2 papers
cs.LG2026
CLOOPD: Closing the Learner Loop in On-Policy Distillation
Keye Zheng, Hanyu Li, Zhan Cheng +1
On-policy distillation (OPD) pays twice for each fresh batch: the student generates trajectories and a stronger teacher scores them. Existing methods improve which trajectories are…
cs.LG2026
Marginal Advantage Accumulation for Memory-Driven Agent Self-Evolution
Mingyu Yang, Keye Zheng, Congchao Cheng +4
In batch-style trace distillation, the same memory operation may receive contradictory feedback across different batches. Existing methods lack a cross-batch, operation-level evide…