The joint research results of OMRON SINIC X Corporation (HQ: Bunkyo-ku, Tokyo; President and CEO: Masaki Suwa, hereinafter “OSX”) has been accepted at the Forty-Second International Conference on Machine Learning (hereinafter “ICML 2025”).

Along with NeurIPS^*1, ICML is one of the premier international conferences with significant authority in the field of machine learning and related areas. The conference will be held in Vancouver, Canada, from July 13 to July 19, 2025 (local time).

The research paper that has been accepted is as follows:

ICML 2025 presentations

_^■ The Harder Path: Last Iterate Convergence for Uncoupled Learning in Zero-Sum Games with Bandit Feedback

Côme Fiegel (ENSAE), Pierre Menard (Meta), Tadashi Kozuno, Michal Valko (INRIA), Vianney Perchet (ENSAE)

We study learning in repeated zero-sum matrix games under the uncoupled bandit feedback setting, where players cannot communicate and observe only their own rewards and actions. We theoretically demonstrate that guaranteeing last-iterate convergence to a Nash equilibrium comes at a significant cost: the best achievable convergence rate is O(1/T^0.25), which is substantially slower than the standard O(1/T^0.5) rate for average-iterate convergence. We then propose two algorithms that achieve this optimal rate up to constant and logarithmic factors.

https://icml.cc/virtual/2025/poster/45419

※Author information is current as of the date of writing or submission. Please be advised that the information may become outdated after that point.

For any inquiries about OSX, please contact us here.