Increasing Iterate Averaging for Solving Saddle-Point Problems

Authors: Yuan Gao, Christian Kroer, Donald Goldfarb7537-7544

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive numerical experiments on zero-sum game solving, market equilibrium computation and image denoising demonstrate the effectiveness of the proposed schemes.
Researcher Affiliation Academia Yuan Gao, Christian Kroer, Donald Goldfarb Columbia University, Department of Industrial Engineering and Operations Research gao.yuan@columbia.edu, christian.kroer@columbia.edu, goldfarb@columbia.edu
Pseudocode Yes Algorithm 1 Nonlinear primal-dual algorithm (PDA); Algorithm 2 Relaxed primal-dual algorithm (RPDA); Algorithm 3 Inertial primal-dual algorithm (IPDA); Algorithm 4 PDAL: Primal-dual algorithm with linesearch; Algorithm 5 Mirror Descent (MD) and Mirror Prox (MP)
Open Source Code No The paper mentions an extended manuscript at https://arxiv.org/abs/1903.10646, which is a link to the paper itself, not explicitly to source code for the methodology.
Open Datasets Yes EFG benchmark instances Kuhn and Leduc poker (see, e.g., (Kroer et al. 2018)).
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes For all algorithms, we use their default stepsizes and Euclidean DGF. We perform T = 2000 iterations. The choices of algorithm hyperparameters are completely analogous to those in solving matrix games. [...] Following Chambolle and Pock (2011), to align it with (4), choose f = 0, g(u) = λ u g 1 with λ = 1.5 and h (p) = δP (p); in this way, the proximal mappings yield closed-form formulas. We use PDA with default, static hyperparameters used in (Chambolle and Pock 2011) and run for T = 1000 iterations.