[논문리뷰] One to More, More to One: Category-Aware Iterative Expert Training for Software Engineering Agents본 논문은 repository-level Software Engineering (SWE) task의 heterogeneous한 특성으로 인해 발생하는 category see-saw 문제를 해결하고자 합니다 .#Review#Software Engineering Agents#Reinforcement Learning#Category-Aware Training#Expert Training#Policy Integration#Multi-Teacher On-Policy Distillation (MOPD)#Refresh-Repair-Expand (RRE)#SWE Labeler2026년 9월 21일댓글 수 로딩 중