduced from arXiv: 2509.07901 by the authors.

Figure 1
Figure 1. Structural Diagram of Our Modular Algorithm. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Structural Detail of Our Modular Algorithm. [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Time-Averaged D-DGaps of Algorithms Acknowledgments and Disclosure of Funding This work was supported by the National Natural Science Foundation of China (Grant No. 62406299). CRediT author statement • Qing-xin Meng: Conceptualization; Methodology; Software; Formal analysis; Investi￾gation; Writing – Original Draft; Visualization. • Xia Lei: Conceptualization; Validation; Writing – Review & Editing; Resources; Fund￾… view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 18 canonical work pages

  1. [1]

    Near-optimal no-regret learning for correlated equi- libria in multi-player general-sum games

    Ioannis Anagnostides, Constantinos Daskalakis, Gabriele Farina, Maxwell Fishelson, Noah Golowich, and Tuomas Sandholm. Near-optimal no-regret learning for correlated equi- libria in multi-player general-sum games. InProceedings of the 54th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2022, pages 736–749, New York, NY, USA,

    work page 2022

  2. [9]

    doi: 10.1016/j.ejor.2021.03.023

    ISSN 0377-2217. doi: 10.1016/j.ejor.2021.03.023. URLhttps://www.sciencedirect.com/ science/article/pii/S0377221721002496. Mark Herbster and Manfred K Warmuth. Tracking the best linear predictor.Journal of Machine Learning Research, 1:281–309,

    work page doi:10.1016/j.ejor.2021.03.023 2021

  3. [11]

    Adaptive and Efficient Algorithms for Tracking the Best Expert

    doi: 10.48550/arXiv.1909.02187. Shiyin Lu, Yuan Miao, Ping Yang, Yao Hu, and Lijun Zhang. Non-stationary duel- ing bandits for online learning to rank. In Bohan Li, Lin Yue, Chuanqi Tao, Xum- ing Han, Diego Calvanese, and Toshiyuki Amagasa, editors,Web and Big Data, pages 166–174, Cham,

    work page Pith review arXiv doi:10.48550/arxiv.1909.02187 1909

  4. [15]

    Learning in games with continuous action sets and unknown payoff functions

    doi: 10.48550/arXiv.1608.07310. Yurii Nesterov and Laura Scrimali. Solving strongly monotone variational and quasi- variational inequalities.CORE Discussion Paper, 2006/107,

    work page Pith review arXiv doi:10.48550/arxiv.1608.07310 2006

  5. [17]

    doi: 10.48550/arXiv.1912.01698. R. Schapire, N. Cesa-Bianchi, P. Auer, and Y. Freund. Gambling in a rigged casino: The adversarial multi-armed bandit problem. In2013 IEEE 54th Annual Symposium on Foundations of Computer Science, page 322, Los Alamitos, CA, USA, October

    arXiv doi:10.48550/arxiv.1912.01698 1912

  6. [19]

    2023.3237486

    doi: 10.1109/TAC. 2023.3237486. Shai Shalev-Shwartz. Online learning and online convex optimization.Foundations and Trends®in Machine Learning, 4(2):107–194,

    arXiv 2023

  7. [21]

    doi: 10.5555/2969442.2969573

    MIT Press. doi: 10.5555/2969442.2969573. 28 A Modular Algorithm for Non-Stationary OCCO Tim van Erven and Wouter M Koolen. Metagrad: Multiple learning rates in on- line learning. In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Gar- nett, editors,Advances in Neural Information Processing Systems, volume

    arXiv

  8. [22]

    John von Neumann

    URLhttps://proceedings.neurips.cc/paper/2016/file/ 14cfdb59b5bda1fc245aadae15b1984a-Paper.pdf. John von Neumann. Zur theorie der gesellschaftsspiele.Mathematische Annalen, 100(1): 295–320, Dec

    work page 2016

Show all 29 references
  1. [25]

    cc/paper/2018/file/10a5ab2db37feedfdeaab192ead4ac0e-Paper.pdf

    URLhttps://proceedings.neurips. cc/paper/2018/file/10a5ab2db37feedfdeaab192ead4ac0e-Paper.pdf. Lijun Zhang, Wei Jiang, Shiyin Lu, and Tianbao Yang. Revisiting smoothed online learning. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan, editors,Advan...

