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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

Uncoupled and Convergent Learning in Monotone Games under Bandit Feedback

Authors: Jing Dong, Baoxiang Wang, Yaoliang Yu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we provide a numerical evaluation of our proposed algorithm in three static games. We repeat each experiment with 50 different random seeds. We ran all experiments with a 10-core CPU, with 32 GB memory. We set t = 1 pt+1, and δt = 0.001 for our algorithm. [...] Figure 1 summarizes our experimental findings, where our algorithm attains comparable performance to online mirror descent and gradient descent with full information.
Researcher Affiliation Academia Jing Dong The Chinese University of Hong Kong, Shenzhen EMAIL Baoxiang Wang The Chinese University of Hong Kong, Shenzhen EMAIL Yaoliang Yu University of Waterloo and Vector Institute EMAIL
Pseudocode Yes Algorithm 1: Doubly regularized online Mirror Descent with bandit feedback
Open Source Code Yes All the codes can be found at https://github.com/jingdong00/monotone_games.
Open Datasets No The experimental environment used in this paper is synthetic.
Dataset Splits No The paper uses synthetic or custom-defined game parameters for its experiments, rather than pre-existing datasets with explicit train/test/validation splits.
Hardware Specification Yes We ran all experiments with a 10-core CPU, with 32 GB memory.
Software Dependencies No The paper specifies algorithmic parameters (e.g., learning rate, δt) but does not list specific software dependencies with version numbers (e.g., Python, PyTorch versions).
Experiment Setup Yes We set t = 1 pt+1, and δt = 0.001 for our algorithm. [...] We set the learning rate to be 0.01 in both zero-sum matrix games and monotone zero-sum matrix games and 0.09 in Cournot competition.