ADWIN: Adaptive Windows for Horizon-Aware On-Policy Distillation
School of Computer Science, Peking University(北京大学计算机科学系);National Key Laboratory for Multimedia Information Processing, Peking University(北京大学多媒体信息处理国家重点实验室);LLM Department, Tencent(腾讯LLM部门)
On-policy distillation (OPD) transfers reasoning behavior by training a student on teacher feedback along student-generated trajectories, but standard full-rollout training ties every update to a costly completion and can over-allocate supervision to late positions with low marginal value for the current student. We revisit this assumption through the useful supervision horizon: student-induced rollouts can drift from teacher-preferred continuations, while aligned prefixes may already preserve the long-horizon OPD update direction. We propose ADWIN, an adaptive-window framework for OPD that treats rollout length as an online admissibility decision, training on short teacher-anchored prefixes while using delayed full-rollout probes to audit prefix--full alignment and adapt the next horizon with staleness control. Across math and code reasoning benchmarks in single-task, multi-task, and strong-to-weak settings, ADWIN improves the accuracy--compute trade-off over full-rollout OPD and prefix-based baselines, reducing end-to-end training cost by up to 4.1 times while achieving comparable or better accuracy.
Designing Adaptive Algorithms Based on Reinforcement Learning for Dynamic Optimization of Sliding Window Size in Multi-Dimensional Data Streams
Department of Computer Engineering, Ferdowsi University of Mashhad(法尔多西大学马什哈德分校计算机工程系);Department of Computer Engineering, University of Tehran(德黑兰大学计算机工程系)
MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
School of Mathematical Sciences, Shanghai Jiao Tong University(上海交通大学数学科学学院);Bio-X Institutes, Key Laboratory for the Genetics of Developmental and Neuropsychiatric Disorders, Shanghai Jiao Tong University(上海交通大学Bio-X研究院、发育与神经精神疾病遗传学重点实验室);Shanghai Key Laboratory of Psychotic Disorders, Brain Science and Technology Research Center, Shanghai Jiao Tong University(上海精神疾病重点实验室、脑科学与技术研究中心,上海交通大学)
Prune-OPD: Efficient and Reliable On-Policy Distillation for Long-Horizon Reasoning
The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));The Hong Kong University of Science and Technology(香港科技大学);MBZUAI;University of California, Merced(加州大学默塞德分校);Sun Yat-sen University(中山大学)
LongReD: Mitigating Short-Text Degradation of Long-Context Large Language Models via Restoration Distillation
Gaoling School of Artificial Intelligence, Renmin University of China(中国人民大学人工智能学院);Department of Computer Science, National University of Singapore(新加坡国立大学计算机科学系);Baichuan Inc.(百川公司)