    2018

  2. [26]

    Lijun Zhang, Guanghui Wang, Jinfeng Yi, and Tianbao Yang

    URLhttps://proceedings.neurips.cc/paper/ 2021/file/70fc5f043205720a49d973d280eb83e7-Paper.pdf. Lijun Zhang, Guanghui Wang, Jinfeng Yi, and Tianbao Yang. A simple yet univer- sal strategy for online convex optimization. In Kamalika Chaudhuri, Stefanie Jegelka, Le Song, Csaba Sz...

    2021

  3. [27]

    Peng Zhao, Yu-Jie Zhang, Lijun Zhang, and Zhi-Hua Zhou

    URLhttps://proceedings.neurips.cc/paper/ 2020/file/939314105ce8701e67489642ef4d49e8-Paper.pdf. Peng Zhao, Yu-Jie Zhang, Lijun Zhang, and Zhi-Hua Zhou. Adaptivity and Non- stationarity: Problem-dependent Dynamic Regret for Online Convex Optimization.arXiv e-prints, December

    2020

  4. [28]

    29 Meng, Lei and Liu Peng Zhao, Yan-Feng Xie, Lijun Zhang, and Zhi-Hua Zhou

    doi: 10.48550/arXiv.2112.14368. 29 Meng, Lei and Liu Peng Zhao, Yan-Feng Xie, Lijun Zhang, and Zhi-Hua Zhou. Efficient meth- ods for non-stationary online learning. In S. Koyejo, S. Mohamed, A. Agar- wal, D. Belgrave, K. Cho, and A. Oh, editors,Advances in Neural Infor- mation...

    doi arXiv

  5. [29]

    Martin Zinkevich

    URLhttps://proceedings.neurips.cc/paper_files/paper/2022/hash/ 4b70484ebef62484e0c8cdd269e482fd-Abstract.html. Martin Zinkevich. Online convex programming and generalized infinitesimal gradient ascent. In Tom Fawcett and Nina Mishra, editors,Proceedings of the Twentieth Intern...

    2022

  6. [1928]

    doi: 10.1007/BF01448847

    ISSN 1432-1807. doi: 10.1007/BF01448847. Lijun Zhang. Online learning in changing environments. In Christian Bessiere, editor,Pro- ceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20, pages 5178–5182. International Joint Conferences...

    doi

  7. [1995]

    doi: 10.1109/SFCS.1995.492488

    IEEE Computer Society. doi: 10.1109/SFCS.1995.492488. Pedro Zattoni Scroccaro, Arman Sharifi Kolarijani, and Peyman Mohajerin Esfahani. Adap- tive composite online optimization: Predictions in static and dynamic environments. IEEE Transactions on Automatic Control, 68(5):2906–2921,

    1995

  8. [1999]

    doi: 10.1006/game

    ISSN 0899-8256. doi: 10.1006/game. 1999.0738. Sini Guo, Jia-Wen Gu, and Wai-Ki Ching. Adaptive online portfolio selection with trans- action costs.European Journal of Operational Research, 295(3):1074–1086,

