TADW: Traceable and Anti-detection Dynamic Watermarking of Deep Neural Networks

Table 11

The performance of test set and trigger set under pruning for different .

Model	Pruning rate (%)
		Testing acc (%)	WM query	Testing acc (%)	WM query
SN-10-1	0	93.46	Lossless	93.46	Lossless
	90	87.41	[21,21,20,21,21, 21,21,21,21,21]	87.57	[31,30,31,31,30, 30,31,30,31,31]
SN-10-2	0	93.47	Lossless	93.42	Lossless
	90	85.75	[21,21,21,20,19, 21,21,21,21,21]	86.20	[31,30,31,30,31, 31,30,31,30,30]
SN-10-5	0	93.59	Lossless	93.39	Lossless
	90	88.66	[21,19,20,21,21, 19,21,20,21,21]	83.67	[31,30,30,29,31, 30,31,29,30,30]
SN-10-10	0	93.39	Lossless	93.39	Lossless
	90	82.95	[21,19,21,21,21, 20,21,21,21,21]	85.70	[30,31,30,31,29, 30,31,31,30,31]