[2605.20722] AGPO: Adaptive Group Policy Optimization with Dual Statistical Feedback
Researchers introduced Adaptive Group Policy Optimization (AGPO) for reinforcement learning in large language models.
AGPO outperforms existing methods, improving performance on math and STEM benchmarks significantly.
This record is extracted from a published AI Today issue and tied to the original source URL. Treat the source as the record of evidence for the summary.