List your company

Mislabel detection on Weak Reference Labels Mislabels

84.3AP4PL Δℓi (proposed)
52.0660.4368.877.17May 28, 2026

Evaluation results

MethodLinks
4PL Δℓi (proposed)Criterion=Forced-ceiling likelihood ratio2026.0584.39895
top-10 disagreement2026.0576.18991.5
plain 4PL, single-stage (low di)Model=4PL, Protocol=single-stage, Selection=low difficulty2026.0571.892.189.5
XGBoost (4PL params)Model=XGBoost, Features=4PL parameters2026.0571.79390
4PL, low diModel=4PL, Selection=low difficulty2026.0570.39587
low riSelection=low response probability2026.0569.49888
plain 2PL (low ai)Model=2PL, Selection=low discrimination2026.0564.29572.5
GLAD2026.0562.29481.5
overall disagreement2026.0553.38073.5