Optimal Acceleration for Minimax and Fixed-Point Problems is Not Unique

Authors: Taeho Yoon, Jaeyeon Kim, Jaewook J. Suh, Ernest K. Ryu

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 8, the paper presents 'numerical simulations illustrating the dynamics of dual-anchor algorithm.' It includes 'Figure 1. Trajectories generated by minimax optimization algorithms' and 'Figure 2. Performance of minimax algorithms in reducing L(xk) 2 for bilinear problem instances.', indicating empirical evaluation.
Researcher Affiliation Academia The authors' affiliations are listed as: '1Department of Mathematical Sciences, Seoul National University' and '2Department of Mathematics, University of California, Los Angeles.' Both are academic institutions.
Pseudocode No The paper presents algorithms like OHM, Dual-OHM, FEG, and Dual-FEG as mathematical update rules within the main text. It does not include dedicated pseudocode blocks or algorithm listings.
Open Source Code No The paper does not include any statement or link indicating the availability of open-source code for the methodology described.
Open Datasets Yes The paper refers to a 'worst-case bilinear example due to Ouyang & Xu (2021)' in Section 8 and provides detailed construction in Appendix I.2, citing 'Ouyang, Y. and Xu, Y. Lower complexity bounds of first-order methods for convex-concave bilinear saddle-point problems. Mathematical Programming, 185(1):1–35, 2021.' This constitutes access to an open dataset via citation.
Dataset Splits No The paper describes experiments on specific problem instances and synthetic data, e.g., 'L(u, v) = uv' or 'L(u, v) = u^2v'. It mentions 'initial points u0 = 0, v0 = 0' or 'u0, v0 with i.i.d. standard normal coordinates'. However, it does not specify any training, validation, or test dataset splits.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as CPU or GPU models.
Software Dependencies No The paper does not specify any software dependencies with version numbers, such as programming languages or libraries.
Experiment Setup Yes Section 8, 'Experiments', provides specific parameter values for the algorithms: 'α = 0.005 and N = 5000', 'α = 0.05 and N = 10000', 'α = 1.0 and N = 10000', 'µ = 0.1', and 'δ = 0.1, α = 0.5 and N = 10^5'. These are explicit hyperparameters.