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

Table 6

Testing accuracy on clean models and watermarked models under different .

Model	SST-2			AG-News
	TextCNN (%)	TextRNN (%)	BERT (%)	TextCNN (%)	TextRNN (%)	BERT (%)
Clean	89.07	88.19	91.49	93.88	93.30	94.25
SN-15-1	88.58	88.19	91.16	93.42	92.82	93.89
SN-15-2	88.69	87.86	91.10	93.45	92.86	93.93
SN-15-7	88.63	87.70	90.99	93.59	93.04	93.99
SN-15-10	88.58	88.03	90.99	93.45	92.91	94.05
SN-20-1	88.58	87.75	91.05	93.57	92.87	93.87
SN-20-2	88.63	87.70	91.38	93.39	92.84	93.97
SN-20-10	88.69	88.08	90.99	93.46	92.80	94.14
SN-20-20	88.80	87.70	91.32	93.49	92.80	93.87