Backdoor Attack Task Recall on WebQuestion (test)

1Top 5 auPRCRISE

Evaluation results

Method
RISE2026.04	1	1	0.9995	0.9996	0.9993	0.9996	0.9985	0.999
RISE2026.04	1	1	0.9999	0.9998	0.9997	0.9997	0.9988	0.9991
RISE2026.04	1	1	1	1	0.9861	0.9903	0.9679	0.9797
TrackStar2026.04	0.9607	0.7	0.9439	0.7506	0.9412	0.8525	0.9527	0.8947
RapidIn2026.04	0.8722	0.8831	0.8554	0.9002	0.8191	0.8928	0.7975	0.8869
RapidIn2026.04	0.7879	0.747	0.7587	0.7943	0.7006	0.8085	0.6578	0.7957
TrackStar2026.04	0.6848	0.8331	0.7909	0.9085	0.9292	0.9797	0.9518	0.986
Embedding Similarity (E5-base-v2)2026.04	0.4089	0.8283	0.374	0.7047	0.3151	0.6411	0.2967	0.6725
ZO-Inf2026.04	0.3701	0.609	0.2646	0.5713	0.1471	0.5761	0.1283	0.575
ZO-Inf2026.04	0.2895	0.5069	0.2164	0.5099	0.1172	0.4844	0.1036	0.4883
BM252026.04	0.2589	0.7776	0.2785	0.7673	0.2733	0.7621	0.2699	0.7689
TrackStar2026.04	0.1025	0.4229	0.3884	0.5783	0.7471	0.852	0.826	0.9119