Tag
这篇论文将多选题评测中的选择偏差和顺序敏感性分解为翻转率(flip rate)和位置偏斜(positional skew)两个因素,并发现先验校正主要降低位置偏斜,而顺序聚合主要降低翻转率,建议同时报告这两个因素以区分缓解效果。
The paper introduces Sim2Signal, a benchmark for systematically measuring sim-to-real gaps in traffic signal control by decomposing them into observation, action, transition, and reward gaps and evaluating various mitigation methods.