Efficient Interpolation between Extragradient and Proximal Methods for Weak MVIs

Authors: Thomas Pethick, Ioannis Mavrothalassitis, Volkan Cevher

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test Algorithm 1 and RAPP on Pethick et al. (2022, Ex. 5) which can be parameterized by ρ and L (see Example F.1). We set γ = 0.99/L for both methods and αk = 0.001 for RAPP and compared on ρ = 0.98/L. Algorithm 1 is additionally run on multiple problem instances of varying ρ to determined the relationship with the (automatically selected) number of inner steps. The results are shown in Figure 2, where Algorithm 1 is observed to converge using substantially fewer inner iterations than the baseline.
Researcher Affiliation Academia Laboratory for Information and Inference Systems (LIONS), EPFL (EMAIL)
Pseudocode Yes Algorithm 1 An explicit hybrid proximal extragradient method
Open Source Code No The paper does not contain an explicit statement about releasing code or a link to a repository for the methodology described.
Open Datasets No The paper evaluates its methods on "Pethick et al. (2022, Ex. 5)", which describes a parameterized mathematical operator (a synthetic problem instance) rather than an empirical dataset that would typically be downloaded or referenced via a specific link/DOI for data.
Dataset Splits No The numerical evaluation is based on a mathematical example (an operator), not an empirical dataset. Therefore, the concept of training, validation, or test splits does not apply to the described experiments.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the numerical evaluation.
Software Dependencies No The paper does not mention any specific software dependencies or their version numbers used for implementing or evaluating the algorithms.
Experiment Setup Yes We set γ = 0.99/L for both methods and αk = 0.001 for RAPP and compared on ρ = 0.98/L. Algorithm 1 is additionally run on multiple problem instances of varying ρ to determined the relationship with the (automatically selected) number of inner steps.