LSRL: Process-Supervised GRPO on Latent Recurrent States Improves Mathematical Reasoning.

Hangliang Ren

Browse the full EMNLP paper archive.