Backdoor Detection

Benchmarks

Dataset Name	SOTA Method	Metric
CIFAR-10	K-Arm	TPR100	135	2mo ago
CIFAR-10 (test)		TPR100	63	2mo ago
SST2	CSO(continuous)	TPR10	56	24d ago
LMSYS-Chat-1M		FPR on LMSYS0.1	50	3mo ago
MMLU		FPR (MMLU)0	50	3mo ago
Calibration Set		FPR (Calibration Set)36	50	3mo ago
Poisoned Prompt Set		Prompt Flagging Rate100	50	3mo ago
EMNIST		TPR100	49	2mo ago
CIFAR-100	Coward	True Positive Rate (TPR)100	49	2mo ago
GTSRB	K-Arm	TPR100	48	2mo ago
Backdoor behaviors (Jailbreak, Watching, Refusal, Password, Country, Sentiment) zero-shot	Best SAE	AUROC (zero-shot)86.8	36	1mo ago
SVHN		TPR100	30	2mo ago
Balanced 50% clean, 50% backdoored (test)	VillanDiffusion	Detection Accuracy98.2	28	4mo ago
Yahoo	CSO(continuous)	Tell me seriously10	24	24d ago
Stable Diffusion ObjRepAtt attacks v1.5	BlackMirror	Precision100	23	4mo ago
StrongREJECT prompts with triggers	WeightWatch	TPR100	20	3mo ago
SST2 high poison rate (test)	MM	Clean Performance Score10	18	24d ago
Tiny ImageNet (test)	PSBD	AUROC0.996	16	4mo ago
Simple IHU Llama 8B	Functional Attribution	AUROC0.992	15	3mo ago
Simple IHU Gemma 2B	Functional Attribution	AUROC1	15	3mo ago
GTSRB BppAttack (test)	ARCGEN	AUC0.9967	15	4mo ago
GTSRB WaNet attack (test)	MNTD	AUC1	15	4mo ago
GTSRB SIG attack (test)	MNTD	AUC97.63	15	4mo ago
MS-COCO v1.4 (val)	T2IShieldFTT	RickBKD Detection Rate99.9	14	3mo ago
Yahoo high poison rate	CSO(discrete)	Tell Me Seriously Score (Clean Label)10	13	24d ago

Showing 25 of 110 rows