Tian, Changyuan, et al. “Rectify Evaluation Preference: Improving LLMs’ Critique on Math Reasoning via Perplexity-Aware Reinforcement Learning”. Proceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 39, Mar. 2026, pp. 33241-9, doi:10.1609/aaai.v40i39.40609.