Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Solving Nonconvex-Nonconcave Min-Max Problems exhibiting Weak Minty Solutions
Authors: Axel Böhm
TMLR 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In Figure 1 we see that beyond the theoretical guarantees OGDA+ can even provide convergence where EG+ does not. [...] 5 Numerical experiments In the following we compare EG+ method from (Diakonikolas et al., 2021) with the two methods we propose OGDA+ and EG+ with adaptive step size, see Algorithm 1 and Algorithm 3 respectively. |
| Researcher Affiliation | Academia | Axel Böhm EMAIL University of Vienna, Austria |
| Pseudocode | Yes | Algorithm 1 OGDA+ [...] Algorithm 2 stochastic OGDA+ [...] Algorithm 3 EG+ with adaptive step size |
| Open Source Code | No | No explicit statement about releasing code for the methodology described in this paper or links to code repositories are provided. The OpenReview link is for peer review, not source code. |
| Open Datasets | No | The paper refers to mathematical problems or toy examples (e.g., "von Neumann s ratio game (von Neumann, 1945)", "min-max toy example with Forsaken solution was proposed in Example 5.2 of (Hsieh et al., 2021)", "min-max problem was introduced in (Pethick et al., 2022)"). These are mathematical formulations used as test cases, not publicly available datasets with specific access information. |
| Dataset Splits | No | The paper does not use traditional datasets; instead, it evaluates methods on mathematically defined min-max problems. Therefore, the concept of training/test/validation splits is not applicable, and no such information is provided. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the numerical experiments. |
| Software Dependencies | No | The paper does not mention specific software dependencies or their version numbers (e.g., programming languages, libraries, frameworks) used for the implementation or experiments. |
| Experiment Setup | Yes | For all experiments, if not specified otherwise, we used for OGDA+ and the adaptive version of EG+ the parameter γ = 1/2. For the step size choice of Algorithm 3 we use τ = 0.99. For the Curvature EG+ method of (Pethick et al., 2022) (with their notation) we use δk equal to ρ/2, where ρ is the weak Minty parameter, if it is known and less than 1/L; and 0.499 times the step size, otherwise. Furthermore we set the parameters of the linesearch to τ = 0.9 and ν = 0.99. |