Boosting Perturbed Gradient Ascent for Last-Iterate Convergence in Games

This repository contains the code for the paper "Boosting Perturbed Gradient Ascent for Last-Iterate Convergence in Games".

Installation

pip install -r requirements.txt

Run Experiments

To run experiments, use the following command:

python main.py --multirun +experiment=<SETTING>/<GAME>/<METHOD>

In the above command, please select <SETTING>, <GAME>, and <METHOD> from the options below:

• <SETTING>: full (for full feedback) or noisy (for noisy feedback)
• <GAME>: random_payoff or hard_concave_convex
• <METHOD>: gabp, apga, og, or aog

Specifically, to reproduce the results in our paper, run the following commands:

# Random Payoff Game with full feedback
python main.py --multirun +experiment=full/random_payoff/gabp,full/random_payoff/apga,full/random_payoff/og,full/random_payoff/aog

# Hard Concave-Convex Game with full feedback
python main.py --multirun +experiment=full/hard_concave_convex/gabp,full/hard_concave_convex/apga,full/hard_concave_convex/og,full/hard_concave_convex/aog

# Random Payoff Game with noisy feedback
python main.py --multirun +experiment=noisy/random_payoff/gabp,noisy/random_payoff/apga,noisy/random_payoff/og,noisy/random_payoff/aog

# Hard Concave-Convex Game with noisy feedback
python main.py --multirun +experiment=noisy/hard_concave_convex/gabp,noisy/hard_concave_convex/apga,noisy/hard_concave_convex/og,noisy/hard_concave_convex/aog

Citation

Kenshi Abe, Mitsuki Sakamoto, Kaito Ariu, and Atsushi Iwasaki. Boosting Perturbed Gradient Ascent for Last-Iterate Convergence in Games. In ICLR, 2025

Bibtex:

@inproceedings{abe2025boosting,
  title={Boosting Perturbed Gradient Ascent for Last-Iterate Convergence in Games},
  author={Kenshi Abe and Mitsuki Sakamoto and Kaito Ariu and Atsushi Iwasaki},
  booktitle={ICLR},
  year={2025},
  url={https://openreview.net/forum?id=Jrt9iWalFy}
}