Yifan Sun

I will be at #NeurIPS2025 presenting our work on data-efficient RL fine-tuning for LLMs today (Dec 4, 11:00-2:00 pm, #2503)! 🚀 We propose difficulty-targeted online data selection and rollout replay, reducing RL training time by 23–62% while matching original GRPO performance.

Huge thanks to my amazing co-authors @evangelinejy99 @ExplainMiracles @RuiYang70669025 @FengLuo895614 @rui4research and our advisors Prof. Tong Zhang & @hanzhao_ml 🙏

🎉 Excited to share that our paper MiCRo: Mixture Modeling and Context-aware Routing for Personalized Preference Learning (aclanthology.org/2025.emnlp-mai…) received the #EMNLP2025 𝐎𝐮𝐭𝐬𝐭𝐚𝐧𝐝𝐢𝐧𝐠 𝐏𝐚𝐩𝐞𝐫 𝐀𝐰𝐚𝐫𝐝 🏆!