Revisiting Inexact Fixed-Point Iterations for Min-Max Problems: Stochasticity and Structured Nonconvexity

Authors: Ahmet Alacaoglu, Donghwan Kim, Stephen Wright

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

Reproducibility Variable Result LLM Response
Research Type Theoretical We obtain optimal or best-known complexity guarantees with cohypomonotonicity or weak MVI conditions for ρ < 1 L. First main insight for the improvements in the convergence analyses is to harness the recently proposed conic nonexpansiveness property of operators. Second, we provide a refined analysis for inexact Halpern iteration... Third, we analyze a stochastic inexact Krasnosel ski ı-Mann iteration... (from Abstract) and sections like 'A. Proofs for Section 2'.
Researcher Affiliation Academia 1University of Wisconsin Madison, USA 2University of British Columbia, Canada 3KAIST, Republic of Korea.
Pseudocode Yes Algorithm 1 Inexact Halpern iteration for problems with cohypomonotonicity (p. 4), Algorithm 2 FBF(z0, N, A, Bin, LB) from (Tseng, 2000) (p. 4), Algorithm 3 Inexact KM iteration for problems with weak MVI (p. 6), etc.
Open Source Code No The paper does not provide any explicit statement about releasing code or a link to a code repository.
Open Datasets No The paper is theoretical and focuses on mathematical analysis and algorithm design. It provides theoretical examples of problem classes (e.g., interaction dominant min-max problems, von Neumann’s ratio game) but does not describe the use of any specific dataset for training or evaluation.
Dataset Splits No The paper is theoretical and does not describe empirical experiments with training, validation, or test data splits.
Hardware Specification No The paper is theoretical and does not describe experimental procedures, therefore no hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and focuses on mathematical analysis. It does not mention any software dependencies with specific version numbers needed for replication.
Experiment Setup No The paper is theoretical and discusses algorithm parameters (e.g., 'Input: Parameters βk = 1 k+2, η > 0, L, ρ, α = 1 ρ η, K 1, initial iterate x0 Rd') as part of its theoretical framework, not as settings for an empirical experimental setup.