List your company

Rule Extraction Functional Fidelity on LLM Injected Behaviors Non-judge output abstractions

0.83MCCRuleSHAP
0.74680.76840.790.8116May 16, 2025

Evaluation results

MethodLinks
RuleSHAPTop-k=102025.050.83
RuleFitTop-k=102025.050.82
RuleSHAPTop-k=32025.050.8
RuleFitTop-k=32025.050.78
RuleSHAPTop-k=12025.050.76
RuleFitTop-k=12025.050.75