Yuxiang Chen

HölderPO unifies LLM policy optimization with Hölder mean

Researchers have introduced HölderPO, a novel framework for optimizing large language models by unifying token-level probability aggregation through the Hölder mean. This approach offers continuous control over the trad…