Doubly Optimal No-Regret Learning in Monotone Games

Authors: Yang Cai, Weiqiang Zheng

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we numerically verify our theoretical results through Example 1. The numerical result is shown in Figure 1.
Researcher Affiliation Academia 1Department of Computer Science, Yale University, New Haven, USA.
Pseudocode Yes Algorithm 1 AOG with step-size adaptation
Open Source Code Yes The code can be found at https://github.com/weiqiangzheng1999/Doubly-Optimal-No-Regret-Learning.
Open Datasets No We consider a convex-concave min-max optimization problem minx X maxy Y f(x, y), which is also a two-player zero-sum game with f 1 = f 2 = f. Details of the choices of H, A, b, h, X, Y and step size η are deferred to Appendix F.
Dataset Splits No No specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce data partitioning was found.
Hardware Specification No No specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running experiments were found.
Software Dependencies No No specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment were found.
Experiment Setup Yes We choose n = 100, X = Y = [-200, 200]^n. We run both AOG and OG with step size η = 0.3 and initial points x1 = y1 = 1/sqrt(n)1 for 10^5 iterations.