    1999

  9. [2006]

    doi: 10.2139/ssrn. 970903. Abhishek Roy, Yifang Chen, Krishnakumar Balasubramanian, and Prasant Mohapatra. Online and Bandit Algorithms for Nonstationary Stochastic Saddle-Point Optimization. arXiv e-prints, December

    doi

  10. [2011]

    doi: 10.1007/ 978-0-387-70914-7

    ISBN 9780387709130. doi: 10.1007/ 978-0-387-70914-7. 26 A Modular Algorithm for Non-Stationary OCCO Nicol` o Campolongo and Francesco Orabona. A Closer Look at Temporal Variability in Dynamic Online Learning.arXiv e-prints, art. arXiv:2102.07666, February

    arXiv

  11. [2012]

    doi: 10.1561/ 2200000018

    ISSN 1935-8237. doi: 10.1561/ 2200000018. Vasilis Syrgkanis, Alekh Agarwal, Haipeng Luo, and Robert E. Schapire. Fast convergence of regularized learning in games. InProceedings of the 28th International Conference on Neural Information Processing Systems - Volume 2, NIPS’15, ...

    1935

  12. [2015]

    doi: 10.1016/j.geb.2014.01.003

    ISSN 0899-8256. doi: 10.1016/j.geb.2014.01.003. Constantinos Costis Daskalakis, Maxwell Fishelson, and Noah Golowich. Near-optimal no-regret learning in general games. In A. Beygelzimer, Y. Dauphin, P. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Proce...

    2014 doi

  13. [2016]

    Panayotis Mertikopoulos and Zhengyuan Zhou

    doi: 10.1287/moor.2016.0778. Panayotis Mertikopoulos and Zhengyuan Zhou. Learning in games with continuous action sets and unknown payoff functions.arXiv e-prints, art. arXiv:1608.07310, August

    2016

  14. [2017]

    Constantinos Daskalakis, Alan Deckelbaum, and Anthony Kim

    doi: 10.1109/TSP.2017.2750109. Constantinos Daskalakis, Alan Deckelbaum, and Anthony Kim. Near-optimal no-regret algorithms for zero-sum games.Games and Economic Behavior, 92:327–348,

    2017

  15. [2018]

    Adrian Rivera Cardoso, Jacob Abernethy, He Wang, and Huan Xu

    doi: 10.48550/ arXiv.1806.08301. Adrian Rivera Cardoso, Jacob Abernethy, He Wang, and Huan Xu. Competing against nash equilibria in adversarially changing zero-sum games. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors,Proceedings of the 36th International Conference o...

    doi arXiv

  16. [2019]

    doi: 10.1007/s10107-018-1262-8

    ISSN 1436-4646. doi: 10.1007/s10107-018-1262-8. 27 Meng, Lei and Liu Shiyin Lu and Lijun Zhang. Adaptive and Efficient Algorithms for Tracking the Best Expert. arXiv e-prints, September

    doi

  17. [2020]

    Early Career

    doi: 10.24963/ijcai.2020/731. Early Career. Lijun Zhang, Shiyin Lu, and Zhi-Hua Zhou. Adaptive online learning in dynamic envi- ronments. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors,Advances in Neural Information Processing Sys...

    2020 doi

  18. [2021]

    Adrian Rivera Cardoso, He Wang, and Huan Xu

    doi: 10.48550/arXiv.2102.07666. Adrian Rivera Cardoso, He Wang, and Huan Xu. The Online Saddle Point Problem and Online Convex Optimization with Knapsacks.arXiv e-prints, June

    doi arXiv

  19. [2022]

    ISBN 9781450392648

    Association for Computing Machinery. ISBN 9781450392648. doi: 10.1145/3519935.3520031. Ioannis Anagnostides, Ioannis Panageas, Gabriele Farina, and Tuomas Sandholm. On the convergence of no-regret learning dynamics in time-varying games. InThirty-seventh Conference on Neural I...

  20. [2023]

    ISBN 978-3-031-25198-6

    Springer Nature Switzerland. ISBN 978-3-031-25198-6. doi: 10.1007/978-3-031-25198-6

    doi

  21. [2025]

    ISBN 978-981-96-6579-2

    Springer Nature Singapore. ISBN 978-981-96-6579-2. doi: 10.1007/978-981-96-6579-2

    doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.