Plausibility Analysis on MLQE-PE v1 (test) benchmark leaderboard · SOTA2 Research

Plausibility Analysis on MLQE-PE v1 (test)

0.67EN-DE src AUCAttention (best head)

Evaluation Results

Method	Links
Attention (best head)2022.04		0.67	0.67	0.7	0.65	0.7	0.7	0.7	0.69	0.73	0.75	0.67	0.6	0.67	0.66
Gradient L22022.04		0.64	0.65	0.65	0.49	0.67	0.61	0.68	0.55	0.72	0.68	0.65	0.54	0.67	0.59
Attention (SMaT)2022.04		0.64	0.65	0.68	0.52	0.66	0.64	0.66	0.54	0.71	0.7	0.61	0.54	0.66	0.6
Attention (best layer)2022.04		0.64	0.65	0.69	0.64	0.64	0.68	0.68	0.68	0.71	0.76	0.64	0.59	0.65	0.65
Attention (all layers)2022.04		0.6	0.63	0.68	0.52	0.6	0.61	0.58	0.55	0.66	0.7	0.62	0.55	0.62	0.59
Integrated Gradients2022.04		0.59	0.6	0.63	0.49	0.6	0.52	0.64	0.48	0.64	0.59	0.6	0.51	0.62	0.53
Gradient x Input2022.04		0.58	0.6	0.61	0.51	0.6	0.54	0.61	0.49	0.64	0.59	0.58	0.51	0.61	0.54
Attention (last layer)2022.04		0.51	0.49	0.61	0.49	0.51	0.5	0.55	0.48	0.52	0.57	0.56	0.5	0.54	0.5