Research Article

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

Table 9

Testing accuracy on clean models and watermarked models under different .

Model
TextCNN (%)TextRNN (%)BERT (%)TextCNN (%)TextRNN (%)BERT (%)

Clean93.8893.3094.2593.8893.3094.25
SN-10-193.4692.8093.8393.4692.8393.91
SN-10-293.4792.9793.8393.4292.8093.82
SN-10-593.5992.8694.0093.3992.8893.78
SN-10-1093.3993.0393.7593.3992.8093.80