Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling

ICLROral2026

Authors: Shuyang Jiang, Yusheng Liao, Ya Zhang, Yanfeng Wang, Yu Wang
Affiliation: Fudan University
Venue: ICLR 2026
Track: Oral

TL;DR

theoretical unveil the underlying limitations of length reward and propose D$^2$yOR to achieve supreme efficiency without performance degradation…

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

← All ICLR 2026 Oral papers · Browse the whole archive