Fast Rates in Time-Varying Strongly Monotone Games

Authors: Yu-Hu Yan, Peng Zhao, Zhi-Hua Zhou

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5. Experiment This section provides empirical evaluations of our proposed method in time-varying strongly monotone games. [...] Results. We report average results with standard deviations of 5 independent runs. [...] Figure 1 plots the tracking error of all methods. Smaller tracking error indicates better performance. The results show the supremacy of our TV-Smog and TV-Smog (Smooth), supporting our theoretical results.
Researcher Affiliation Academia 1National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China. Correspondence to: Peng Zhao <zhaop@lamda.nju.edu.cn>.
Pseudocode Yes Algorithm 1 TV-SMOG for the i-th player, Algorithm 2 TV-SMOG (smooth) for the i-th player, Algorithm 3 Deploying ADER (Zhang et al., 2018) for the i-th player, Algorithm 4 FLH-OGD (Hazan & Seshadhri, 2007), Algorithm 5 Known strong monotonicity parameter µ and box domain (Baby & Wang, 2022), Algorithm 6 Known strong monotonicity parameter µ (Baby & Wang, 2022).
Open Source Code No The paper does not contain any statement about open-source code release or provide links to a code repository for its methodology.
Open Datasets Yes We use four LIBSVM datasets to initialize the logistic loss.
Dataset Splits No The paper does not provide specific training/test/validation dataset splits, percentages, or absolute sample counts.
Hardware Specification No The paper does not explicitly describe the hardware (e.g., specific GPU or CPU models, cloud resources) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependency details with version numbers (e.g., libraries, frameworks, or programming language versions).
Experiment Setup Yes We set µ = 0.005, i.e., ut is 0.01-strongly monotone. [...] We focus on the standard linear model: P(x) a b PN i=1 xi, where a = 0.5 and b = 1/(NR). [...] We consider time-varying zero-sum strongly convex-concave games with utility ut(x, y) 1 2 x 2 + x Aty 1 2 y 2, which is 1-strongly monotone and 1-smooth. [...] All hyper-parameters are set to be theoretically optimal.