The authors designed a paired evaluation framework with two complementary tests.
Fallback rules cause typed decision models to err even when the category is known.
Paired tests across three benchmarks show models fail to reassign inputs correctly and disagree on logically equivalent rule formulations.
Research Lab
Yuanbo Xie · Mingli Zhu · Tianyun Liu · Yingjie Zhang · Haixiao Xiong · Tingwen Liu
Li Auto Inc · Institute of Information Engineering, Chinese Academy of Sciences · University of Chinese Academy of Sciences
Research Digest··2 min read
The authors introduce the Paired Fallback Robustness (PFR) Test to measure how reliably typed decision models (TDMs) follow fallback rules in non-adversarial settings.
§